Winning by Losing: Evidence on Overbidding in Mergers

Winning by Losing:

Evidence on Overbidding in Mergers

Ulrike Malmendier

UC Berkeley and NBER

Enrico Moretti

UC Berkeley and NBER

Florian Peters

U Amsterdam and DSF

April 2011

Abstract

Do acquiring companies profit from acquisitions, or do acquiring CEOs overbid and destroy

shareholder value? Answering this question empirically is difficult since the hypothetical

counterfactual is hard to determine: A negative stock reaction of the acquiror to a merger

announcement is consistent with value-destroying mergers but also with overvaluation of the

acquiror. We argue that bidding contests where at least two bidders have a significant chance

of winning and acquiring the target help to address the identification issue: the post-merger

performance of the loser allows to calculate the counterfactual performance of the winner

without the merger. We construct a novel data set of all mergers with overlapping bids between

1983 and 2009. We find that the stock returns of bidders are not significantly different before

the merger contest, but diverge post-merger. In the full sample, winners underperform losers

over a three-year horizon, but there is large heterogeneity and the effect is not significant. In

the subsample of deals where at least two bidders have a significant chance at winning losers

outperform winners, while the opposite is true in cases with a predictable winner.

Keywords: Mergers; Acquisitions; Misvaluation; Hypothetical Counterfactual

JEL classification: G34; G14; D03

1 Introduction

Do acquiring companies profit from acquisitions? Or do acquiring CEOs, on average, overbid and

destroy shareholder value? The large payments for acquisitions and negative announcement effects

observed for a large set of acquirors have attracted a lot of attention to these questions. Moeller,

Schlingemann, and Stulz (2005) calculate that, during the last two decades, U.S. acquirors lost

in excess of $220 billion at the announcement of merger bids alone. Such findings have been

interpreted as empire building of acquiring CEOs (Jensen (1986)), other misaligned personal

objectives of CEOs (Morck, Shleifer, and Vishny (1990)), or the result of CEO overconfidence

about their proposed mergers (Roll (1986), Malmendier and Tate (2008). However, the evaluation

of the causes and consequences of mergers has been hampered by the empirical difficulty in

assessing the value created (or destroyed) in mergers. Using the announcement effect as a proxy,

one may underestimate the value creation of a merger due to price pressure around mergers

(Mitchell, Pulvino, and Stafford (2004)), information revealed in the merger bid (Asquith, Bruner,

and Mullins (1987)), or failure of the efficient markets hypothesis (Loughran and Vijh (1997)).

To the extent that the returns to mergers are revealed only over time and the announcement

effect is insufficient, it is hard to measure what portion of the returns can be attributed to a merger

decision rather than other corporate events or market movements. Consider, for example, the

argument of Shleifer and Vishny (2003) and Rhodes-Kropf and Viswanathan (2004) that CEOs

undertake mergers when their own firms are overvalued. Under this scenario, they purchase

assets (i.e. targets) that are less overvalued at a cheap price by using their overvalued stock.

The announcement effect of mergers is negative because bids reveal the acquirors’ overvaluation

and because the targets are less highly valued. However, in the long-run such mergers might still

be in the best interest of current shareholders: Had the CEOs not used the overvalued stock for

acquisitions, the stock price would have fallen even more. Moreover, long-term returns may also

be affected by slowly declining overvaluation. In other words, it is hard to evaluate the returns

to mergers empirically due to the lack of a clear counterfactual.

This paper studies bidding contests to address the identification issue and, in particular, the

1

concern about prior run-ups in stock prices. Our research design exploits the concurrent bidding

of two or more companies for the same target. We identify cases in which at least two bidders had

ex ante, i.e., at the time they made their offer, a significant chance at winning the contest and

acquiring the target. We show that, in these cases, bidders show very similar stock performance

prior to the merger contest and argue that the losing bidder’s post-merger performance helps

to calculate the hypothetical post-merger performance of the winning bidder had the merger

not taken place. The basic idea is that participation in such a bidding contest provides an

additional matching criterion, beyond the usual market-, industry-, and firm level controls. To

the extent that overvaluation affects the propensity to acquire, participation in a bidding fight

should account for that. It is also likely to capture strategic considerations that affect the decision

to attempt a takeover, which are hard to control for with the set of standard financial variables.

More generally, our empirical methodology allows us to include controls beyond time-varying firm

characteristics and beyond unspecified time-invariant firm characteristics (i. e. beyond firm fixed

effects). In particular, we are able to control for unspecified merger case effects. The ability

to include an additional set of case fixed effects and to test the validity of our hypothetical

benchmark directly, by comparing price paths and other characteristics of bidders prior to the

merger fights, provides a methodological improvement relative to prior empirical approaches to

measuring the returns to mergers for the acquiring company. While our findings are specific

to the set of contested mergers, the discrepancy between our results and those generated with

standard approaches of measuring returns to mergers suggest that the well-known biases in prior

approaches are economically important.

We construct a novel data set of all mergers with overlapping bids of at least two potential

acquirors between 1983 and 2009. We then compare adjusted abnormal returns of all candidates

both before and after a merger fight. Analyzing the price paths of bidders before the merger

fight allows us to check directly the validity of our key identifying assumption - that losing

bidders are a valid counterfactual for the winner, after employing the usual controls and matching

criteria. We find that stock returns of bidders are not significantly different before the merger

fight, but diverge significantly after one bidder has completed the merger. On average, winners

2

underperform losers over a three-year horizon post-merger, but there is large heterogeneity and

the effect is not significant. We then identify the subsample of cases where both the (ultimate)

winner and the (ultimate) losers of a given merger contest appear to have, ex ante, a significant

chance at winning the merger contest. Specifically, we exploit contest duration and press reports

to identify cases of actual ”bidding wars.” In short-duration merger contests (e.g., the lowest

quartile of two- to four-month merger contests), there is typically a clear candidate for winning,

while in long-duration contest (e.g., the highest quartile of one-year and longer contests), the

back and forth between bidders indicates that at least two bidders have a significant chance at

winning the contest. Splitting the sample into quartiles, we find that winners outperform in the

bottom quartile of contest duration but underperform in the top quartile. Each effect is at least

marginally significant, and the difference is strongly significant in all specifications. The results

exploiting press reports show the same pattern.

These results are robust to various sample selection criteria and controls. In particular, we

address the alternative explanation that winners undergo a change in risk profile due to the

merger. We test for significant differences in winner- versus loser pre- and post-merger fight

betas to address out this hypothesis. Moreover, a risk- shift explanation would require a jump in

price level of winners at merger time relative to the losers, which we do not observe.

Our results suggest that, for the subset of mergers that involve at least two bidders either one

of whom could be winning the contest, making the acquisition destroys shareholder value of the

acquiror, on average. The leading alternative hypothesis, acquiror overvaluation, would require

winning bidders to be more overvalued than losing bidders. The closely matched price paths of

winners and losers prior to the merger contest suggest that this is not the case. In addition, the

use of firm and contest dummies addresses this concern.

The research design in this paper is motivated by Greenstone, Hornbeck, and Moretti (2011).

There, the authors analyze the local consequences of attracting a million-dollar plant to a county,

including the effects on labor earnings, public finances, and property values. Compared to the

county-level analysis in their paper, mergers allow for considerably more exhaustive controls

of heterogeneity among bidders. Moreover, stock prices of bidding companies incorporate the

3

markets expectations about future cash flow. At a minimum we view this methodology as an

improvement over previous matching procedures as sole the basis for alternative (counterfactual)

returns.

The paper proceeds as follows. Section 2 describes the data sources and presents some sum-

mary statistics. Section 3 explains the econometric model. Section 4 describes the results while

Section 5 concludes.

2 Data

2.1 Sample Construction

Our initial dataset contains publicly submitted bids in merger contests between January 1983

and December 2009 recorded in the SDC Mergers and Acquisitions database.

The takeover process typically begins long before a bid is publicly placed, and can be divided

into a private and a public stage. This process has been described in detail by Boone and Mul-

herin (2007). It usually starts with an investment bank soliciting interest of potential acquirors,

after which interested parties sign confidentiality agreements and obtain access to non-public in-

formation about the target. Following this, several non-binding rounds of bidding are conducted

in order to identify a small group of seriously interested parties. A final auction among these

bidders then leads to a binding offer, and the prevailing bid becomes public. Any competing bid

placed from then on, is public and binding. It is the public bids, i.e. the prevailing bid of the

final private auction and any rival bid placed after that, that SDC records. In the spirit of our

identification strategy, we focus on these binding and public bids, because the respective bidders

are most seriously interested in acquiring the target and thus more likely to be ex ante similar.

For each bid we collect the SDC deal number, the acquiror’s SDC assigned company identifier

(CIDGEN), six-digit CUSIP, ticker, nation, company type, and the SIC and NAICS codes. We

also collect the following bid characteristics: announcement date, effective or withdrawal date,

the percentage of the transaction value offered in cash, stock or other means, the deal attitude

(friendly or hostile), and the acquisition method (tender offer or merger).

4

We restrict the initial sample of bids in the following ways:

• Deal status: We eliminate merger contests that are not completed by December 31, 2009.

• Acquiror nation: We restrict our analysis to U.S. acquirors.

• Acquiror type: We consider bids by public companies only, and exclude privately held

and government-owned firms, investor groups, joint ventures, mutually owned companies,

subsidiaries, and firms whose status SDC cannot reliably identify.

• Parent/subsidiary relationship: We exclude contests in which the acquiror is the ultimate

parent of the target company.

• White knights: Bids by companies that enter as white knights are excluded.

We define bidders as contestants in the same merger fight if their bids are for the same target

and the time periods in which the bids are effective overlap. SDC contains a variable indicating

whether a particular bid was contested, and provides the bid number(s) of the competing bids.

We code two bids as competing if, for a given target, the first bid is recorded as a competing

bid of the second bid and vice versa, or if both bids have a common third competing bid. We

then check that all bids classified as contested in the first step are placed before the recorded

completion date. Companies that succeed in completing the merger are classified as winners, all

other bidders participating in the same contest are coded as losers. In our initial dataset, there

are three contests to which SDC erroneously assigns two winners. We hand-check these cases and

identify the unique winner by a news wire search.

The beginning of a merger fight is defined as the announcement date of the earliest bid. The

end of a merger fight defined as the date of completion, or the last available withdrawal date

if there is no winner. We create an event time variable, t, which counts the months relative to

the contest period. We set t equal to 0 for the end of the month preceding the start date of

the merger contest, and equal to +1 for the end of the month in which the merger fight ends.

Figure 1 illustrates the variable construction with a stylized example and with a concrete example

5

from our dataset, the merger contest between Westcott Communications and Automatic Data

Processing for Sandy Corporation.

[Figure 1 approximately here]

Next, we merge the SDC data with financial and accounting information from CRSP and

COMPUSTAT using each company’s CRSP permanent company and security identifiers (PERMCO

and PERMNO). To obtain the PERMCO pertaining to each SDC bidder, we match the 6-digit

CUSIP provided by SDC with the first six digits of CRSP’s historical CUSIP (NCUSIP). In

general, the CUSIP of a particular firm can change over time as CUSIPs are sometimes reas-

signed. Thus, the 6-digit CUSIP uniquely identifies a firm only at a given point in time. Since

reassignment of CUSIPs is particularly common following a merger, we are careful to take this

into account when determining bidder PERMCOs. Specifically, we match, for any given bid,

SDC’s 6-digit bidder CUSIP with CRSP’s 6-digit NCUSIP for the month end preceding the an-

nouncement date, and extract the respective PERMCO. This procedure ensures that we obtain

the correct PERMCO for any given bidder. In a second step, we manually check that, for the

sample of matched CUSIPs, SDC company names correspond to CRSP company names. This

is indeed the case for all matched bidders. Another complication arises from the fact that firms

often have multiple equity securities outstanding. In these cases we use (1) the common stock

if common and other types of stock are traded; (2) Class A shares if the company has Class A

and Class B outstanding; (3) the stock with the longest available time series of data if there are

multiple common stocks traded.

We extract the following data for 5-year periods around the start and end date of each contest

and bidder:

• CRSP Monthly Stock Database: holding period stock return (RET), distribution event code

(DISTCD), delisting code (DLSTCD), and delisting return (DLRET).

• CRSP-COMPUSTAT Fundamental Annual Database: Annual accounting data such as to-

tal assets, book and market value of equity, operating income, and property, plants and

6

equipment.

• Market data: monthly CRSP value-weighted index returns, T-bill yields, and Fama-French

factor returns.

Using the monthly CRSP stock return series, we then construct the time series of monthly

bidder stock prices for an initial window of 5 years around the merger fight (t= -59 to t=60). We

normalize the price series to 100 at t=0. We emphasize that the CRSP holding period return is

adjusted for stock splits, exchanges, and cash distributions, and thus properly accounts for such

events which are particularly common around mergers.

We then construct the time series of target stock prices in the same way as we do for bidders.

We use this time series to compute the offer premium as the run-up in the target stock price

from one month preceding the start of the contest (t = −1) until completion of the merger. We

compute the offer premium both in percent of the target equity value, and in percent of the

acquiror equity value.

Our initial sample contains 193 takeover contests representing 416 bids by 402 bidders of

which 154 are winners and 248 are losers. We first drop repeated bids by the same bidder, and

keep only the final bid. This eliminates 14 bids. We then drop 45 contests that had not been

completed until December 31, 2009. Next, we drop two contests for which either the winner or

the loser could not be matched to a CRSP PERMNO. We then delete contests where the winner

is the ultimate parent company of the target (10 contests) since ultimate parents are unlikely to

provide a good comparison for other bidders. When we require both the winner and the loser

to have non-missing stock price data for at least periods 0 (the month preceding the start of the

contest) to period 1 (the month following the end of the contest), another 4 contest drop out of

the sample. We also eliminate seven bids where the bidding firm has extreme stock price volatility

over the event window, with the standard deviation of the price exceeding 200,1 since these firms

1The volatility is calculated using the full event window of +/- three years. Four of these firms are in the HighTech sector: CTS Corp, Yahoo!, Liquid Audio Inc, and QWest Communications. Two are in the Healthcare sector:Inamed Corp and Hyseq Pharmaceuticals. One firm, Cannon Group, operates in the Service sector. All of thesefirms show 10 to 20-fold increases and reversals in their stock market valuations, mostly occurring in the pre-mergerperiod.

7

appear to be be influenced by idiosyncratic factors and are, ex ante, a poor benchmark for their

respective contestants. (If we keep these firms in our sample, our qualitative results remain

unchanged, but the confidence bounds in the pre-merger period increase substantially.) For our

baseline sample we further require non-missing stock price data for at least periods −35 to +36

(i.e., 3 years before and after the contest). This reduces the number of contests to 114. Finally,

we balance the sample and keep only those contests for which we have contests for both the

winner and the losers. This reduces the sample by another 20 contests. The final sample contains

82 merger contests, representing 172 bids by 82 winners and 90 losers. Table 1 summarizes the

construction of our dataset, and Figure 2 illustrates the frequency distribution of the contests

over the sample period. We observe between zero and eight contested mergers per year, with

spikes in the mid-1980s and mid-1990s.

[Table 1 approximately here]


In the 3-year period following a merger contest, many bidder stocks disappear from CRSP

due to delisting. We are careful to account for delisting events and their implications for stock

holders using all available delisting information provided by CRSP. CRSP’s delisting code (DL-

STCD) classifies delists broadly into mergers, exchanges for other stock, liquidations, and several

categories of dropped firms. In addition, CRSP provides delisting returns and distribution in-

formation.2 We track the performance of a delisted firm as if shareholders were buy-and-hold

investors whenever they receive payments in stock, mirroring the underlying assumption when

tracking performance of listed firms. Specifically, we assume that stock payments in takeovers are

held in the stock of the acquiring firm; exchanges for other stock are held in the new stock. When

shareholders receive payments in cash (in mergers, liquidations, and bankruptcies), or CRSP

cannot identify or does not cover the security in which payments are made, we track performance

2Delisting returns are defined as shareholder returns from the last day the stock was traded to the earliestpost-delisting date for which CRSP could ascertain the stock’s value. Distribution data contains information aboutwhether and to what extent shareholders of a takeover target were paid in cash or stock

8

as if all proceeds were invested in the market portfolio. We use the value-weighted CRSP index

as a proxy for the market portfolio.

2.2 Descriptive Statistics

In this section, we provide descriptive statistics of bidder and deal characteristics. Panel A of

Table 2 contains the bidder characteristics, Panel B contains the bid characteristics. In Panel A,

we report the statistics separately for winners and losers. The two rightmost columns report the

winner-loser differences in means and medians. The variables are computed from yearly balance

sheet and income data, and refer to the fiscal year end preceding the beginning of the contest.

The first two rows of Panel A indicate that both winning and losing firms in contested bids are

very large compared to the average Compustat firm. This is due to the fact that we require

firms to be public, but also that SDC is more likely to record bids of large firms. In terms of

firm size, winners tend to be larger than losers. However, the size difference is much smaller

than that between the average acquiring and non-acquiring firm. Winners also have a somewhat

higher Tobin’s Q than losers. Profitability and leverage are virtually identical between winners

and losers.

The tests for differences in means and medians reveal that none of these differences in charac-

teristics is statistically significant. This is a first indication that our identifying assumption that

losers are a valid counterfactual for the winners is indeed supported by the data.

Panel B shows that transaction values in contested mergers are quite large compared to bidder

size. The average transaction value is about one quarter of the losers’ market capitalization and

about 16 percent of the winner’s market capitalization. Another interesting aspect is that most

contests in SDC are between two bidders, while higher numbers of contestants are very rare. Also

notable is the fact the final offer premium in our sample of contested bids is greater than that of

non-contested bids. Betton, Eckbo, and Thorburn (2008) find that the average offer premium in

a sample of 4,889 control bids for US targets during 1980-2002 to be 48 percent. The final offer

premium in our sample is about 58 percent. This may be an indication of some degree of winner’s

curse brought about by the presence of competing offers. Below we explore this possibility in more

9

detail. 3 Maybe the most important difference between contested and non-contested acquisitions

is contest duration. While the average time to completion in single-bidder mergers is about 65

trading days (see Betton et al. (2008)), merger contests take, on average, 9.5 months - and thus

about three times as long - to close. This seems to be a defining aspect of merger contests. In

Section 4.3 we therefore explore the correlates of contest duration in more detail. One plausible

hypothesis is that contest duration is a proxy for bidder similarity. Single-bidder acquisitions

may fail to elicit competing bids - and thus have short completion times - precisely because other

potential acquirors differ too much in terms of the synergies they could generate. Competing

bids are only launched if these synergies are similar enough for at least two potential acquirors.

Finally, a contest of multiple competing bidders is likely to take longer the more similar the

synergies are.


3 Econometric Model

A simple estimator of the effect of mergers on firm performance can be obtained by regressing

stock market price on a dummy for whether a firm successfully completes a merger, controlling

for observable characteristics of the firm. Alternatively, a matching estimator can be obtained

by comparing the stock market price of firms that successfully complete a merger to the stock

market price of the average firm in the market with a similar set of observable characteristics.

The consistency of both types of estimators crucially depends on the assumption that in the

absence of the merger the stock market price of the acquiring firm would have evolved like the

stock market price of the average firm with similar observable characteristics. In other words,

both the regression and the matching estimator are based on the assumption that the acquiring

3Offer premia expressed as a percentage of the acquiror equity value are smaller since acquirors tend to besignificantly larger than targets.

10

firm and the average firm in the market have identical unobserved determinants of stock prices,

conditional on covariates.

Of course, this assumption could be violated in reality. The acquiring firm is likely to be

different along many unobserved dimensions from the average firm in the market. Both positive

and negative selection, in terms of future stock price performance, is a priori possible. Positive

selection would occur if firms that successfully acquire other firms have better unobservable

characteristics. This would be the case if, for example, firms that are outperforming other firms

in the same industry tend to grow by mergers and acquisitions. Negative selection would occur if

firms that successfully acquire other firms have worse unobservables. This would be the case if,

for example, firms in industries that are experiencing declines in demand tend to consolidate and

merge with other firms. In terms of the naiıve regression or the matching estimator described

above, positive selection would lead to an overestimate of the true effect of a merger on firm

performance. Negative selection would lead to an underestimate of the true effect of a merger on

firm performance.

One important contribution of this paper is the use of contested mergers to eliminate - or at

least reduce - the scope of this type of omitted variable bias. The idea is simple: in mergers where

there is only one potential acquiring firm, any evaluation of future long-run returns is necessarily

relative to other firms with similar observable characteristics. In contested mergers, however, two

firms fight to acquire the same target firm. Our key assumption is that the losing firm provides

a valid counterfactual for what would have happened to the winning firm in the absence of the

merger. Even if this assumption is not true exactly, it is likely that winners are more similar to

losers than to the average firm in the market, or even to non-acquiring firms that have similar

observable characteristics.

Using this identification strategy, our goal is to estimate the causal effect of prevailing in the

contested merger on stock market valuation. Here we describe the econometric model used to

estimate this effect. We use a panel of firms observed for 3 years before and after the merger. We

11

fit the following regression equation:

Pi,j,τ =

T∑τ ′=T

πτ ′Wτ ′i,j,τ +

T∑τ ′=T

δτ ′Cτ ′i,j,τ + γ′Xi,j,t + ηj + ξt + εi,j,τ (1)

In this equation, i references firms, j indicates a bidding contest, and τ indexes the month in

event time. As explained above, event time is defined such that τ = 0 indicates the month pre-

ceding the start of the merger contest, while τ = 1 refers to the month immediately following the

conclusion of the contest. Observations during the merger contest are not used in this regression.

The outcome variable in equation (1), Pi,j,τ , is the stock price of firm i in month τ relative to

the contest period of contest j. All stock prices are end-of-month values, normalized to 1 in the

month preceding the beginning of the contest, i.e at τ = 0. The vector ηj is a full set of case

fixed effects that adjusts for permanent case-specific differences in the intercept of the outcome

variable. These dummies account for all fixed characteristics of each pair or group of contes-

tants. ξt is a vector of calendar month fixed effects, capturing any calendar time-specific effects

on winner or loser stock prices. These indicators essentially control for aggregate market-driven

fluctuations of bidder prices. Xi,j,t is a vector of time-varying observable firm characteristics,

εi,j,τ is a stochastic error term.

The key variables are the W τ ′i,j,τ and the Ct

′i,j,t indicators. The Cτ

′i,j,τ variables are simply a

set of dummies indicating event time, i.e. event time fixed effects. That is, Cτ′i,j,τ is equal to 1

- whether firm i is a winner or loser - if the event period is equal to the running index τ ′, i.e.

Cτ′i,j,τ = 1(τ = τ ′). The Wi,j,τ variables are a set of event time-winner indicators. That is, W τ ′

i,j,τ

takes the value of 1 if firm i is a winner in contest j and the observation pertains to event period

τ ′. Formally, W τ ′i,j,τ = 1(τ = τ ′ and firm i wins contest j). Given these two sets of indicator

variables, the coefficients δτ ′ measure the average loser price in period τ ′ relative to the contest

period, while the coefficients πτ ′ estimate the average price difference between the winner and the

loser in event period τ ′. In this way, the effect of winner or loser status is allowed to vary with

event time. For example, for τ ′ = 3, δτ ′ is the conditional mean of the loser price 3 months after

the end of the bidding contest, and πτ ′ is the conditional mean difference between the winning

12

and losing firms’ stock prices 3 months after the completion of the merger.

A few details about the identification of the π coefficients deserve highlighting. First, and most

importantly, including case fixed effects guarantees that the π-series is identified from comparisons

within a winner-loser pair. Including them allows us to retain the intuitive appeal of pairwise

differencing in a regression framework. Second, it is possible to separately identify the π’s, and

calendar time effects because the merger announcements occur in multiple years. Third, some

firms are winners and/or losers two times or more, and any observation from these firms will

simultaneously identify multiple π’s.

For more refined statistical tests, we specify a regression that estimates a piecewise-linear

approximation of the period-specific π-coefficients of equation (1):

Pi,j,τ = α0 + α1 Wi,j,τ + α2 τ + α3 τ ×Wi,j,τ + α4 Posti,j,τ + α5 Posti,j,τ ×Wi,j,τ

+ α6 τ × Posti,j,τ + α7 τ × Posti,j,τ ×Wi,j,τ + ηj + ξt + δ′Xi,j,τ + εi,j,τ (2)

This specification allows for linear, and different pre-contest and post-contest trends in the winner

and loser prices as well a level shift of the winner-loser price difference at the time of the merger

contest. Our statistical tests focus on the following parameters: To establish validity of our

identifying assumption that losers are a valid counterfactual, we are interested in the following

parameters: (1) α1, the average level difference between winner and loser price before the merger

contest. And (2) α3, the trend difference between winner and loser price before the merger

contest.

A test of [α1+# of pre-merger periods ·α3] = 0 is informative about whether winner and loser

stock prices diverge significantly before the beginning of the contest. An insignificant t-statistic

supports our identifying assumption.

For the causal effect of the merger we are interested in the long-run price divergence of winners

and losers. If winners and losers do not differ in their exposure to risk factors, there is no reason,

apart from a ”merger effect”, to expect them to show any performance divergence. When we use

cumulative abnormal returns, instead of prices, as the dependent variable, the effect of possibly

13

different risk exposures on expected returns is purged. Therefore, a statistical evaluation of a

merger effect would simply test for the winner- loser performance difference at the end of the

post-merger event window. With our piecewise-linear regression specification this amounts to a

test of [α1 + α5 + # of post-merger periods · (α3 + α7)] = 0.

A significant t-statistic suggests a causal effect of the merger on the winning firm. The

parameter measuring the pre-merger trend, α3, is included in the test equation, since our aim is

to measure the total slope of the post-merger trend, not just the incremental trend shift, if the

identifying assumption is not rejected in our data. In a similar way, α1, the pre-merger winner-

loser difference, is included in the equation, because, even though winner-loser differences are

normalized to zero in period t = 0, the regression does not estimate α1 to be precisely equal to

zero. And so the piecewise-linear approximation of the post-merger performance difference would

be misstated if α1 were not accounted for.

To illustrate our estimation method, the top graph of Figure A-1 plots the series of π-

coefficients against event time τ , along with 95% confidence bands and the piecewise-linear

approximation obtained from regression (2). As a starting point, we run regressions (1) and

(2) with the sets of Cτ′i,j,τ and W τ ′

i,j,τ dummies and case fixed effects only, i.e. without additional

controls. For illustrative purposes, we use a symmetric 3-year window around the contest period.

The point estimates of the π’s represent the period-specific mean difference between winner and

loser stock prices around merger contests. From this graph, it is evident that the winning and

losing firms have very similar price trends during the 3 years before the start of the contest. The

average trend in the winner-loser price difference, as estimated by the solid linear line in the left

half of the graph, is statistically indistinguishable from zero. This is quite important. Because

stock market prices reflect both current economic conditions as well as future expectations, the

similarity in trends in the years before the announcement lends credibility to our identifying

assumption that losing firms provide a valid counterfactual.

In addition, Figure A-1 shows that winners experience a price drop during the contest period

and in the month immediately following merger completion. From then on, the winners and

14

losers exhibit essentially identical price paths - the winner-loser difference remains flat. Though

the price drop of winners relative to losers is not statistically significant, below we turn to a more

detailed analysis of the various factors driving the winner-loser price divergence following merger

completion.

[Figure A-1 approximately here]

In principle, the parameters in equations (1) and (2) can alternatively be estimated using ”dif-

ferenced” data. For example, the OLS estimate of the π-vector in regression (1) is approximately

equal to the estimate of the π-vector of the following regression:

∆Pj,τ =

T∑τ ′=T

πτ ′Cτ ′j,τ + β′∆Xj,t + ξt + ζj,τ (3)

Here, the dependent variable is the period-specific winner-loser price difference within a contest.

Similarly, the ∆Xj,t-vector is the vector of period-specific winner-loser differences in the observable

firm characteristics. Because, in this specification, the regression is run on the within-case winner-

loser differences, the coefficients of the event time dummies, Cτ′j,τ , directly estimate the average

period-specific winner-loser differences. The series of period-winner indicators, W τ ′i,j,τ , thus drops

out, as do the case fixed effects. It can be shown that, on a balanced sample with only one loser

per contest, the OLS estimates of π and π are numerically identical. However, these estimates

generally differ on unbalanced samples and on samples that contain contests with multiple losers.

In essence, the ”level” specification of equations (1) and (2) makes better use of multiple losers,

and it is therefore our specifications of choice.

15

4 Results

4.1 Testing the Identifying Assumption

We start by analyzing statistically the validity of our identifying assumption that the loser in

a merger contest is a valid counterfactual for the winner. The task is to establish similarity of

winner-loser pairs both in observables and unobservables. To do this, we use the pre-merger stock

price paths of each winner-loser pair. We first estimate the model

rijt − rft = αij + βij(rmt − rft) + εijt (4)

on the pre-merger period, separately for the winner and loser in each contest. In this equation, i

indicates the bidder (either winner or loser), j references the contest, and t denotes the month.

The above decomposition of stock returns allows us to distinguish observable from unobserv-

able determinants of bidder prices. The component βij(rmt−rft) of the bidder return is explained

by the exposure to market risk and the return of the market portfolio. In contrast, αij and εijt

are due to factors unobserved by the econometrician: αij is the average excess return, i.e. the

part of the price trend that cannot be explained by market risk. εijt, on the other hand, is the

monthly unexplained residual return.

If our identifying assumption indeed holds, we expect that both observable and unobservable

determinants of stock returns are similar within a winner-loser pair before the merger. This would

imply that winner and loser alphas, betas and residuals are correlated. To test this assumption,

we regress the winner alpha on the loser alpha, the winner beta on the loser beta, and the winner

residuals on the loser residuals. If winners and losers have a common determinant that is omitted

from equation (4), alphas and residuals will be correlated through this omitted factor. Such

a determinant could be observable but simply omitted. For instance, winners and losers may

load similarly on some risk factor which is observable but omitted from equation (4). In contrast,

winners and losers may also share a common determinant which is unobservable or unknown. Such

an unobservable factor could be the essential source of endogeneity in the decision to acquire. By

16

including the standard observable risk factors in this equation we reduce the scope of observables

driving winner-loser correlation in alphas and residuals, and we shift it towards unobservables.

Hence, while we cannot exclude the possibility that the winner-loser correlation is in part also

driven by omitted, observable factors, it is likely to be mostly due to common, unobservable

factors.

The results are shown in Table 3. We report regressions both for the pre-merger (column 1) as

well as for the post-merger period (column 2). While we are mostly interested in the pre-merger

relationships, comparing these with the post-merger results is informative about which aspects

of similarity change through the merger.


Consistent with our assumption, the first column of Panel A of Table 3 shows that the pre-

merger alphas of winners and losers are highly correlated prior to the merger. This means that

winning bidders who experience run-ups during the three years preceding the merger are typically

challenged by rival bidders that experience a similar run-up during that period. In the context

of mergers, this is an important aspect of similarity. Contestants with markedly different pre-

merger price trends may vary significantly in their motives for and prospects of acquisitions. For

example, the post-merger performance of acquirors motivated by overvaluation of their own stock

- possibly following a pre-merger run-up - might be systematically different from the post-merger

performance of acquirors that did not experience a recent run-up. The pre-merger winner-loser

correlation in run-ups is further illustrated in Figure 3. The four panels of the figure plot the

average cumulative abnormal returns of winners and losers for subsamples sorted by the pre-

merger abnormal return of the winning bidder. Quartile 1 contains the poorest-performing,

and quartile 4 the best-performing winners. We compute the monthly CARs for each bidder

as the risk-adjusted excess return, rijt − rft − βij(rmt − rft), where βij is estimated via OLS

from regression (4), separately for the pre- and post-merger period.4 The monthly CARs are

4See Brennan, Chordia, and Subrahmanyam (1998) for a similar approach for individual stocks.

17

then normalized to zero at t = 0 and cumulated over the event window. Consistent with the

regression results, losers exhibit an abnormal performance that is very similar to winners in the

years preceding the merger. In fact, the abnormal returns of winners and losers in all quartiles are

not statistically significantly different at any point in time in the pre-merger period and they are

virtually indistinguishable in the six months preceding the merger. Most importantly, the return

paths of winners and losers in the fourth quartile, which is most likely to contain overvalued

acquirors, overlap for an even longer period, approximately eighteen months.


The second column of Table 4, Panel A, shows that the winner-loser correlation in alphas

drops to one half after the merger and remains only marginally significant. Moreover, R-squared

drops from 15.5 to 3.6 percent, i.e. to about one fourth, from the pre- to the post-merger period.

This drop in trend correlation is indicative of a causal effect of the merger.

Panel B of Table 3 reports regressions of winner betas on loser betas. Here, again, we expect,

and find, a high pre-merger correlation, capturing the similarity of winners and losers in their

correlation with the market. Comparing the pre- and post-merger period, we do not observe a

significant change in beta correlation. The coefficient drops only marginally from 0.366 to 0.329

and the R-squared drops from 12.3 to 11.3 percent. Thus, in our sample, mergers do not lead to

significant changes in exposure to market risk.

Finally, Panel C tabulates regressions of winner on loser market model residuals. The coeffi-

cient of the residual regression is identical to the coefficient, γ, of the loser return of an extended

market model, where the loser return is added as an explanatory variable of the winner return:5

rWjt − rft = αWj + βWj (rmt − rft) + γWj (rLjt − rft) + νWjt (5)

Thus, a positive and significant coefficient means that the loser return has explanatory power for

the winner return over and above its exposure to market risk. The results shown in Panel C of

5See the Frisch-Waugh-Lovell Theorem

18

Table 3 confirm that this is indeed the case.

Taken together, the similarity of winner-loser pairs in pre-merger trends, market exposures

and market model residuals strongly supports the credibility of the identifying assumption that

the losers form a valid counterfactual for the winners.

4.2 Main Results

We now turn to the analysis of the causal effects of mergers using our econometric method

described in Section 3. We start by presenting the results for the final sample of 82 merger

contests, i.e. the balanced and matched sample. The top panel of Figure A-1 illustrates the basic

findings, i.e. the estimates from regressions (1) and (2) without controls. In this section, we focus

on regression (2), the piecewise-linear approximation of the period-specific winner-loser differences

as illustrated by the black solid lines in Figure A-1. Table 4 reports six different specifications

of that regression. The first column corresponds to the top panel of Figure A-1, and uses no

additional controls besides the Cτ′i,j,τ and W τ ′

i,j,τ dummies and case fixed effects. Columns two and

three add firm-specific controls and year fixed effects. Columns four to six use the cumulative

abnormal return as the dependent variable instead of the raw price.

The rows named ”Identification” and ”Merger Effect” report the test statistics of interest.

The top two rows report the test of the identifying assumption that price paths of winners and

losers are statistically indistinguishable before the contest. The last rows report a test of whether

the price paths diverge significantly during the contest period and following the completion of

the merger. If the identifying assumption is correct, then the latter is a test of the causal effect

of the merger on the acquiring firm’s stock performance.


As Table 4 shows, the identifying assumption of equality of winner and loser price paths before

the merger cannot be rejected in any of the specifications. The statistical test reported in the

19

second-to-last row formalizes the evidence illustrated in the top panel of Figure A-1, in particular

the wide confidence bounds of the period-specific winner-loser differences. We also do not find

statistical significance of the causal effect of the merger, i.e we do not find a significant jump and

trend shift following the merger. Again, this result is expected given the wide confidence bands

of the period-specific coefficients.

We find the same qualitative results when we run the parameter tests using less restrictive

sample selection criteria: when the sample is balanced but not matched, when the sample is

matched but not balanced, or when the sample is neither balanced nor matched.

4.3 Subsample Analysis

The last section has shown that, on average, winners of merger contests do not perform signifi-

cantly worse than losers in the years following merger completion. We now turn to sample splits

to investigate the reasons for variation in post-merger performance.

One reason why a merger can be value-destroying for stockholders of the acquiring firm is

overbidding. Overbidding can be due to an inflated perception of the target’s standalone value or

to an overvaluation of potential synergies. If the market realizes and prices overbidding quickly,

the acquiror’s stock price will experience a drop immediately following the bid announcement

and the completion of the deal. Given the way we code event time in this study, such a drop

would occur between period zero (the month end before the first bid announcement) and period 1

(the month end following completion of the transaction). In contrast, if the market incorporates

overbidding only gradually into prices, one should see a decline of the winner’s performance in the

periods following deal completion.Another reason for poor post-merger performance of winning

firms could be that winning a merger contest could be correlated with managerial characteristics

that lead to weak performance. For instance, CEO overconfidence implies a higher probability of

winning the merger contest and may also lead to worse post-merger performance.

The descriptive statistics in Section 2.2 show that contest duration is an aspect that seems to

set multi-bidder takeover contests apart from single-bidder acquisitions. Time to completion is

20

more than three times longer in contested bids than in single-bidder acquisitions. Reiterating the

arguments made above, it is plausible to expect that longer contest durations are associated with

more similarity of winners and losers in takeover battles. Single-bidder takeovers may attract

only one contestant - and thus be completed quickly - because other potential bidders are too

different from the actual bidder, and the synergies they can generate are significantly smaller. In

contrast, multi-bidder contests may be particularly fiercely fought - and hence take a long time

to complete - if bidders, and the synergies they can obtain with the target, are very similar. This

may also explain the higher takeover premia that we observe in our sample of contested bids.

We now investigate the effect of contest duration in more detail. We first analyze the rela-

tionship between contest duration and stock price performance. We split our sample of contested

mergers into duration quartiles. The quartiles (from first to fourth) contain merger contests that

last, respectively, two to four months, five to seven months, eight to twelve months, and more

than a year (up to 43 months). The contest duration in quartile one roughly corresponds to the

average duration of non-contested mergers, while all other quartiles contain significantly longer

fight durations.

To illustrate our findings, the bottom panels of Figure 4 plot the average winner and loser

prices, rather than price differences, along with the normalized industry index of the winner for

each quartile of contest duration. We add the industry index of the winner to better understand

the nature of abnormal performance, i.e. whether performance is strong or poor only relative to

the loser or also compared to the industry.6

The figures show that the effect of the merger on the acquiring firm’s stock performance varies

monotonically and significantly with the duration of the merger contest. In fact, the acquirors with

the shortest contest duration perform very well relative to the loser following merger completion.

While winner-loser price differences are virtually zero in the three years leading to the merger

contest, there is a sharp trend break in price paths following in the contest period. Winners

outperform losers by 36 percentage points in the three years following the merger. Despite

6We define industries according to the Fama-French 12 industry classification, and we obtain thecorresponding performance data from Ken French’s website (http://mba.tuck.dartmouth.edu/pages/fac-ulty/ken.french/data library.html)

21

the small sample (21 contests), the outperformance is statistically significant at the 10 percent

level. In contrast, in the subsample of the longest contest durations, winners fare much worse

than losers. Here, the cumulative underperformance of winners over three years is about 50

percentage points and also statistically significant at the 10 percent level. Thus the interquartile

range of underperformance is 86 percentage points. We also perform the same regressions with the

cumulative abnormal return as the dependent variable. This removes any performance differences

due to varying market exposure. The graphs shown in Figure A-2 show a very similar picture to

that which uses raw prices. The Q4-Q1 difference is only slightly smaller.

We obtain the statistical tests for the winner-loser differences by re-estimating regressions (1)

and (2), separately for each contest duration quartile of the sample. The respective difference

plots are shown in Panels B to E of Figure A-1 for raw prices and in Panels B to E of Figure A-2

for cumulative abnormal returns.

Interestingly, Figure 4 also shows that winners never outperform their industry significantly.

In the shortest contests outperformance is only significant relative to the loser, not relative to the

industry. In contrast, the winners of the longest merger fights, underperform both the loser and

their industry.

Table 5 reports the corresponding regression estimates of equation (2) for each of the four

quartiles of contest duration, and for both raw prices (leftmost columns) and cumulative abnormal

returns (rightmost columns). A statistical test of the Q4-Q1 difference in the long-term value

effect of the merger is economically large and highly statistically significant.

[Figure A-1 approximately here]


4.4 Economic Mechanism

The results so far show that post-merger performance differs substantially with the duration of

the contest. Losing is better than winning for the subsample of long-lasting bidding contest. But

for mergers with a short duration, e.g., the lowest quartile, the result reverses. In fact, under any

22

econometric specification, the results for the lowest and highest quartile are always significantly

different.

What explains these differences across fight duration quartiles? A first possible explanation

is that longer-lasting bidding wars increase the premium paid to target shareholders, and higher

premia may explain the worse performance of acquirors. However, our prior results indicate, for

the fourth quartile, an underperformance trend over the three years following merger completion

and, for the first quartile, an overperformance trend over the following three years. This long-term

trend is hard to explain with a one-time payment.

Nevertheless we investigate a possible causal role of differences in offer premia empirically.

For this analysis we use the target price run-up prior to the contest resolution to construct the

offer premium as described in Section 2.1. Therefore, we need to restrict the sample to the 66

winning bids for which the target stock price is available. The top panel of Figure 5 plots the offer

premium against the duration of the bidding contest (in months). We observe a weak positive

correlation. A simple linear regression of the offer premium on the duration of the merger contest

(not shown) reveals a weakly significant effect (p-value< 0.1): one additional month implies a

premium that is 1.74 percentage points higher. As the scatter plot also shows, however, the

positive correlation is driven by a few outliers with extreme fight durations.


In order to gauge the potential impact of overbidding on acquiror stock prices, we also estimate

the effect of contest duration on the offer premium expressed as a percentage of the acquiror’s

market valuation. This allows us to directly evaluate how much of the over- or underperformance

of the winner can be explained by contest-duration-induced additional payments. Recall that

we normalize bidder stock prices to 100 in the month prior to the beginning of the contest,

so that winner-loser differences in performance are effectively percentage differences in bidder

valuations. The bottom panel of Figure 5 shows a scatter plot of this relationship. The figure

show that, when the offer premium is expressed as a percentage of the acquiror value rather than

as a percentage of the target value, the correlation with contest duration becomes an order of

23

magnitude smaller. In an unreported regression of the offer premium and contest duration we

show that the relationship is statistically insignificant. Therefore, duration-induced overbidding

cannot explain the underperformance of long merger contests.

A second possible explanation of the large differences between bidder returns in the short-

duration and the long-duration subsamples relates closely to our initial motivation: Our analysis

aims to exploit bidding contests to construct a hypothetical counterfactual for acquiror perfor-

mance had the merger not taken place. In the case of bidding contest of short duration, however,

it is likely that the ultimate winner entered the contest as the likely winner, e.g., due to higher

synergies. In other words, winners and losers in short-duration contests of two to four months

may be significantly different along merger-relevant dimensions while winners and losers in long-

duration fights, which may last a year or more, are more similar. In that case, return differences

in the short-duration sample are not a good measure of the hypothetical counterfactual while the

long-duration estimates provide the correct estimate of the hypothetical counterfactual.

Measuring differences in synergies or, more broadly, differences in characteristics that affect

the returns to mergers, is difficult. Instead, we attempt to test for such differences by comparing

observable characteristics. First, we compare, one-by-one, the differences in characteristics be-

tween winners and losers, separately for the samples of long-duration and short-duration contests.

The results, reported in Table 2, Panel B, indicate some differences across the duration quartiles.

Size differences, whether measured in terms of book assets, market capitalization or sales, tend

to increase across from short to long-lasting contests. There is, however, evidence that bidders

in long-duration contests tend to be in the same industry.

Next, we aggregate these differences into a scalar distance measure. We compute the distance

between the multivariate characteristics of two bidders as the vector norm, D = x′V x, where

x = xW − xL is the vector of differences between winner and loser characteristics, and V is a

weighting matrix. We use a diagonal matrix for V , with the inverse of the variance as diagonal

elements. [TO BE DONE.]

We also examine whether characteristics of the completed deal, i.e. that of the winning bid,

exhibit differences across contest duration. Panel C of Table 2 shows a range of deal characteristics

24

for the aggregate sample and the four quartiles of contest duration. Two characteristics stands

out. First, long-duration contests are far more likely to be concluded with a stock-financed deal.

The median percentage of stock financing increases monotonically from 0 percent in the short

contest to 71 percent in the long contests. This pattern is in part explained the more extensive

documentation and filing necessary for stock offerings. Given these differences, it could be the

case that the means of financing explains the variation of performance across contest duration. In

fact, Loughran and Vijh (1997) find that stock mergers tend to have negative abnormal returns in

the five years following the merger. However, as we show below in Section 4.6, stock mergers do

not perform significantly worse than cash mergers or mergers with both stock and cash payments

in our sample of contested bids. Hence, the duration result cannot be driven by means of payment.

Second, contests with long-durations are for significantly larger targets than short contests.

The median transaction value increases from $146m to $648m from the first to the fourth quartile.

However, we show below in Section 4.6 that acquisitions of large targets do not perform worse

than those for small targets in our sample of contested bids. So target size also cannot explain

the duration result.

4.5 Comparison with Announcement Returns

An interesting question is whether the long-run divergence of winner and loser performance is

consistent with the sign and magnitude of short-run announcement returns. Because of the

difficulty of identifying a valid benchmark for long-run performance, announcement returns are

commonly viewed as a good market-based measure of the causal effect of merger.

Table 6 reports basic descriptive statistics of 3-day cumulative abnormal announcement re-

turns (CARs) of the winning bid. Panel A shows that the winners’ CARs of our sample of

contested bids are negative and economically large. The average CAR is -3.9 percent (median:

-2.3 percent). The announcement reaction is much more negative than for non-contested acquisi-

tions where CARs are typically slightly negative but greater than -1 percent and not significant.

Interestingly, Panel B shows that announcement returns do not vary systematically with the

length of the contest duration, as long-run winner-loser performance differences do.

25

In Table 7 we regress the long-run winner-loser performance difference on the winning bid’s

announcement return in order to explore whether the market is, on average, correct in its as-

sessment of the value effects of acquisitions. Surprisingly, we do not find a significant relation,

and the sign of the coefficient is even opposite of what would be expected if markets were on

average correct. These result hold whether the raw (column 1) or the risk-adjusted performance

difference (column 2) is used.

4.6 Robustness

In this section, we explore several alternative explanations for the heterogeneity observed in our

data. Mergers vary on a wide range of aspects, from bidder and target properties to deal charac-

teristics, and prior literature has shown - for single-bidder takeovers - that a number of charac-

teristics are significantly associated with long-term post-merger performance. In this section, we

explore whether these characteristics also correlate with long-term performance in our sample of

contested deals. If this were the case, our result that losers do better than winner in long-lasting

contest could be driven by these characteristics, given that some of these characteristics also vary

with contest duration.

Cash vs stock. We test whether the winner-loser comparison yields systematically different

results if we subsample by means of payment: cash, stock, and mixed payments. For example,

Loughran and Vijh (1997) find that stock mergers have negative abnormal returns relative to

size and market-to-book matched firm portfolios while cash mergers outperform the matched

firms in the five year period following deal completion. We run regressions (1) and (2) separately

for cash, mixed, and stock deals. The results are shown in Figure A-5. Interestingly, post-

merger underperformance is least pronounced for stock mergers, though the differences are not

statistically significant. This has two interesting implications. First, it shows that our duration

result cannot be explained by the fact that long-lasting contests tend to result in poorly performing

stock deals. Second, it suggests that losers may be a better counterfactual to winners than

characteristics-matched firms. Shleifer and Vishny (2003) argue that overvalued firms may use

their stock as cheap currency to buy less overvalued targets. Such mergers would tend to perform

26

poorly due to a mean reversion of the acquiror’s stand-alone valuation, not because the merger

is value-destroying. Our analysis reveals that stock mergers perform better than cash or mixed

mergers, when benchmarked against the losing contestant.

Hostile vs friendly. Since hostile bids are more prone to overbidding, we might expect the

underperformance of winners relative to losers to be more pronounced in the subsample of hostile

bids than among friendly takeover attempts. Thus, we run regressions (1) and (2) separately for

contests whose winning bids were hostile and for those that were friendly. Hostile bids tend to be

quite rare, as our resulting subsample of only of eight cases shows. The top two panels of Figure A-

6 illustrate the results for the two subsamples. As expected, hostile mergers exhibit a steep drop in

the winner’s stock price relative to loser’s around the contest period, and continue to underperform

in the long run. Despite the very small sample, this underperformance is statistically significant

at the 10 percent level. In contrast, friendly deals are followed by virtually zero performance

difference post-merger. Given the underperformance of hostile acquirors in our sample, it is

important to note that the finding of contest duration dependence of performance cannot be

driven by hostility. Hostile bids are more common in short contest rather than in long ones, as

Table ?? shows. So, the worse performance of hostile takeovers tends to weaken our duration

finding.

Acquiror Q. Prior research shows that high-Q acquirors underperform in the long run relative

to characteristics-matched firm portfolio (Rau and Vermaelen, 1998). To see whether such a

pattern is present in our data, we rank contests by the Tobin’s Q of the winning firm at the fiscal

year end preceding the beginning of the contest, and estimate equations (1) and (2) separately

for each quartile of acquiror Q. The results, illustrated in Figure A-7, show that in our data,

and using our winner-loser methodology, we find no significant performance differences across

acquiror’s Q. This result suggest that previous findings need to be interpreted with caution when

a proper counterfactual is not available. As mentioned above, Shleifer and Vishny (2003) suggest

that high-Q acquirors may be overvalued firms seeking to soften the reversal in valuation by

means of acquisitions. The poor post-merger performance of such firms may hence occur not

27

because of but despite the acquisition. When benchmarked against a valid counterfactual, such

mergers should not show any negative performance.

Firm size. Harford (2005) provides evidence of poor post-acquisition performance of large

acquirors. Since acquirors tend to be somewhat larger in long-duration contests (results not

reported), size effects could potentially explain our duration result. We thus examine sample splits

based on the market capitalization of the ultimate acquiror. Figure A-8 illustrates the results.

Again, we observe no significant differences in post-merger performance across size quartiles.

The graphs are, at best, suggestive of more negative announcement return for larger acquirors,

but there is no long-term underperformance. Thus, size effects do not explain why winners in

long-lasting contest underperform.

Relative deal size. We also use relative deal size (transaction value relative to the acquiror’s

market capitalization) as a sorting variable. This is to check whether acquisitions have a signifi-

cant effect only when they are large relative to the acquiror. Figure B-2 confirms that, even for

the large acquisitions in quartile three and four, winners do not perform significantly worse than

losers.

Number of Bidders. Another explanation is that acquirors do not account for the winner’s

curse, leading to greater overbidding if more bidders participate in the contest. Splitting the

sample into contests with two and contests with more than two bidders, however, suggests no

such pattern, see Figure B-3.

Diversification. We also analyze separately diversifying and non-diversifying mergers. A

merger is coded as diversifying if the winning bidder has a Fama-French 12-industry classification

that is different from the target’s. The corresponding graphs, however, suggest no such pattern,

see Figure B-4.

28

5 Conclusion

We argue that bidding contests help to address the identification issue caused by the missing

counterfactual in corporate acquisitions: In contests where at least two bidders have a significant

chance of winning, the post-merger performance of the loser allows to calculate the counterfactual

performance the winner would have had without the merger. We construct a novel data set of all

mergers with overlapping bids between 1983 and 2009. We find that the stock returns of bidders

are not significantly different before the merger contest, but diverge post-merger. In the full

sample, winners underperform losers over a three-year horizon, but there is large heterogeneity

and the effect is not significant. We then identify the subsample of deals where at least two

bidders have a significant chance at winning, exploiting long contest duration and a bidder’s

intermediate stock price reaction to the competing bid. We show that, for cases where either

bidder was ex ante likely to win the contest, losers outperform winners, while the opposite is true

in cases with a predictable winner.

29

t= 3 t= 2 t= 1 t=0 t=1 t=2 t=3Event time

First bid of company A

First bid of company B Completion of bid by company B

Withdrawal of bid by company A

(a) Stylized example.

01/95 02/95 03/95 04/95 05/95 06/95 07/95 08/95 09/95 10/95 11/95 12/95 01/96 02/96 03/96 04/96t=-2 t=-1 t=0 t=1 t=2 t=3

Loser: Westcott Communications; announced: 7 Apr 1995; withdrawn: 4 Jan 1996

Winner: Automatic Data Processing; announced: 5 June 1995; completed: 4 Jan 1996

(b) Example from our dataset.

Figure 1: Illustration of Event Time in a Merger Contest This figure illustrates the constructionof event time for merger contests. The top figure shows a stylized example, the bottom figure a concreteexample from our dataset.

30

0

0

02

2

24

4

46

6

68

8

8Number of contests

Num

ber o

f con

test

s

Number of contests1985

1985

19851990

1990

19901995

1995

19952000

2000

20002005

2005

2005Year

Year

Year

Figure 2: Frequency of Merger Contests over Time This figure shows the frequencydistribution of merger contest over the sample period 1985-2006. We count the calendar year ofthe starting month of the merger contest as the contest year.

31

-100

-100

-100-50

-50

-500

0

050

50

50100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period

PeriodPoint Estimate, Avg. Cumulative Abnormal Return of Winner

Point Estimate, Avg. Cumulative Abnormal Return of Winner

Point Estimate, Avg. Cumulative Abnormal Return of WinnerPoint Estimate, Avg. Cumulative Abnormal Return of Loser

Point Estimate, Avg. Cumulative Abnormal Return of Loser


(a) Pre-merger run-up of winner: 1st quartile

-100

-100

-100-50

-50

-500

0

050

50

50100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period






(b) Pre-merger run-up of winner: 2nd quartile

-100

-100

-100-50

-50

-500

0

050

50

50100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period






(c) Pre-merger run-up of winner: 3rd quartile

-100

-100

-100-50

-50

-500

0

050

50

50100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period






(d) Pre-merger run-up of winner: 4th quartile

Figure 3: Pre-merger Run-up The graphs show the cumulative abnormal returns (CARs)of winning and losing bidders in the three years around the merger contest by quartile of thewinner’s pre-merger abnormal return (run-up). Quartile 1 (top left figure) contains the winnerswith the worst abnormal performance in the 3-year pre-merger period. Quartile 4 (bottom rightfigure) contains the winners with the strongest pre-merger performance. Cumulative abnormalreturns are computed by first fitting the model rit− rft = αi+βi(rmt− rft) + εit, for each bidder,and for the pre- and post-merger period separately, and then cumulating abnormal performance:CARit = (1 + CARit−1) · (1 + ARit) − 1, where ARit = rit − rft − βi(rmt − rft) and CARit isnormalized to zero at t = 0..

32

60

60

6080

80

80100

100

100120

120

120140

140

140160

160

160-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period

PeriodPoint Estimate, Avg. Winner Price

Point Estimate, Avg. Winner Price

Point Estimate, Avg. Winner PricePoint Estimate, Avg. Loser Price

Point Estimate, Avg. Loser Price

Point Estimate, Avg. Loser PricePoint Estimate, Avg. Industry Price

Point Estimate, Avg. Industry Price


(a) Full sample

60

60

6080

80

80100

100

100120

120

120140

140

140160

160

160-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period








(b) Contest duration: 1st quartile

60

60

6080

80

80100

100

100120

120

120140

140

140160

160

160-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period








(c) Contest duration: 2nd quartile

60

60

6080

80

80100

100

100120

120

120140

140

140160

160

160-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period








(d) Contest duration: 3rd quartile

60

60

6080

80

80100

100

100120

120

120140

140

140160

160

160-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period








(e) Contest duration: 4th quartile

Figure 4: Pre- and Post-Merger Peformance The figure shows the pre- and post-mergerstock price performance of winning and losing bidders and the average performance in the win-ner’s industry. The winner’s industry is defined according to the Fama-French 12 industryclassification, and the corresponding performance data are obtained from Ken French’s website(http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data library.html). The five figurescorrespond to the entire sample (top figure) and the different quartile of contest duration (bottomfigures). Prices are normalized to 100 in the month preceding the start of the contest.

33

100

010

020

030

040

0O

ffer p

rem

ium

[% o

f tar

get]

0 10 20 30 40Duration of merger contest

Fitted values

100

010

020

030

040

0O

ffer p

rem

ium

[% o

f acq

uiro

r]

0 10 20 30 40Duration of merger contest

Fitted values

Figure 5: Offer Premium The figure shows scatter plots of the offer premium against contestduration. In the left figure, the offer premium is expressed in percentage of the target’s equitycapitalization. It is computed as the percentage run-up in the target stock price from one monthbefore the beginning of the merger contest until completion. In the right figure, the offer premiumis expressed as a percentage of the acquiror’s equity market capitalization. Contest duration isexpressed in months. The fitted values are the predictions of an OLS regression of the offerpremium on the contest duration.

-200

-200

-200-100

-100

-1000

0

0100

100

100200

200

200Winner-loser price difference at t=36

Win

ner-

lose

r pric

e di

ffer

ence

at

t=36

Winner-loser price difference at t=36-.2

-.2

-.2-.15

-.15

-.15-.1

-.1

-.1-.05

-.05

-.050

0

0.05

.05

.05Announcement CAR of winner

Announcement CAR of winner

Announcement CAR of winnerFitted values

Fitted values

Fitted values -200

-200

-200-100

-100

-1000

0

0100

100

100W-L difference of risk-adj. performance at t=36

W-L

diff

eren

ce o

f risk

-adj

. per

form

ance

at

t=36

W-L difference of risk-adj. performance at t=36-.2

-.2

-.2-.15

-.15

-.15-.1

-.1

-.1-.05

-.05

-.050

0

0.05

.05

.05Announcement CAR of winner

Announcement CAR of winner

Announcement CAR of winnerFitted values

Fitted values

Fitted values

Figure 6: Comparison with Announcement Returns The figure shows scatter plots ofthe winner-loser performance difference three years after the acquisition against the three-daycumulative abnormal announcement return (CAR) of the winning bid. In the left figure, long-runperformance is measured in terms of the raw (normalized) price. In the right figure, long-runperformance is measured as the cumulative abnormal return (CAR). CARs are computed by firstfitting the model rit − rft = αi + βi(rmt − rft) + εit, for each bidder, and for the pre- and post-merger period separately, and then cumulating abnormal performance: CARit = (1 +CARit−1) ·(1 +ARit) − 1, where ARit = rit − rft − βi(rmt − rft) and CARit is normalized to zero at t = 0.The fitted values are the predictions of an OLS regression of the long-run winner-loser differenceon the CAR.

34

Tab

le1:

Sam

ple

Con

stru

cti

on

Sam

ple

sele

ctio

ncr

iter

ion

Con

test

sB

ids

Bid

der

sW

inn

ers

Los

ers

Init

ial

sam

ple

193

416

402

154

248

less

rep

eate

db

ids

ofsa

me

bid

der

for

sam

eta

rget

193

402

402

154

248

less

conte

sts

that

hav

en

ot

bee

nco

mp

lete

d14

831

031

014

816

2le

ssb

idd

ers

wit

hou

tC

RS

PP

ER

MN

O14

629

829

814

215

6le

ssco

nte

sts

wh

ere

win

ner

was

par

ent

com

pan

y13

628

028

013

214

8le

ssb

idd

ers

wit

hm

issi

ng

stock

retu

rns

inp

erio

ds

-1to

113

427

027

013

213

8le

ssb

idd

ers

wit

hex

trem

ep

rice

vola

tili

ty(S

td>

200)

132

263

263

128

135

less

bid

der

sw

ith

mis

sin

gp

rice

dat

ain

the

even

tw

ind

ow[-

35,+

36]

114

207

207

104

103

less

bid

der

sin

conte

sts

wh

ere

conte

stan

tis

mis

sin

g:82

172

172

8290

35

Tab

le2:

Desc

rip

tive

Sta

tist

ics

Pan

elA

:B

idd

erC

har

acte

rist

ics

Win

ner

sL

oser

sM

ean

Med

ian

Std

NM

ean

Med

ian

Std

N

Tota

lass

ets

[$m

]14

930.

5833

26.2

438

841.

3579

9078

.59

2440

.52

1678

3.93

88M

arke

tca

pit

ali

zati

on

[$m

]20

987.

7446

76.1

749

163.

3379

1302

2.12

2840

.48

2653

3.84

88S

ale

s[$

m]

5676

.67

1835

.20

1242

0.49

7933

77.2

110

90.0

158

63.7

087

Tob

in’s

Q1.

881.

341.

5079

1.79

1.19

1.41

88P

P&

E0.

270.

220.

2478

0.27

0.20

0.26

86P

rofi

tab

ilit

y0.

120.

110.

0978

0.13

0.12

0.12

86B

ook

leve

rage

0.25

0.21

0.21

790.

230.

200.

1784

Mar

ket

leve

rage

0.18

0.15

0.16

790.

180.

140.

1684

An

nou

nce

men

tC

AR

[%]

-3.8

9-2

.32

6.69

76-3

.78

-3.3

64.

1686

An

nou

nce

men

tC

AR

[%]

-Q

1-3

.40

-2.7

63.

4121

-3.3

6-2

.88

3.39

23A

nn

ou

nce

men

tC

AR

[%]

-Q

2-5

.70

-2.3

211

.07

18-2

.61

-2.6

73.

5420

An

nou

nce

men

tC

AR

[%]

-Q

3-3

.20

-1.2

05.

6920

-4.2

2-3

.36

5.00

22A

nn

ou

nce

men

tC

AR

[%]

-Q

4-3

.40

-2.7

34.

6917

-4.8

9-5

.19

4.41

21A

nn

ou

nce

men

tC

AR

[Dol

lar]

-295

.62

-22.

4912

36.7

176

-211

.10

-30.

4368

5.72

86A

nn

ou

nce

men

tC

AR

[Dol

lar]

-Q

1-9

0.82

-29.

8512

8.35

21-6

3.61

-41.

9216

623

An

nou

nce

men

tC

AR

[Dol

lar]

-Q

2-1

64.4

6-2

1.67

513.

9218

-94.

76-1

8.52

312.

3120

An

nou

nce

men

tC

AR

[Dol

lar]

-Q

3-1

50.4

2-1

6.52

441.

0220

-261

.89

-34.

6390

0.35

22A

nn

ou

nce

men

tC

AR

[Dol

lar]

-Q

4-8

58.2

9-2

6.38

2489

.99

17-4

30.2

3-5

4.43

965.

6421

36

Tab

le2:

Desc

rip

tive

Sta

tist

ics

-C

onti

nu

ed

Pan

elB

:W

inn

er-L

oser

Diff

eren

ces

inC

har

acte

rist

ics

Med

ian

Diff

eren

ceby

Con

test

Du

rati

onM

ean

Med

ian

Std

NQ

1Q

2Q

3Q

4

Tot

alas

sets

[$m

]54

85.8

119

6.86

3139

1.29

7820

9.94

116.

7792

.90

1213

.69

Mark

etca

pit

aliz

atio

n[$

m]

5987

.80

153.

5740

037.

0478

288.

5211

1.27

-271

.64

3062

.34

Sal

es[$

m]

1531

.02

27.3

211

605.

7377

27.3

211

6.15

-12.

5827

6.39

Tob

in’s

Q0.

050.

011.

5178

-0.0

2-0

.02

-0.0

40.

12P

P&

E-0

.01

0.00

0.15

76-0

.03

0.02

0.00

0.00

Pro

fita

bil

ity

-0.0

2-0

.01

0.13

76-0

.02

-0.0

1-0

.01

0.00

Book

leve

rage

0.01

-0.0

10.

2175

-0.0

30.

07-0

.02

-0.0

3M

ark

etle

vera

ge

-0.0

10.

000.

1775

0.00

0.03

-0.0

2-0

.06

An

nou

nce

men

tC

AR

[%]

-0.2

60.

416.

4076

0.13

-0.6

00.

931.

14

Pan

elC

:D

eal

Ch

arac

teri

stic

s

Med

ian

by

Con

test

Du

rati

onM

ean

Med

ian

Std

NQ

1Q

2Q

3Q

4

Tra

nsa

ctio

nva

lue

[$m

]34

36.5

234

4.83

1240

0.92

8114

8.36

229.

851

8.26

648.

45T

end

eroff

er0.

340.

000.

4882

10

00

Hos

tile

0.10

0.00

0.30

820

00

0P

erce

nta

ge

pai

din

stock

36.9

48.

0443

.21

820

050

.35

71.4

2P

erce

nta

ge

pai

din

cash

43.9

330

.76

44.7

382

100

43.8

10

10.7

Nu

mb

erof

bid

der

s2.

432.

000.

7282

22

22

Off

erp

rem

ium

[%of

targ

et]

57.8

242

.93

63.0

267

67.4

451

.75

40.3

435

.97

Off

erp

rem

ium

[%of

acq

uir

or]

9.66

6.64

21.2

667

3.39

6.15

13.1

66.

73C

onte

std

ura

tion

(month

s)9.

487.

507.

3782

46

1015

.5

37

Table 3: Tests of Identifying Assumption

Panel A: Alpha Regressions

Dept. variable: CAPM Alpha - LoserPre-merger Period Post-merger Period

CAPM Alpha - Winner 0.364*** 0.175*(0.095) (0.101)

Constant 0.008*** -0.002(0.002) (0.002)

Observations 82 82R-squared 0.154 0.036

Panel B: Beta Regressions

Dept. variable: CAPM Beta - LoserPre-merger Period Post-merger Period

CAPM Beta - Winner 0.366*** 0.329***(0.109) (0.103)

Constant 0.619*** 0.712***(0.142) (0.126)


Panel C: Residual Regressions

Dept. variable: Market model residual - LoserPre-merger Period Post-merger Period

Market model residual - Winner 0.169*** 0.227***(0.029) (0.057)

Constant -0.000 0.000(0.000) (0.000)


38

Tab

le4:

Regre

ssio

ns

Est

imati

ng

the

Eff

ect

of

aM

erg

er

Dep

end

ent

vari

ab

le:

Pri

ceC

um

ula

tive

abn

orm

alre

turn

(1)

(2)

(3)

(4)

(5)

(6)

Win

ner

(α1)

0.12

3-3

.913

-4.0

53-3

.432

-7.7

80-8

.470

(2.9

21)

(4.3

07)

(4.2

33)

(5.6

48)

(6.9

88)

(7.0

89)

Per

iod

(α2)

0.87

1***

0.58

9***

-0.4

07-0

.297

-0.5

87*

-0.7

24*

(0.1

40)

(0.1

45)

(0.3

89)

(0.2

89)

(0.3

11)

(0.4

09)

Win

ner

xP

erio

d(α

3)

-0.2

35-0

.251

-0.2

53-0

.636

-0.6

70-0

.663

(0.1

82)

(0.1

62)

(0.1

60)

(0.4

97)

(0.5

12)

(0.5

02)

Post

mer

ger

(α4)

8.24

82.

187

-6.8

96-0

.321

-4.3

65-4

.810

(6.0

13)

(5.2

64)

(5.6

86)

(5.4

37)

(4.9

89)

(5.5

13)

Win

ner

xP

ost

mer

ger

(α5)

-4.5

182.

494

2.81

10.

332

5.67

57.

693

(5.4

37)

(5.4

75)

(5.4

79)

(6.5

39)

(6.8

79)

(7.1

74)

Per

iod

xP

ost

mer

ger

(α6)

0.25

70.

861*

*1.

035*

**0.

241

0.86

5*0.

828*

(0.2

92)

(0.3

44)

(0.3

40)

(0.3

16)

(0.4

74)

(0.4

81)

Win

ner

xP

ost

mer

ger

xP

erio

d(α

7)

0.20

20.

0897

0.10

80.

649

0.56

10.

559

(0.3

17)

(0.3

03)

(0.3

04)

(0.5

24)

(0.5

58)

(0.5

60)

Ln

(mar

ket

cap

)10

.60*

**9.

845*

**8.

688*

**8.

723*

**(2

.503

)(2

.419

)(2

.153

)(2

.051

)T

ob

in’s

Q8.

135*

**7.

456*

**6.

711*

**8.

362*

**(2

.168

)(1

.963

)(2

.428

)(2

.101

)P

rofi

tab

ilit

y-4

2.32

*-3

9.80

*-5

2.90

**-6

0.07

***

(21.

88)

(21.

24)

(23.

95)

(22.

43)

Mar

ket

leve

rage

-91.

66**

*-8

5.70

***

-63.

88**

*-5

8.63

***

(22.

83)

(21.

46)

(22.

77)

(19.

72)

Con

test

fixed

effec

tsX

XX

XX

XY

ear

fixed

effec

tsX

X

Iden

tifi

cati

on:

Poi

nt

esti

mat

eofα1

+36α3

-8.3

39-1

2.96

-13.

15-2

6.32

-31.

91-3

2.33

Iden

tifi

cati

on:

P-v

alu

eofα1

+36α3

0.33

50.

137

0.12

80.

253

0.19

20.

182

Mer

ger

effec

t:P

oin

tes

tim

ate

ofα1

+α5

+36

(α3

+α7)

-5.5

99-7

.238

-6.4

61-2

.619

-6.0

34-4

.517

Mer

ger

effec

t:P

-valu

eofα1

+α5

+36

(α3

+α7)

0.56

40.

457

0.50

60.

690

0.38

20.

513

Ob

serv

atio

ns

1238

411

432

1143

212

384

1143

211

432

R-s

qu

ared

0.35

60.

493

0.53

50.

217

0.35

60.

396

Nu

mb

erof

conte

sts

8282

8282

8282

39

Tab

le5:

Regre

ssio

ns

Est

imati

ng

the

Merg

er

Eff

ect

Dep

end

ent

Vari

able

:P

rice

Cu

mu

lati

veA

bn

orm

alR

etu

rn

Qu

arti

leof

Conte

stD

ura

tion

:Q

1Q

2Q

3Q

4Q

1Q

2Q

3Q

4

Win

ner

(α1)

2.14

70.

837

1.68

4-5

.250

2.63

0-1

.823

0.28

9-1

6.37

(3.8

92)

(3.9

86)

(4.3

17)

(9.4

89)

(3.7

20)

(5.5

48)

(4.1

89)

(22.

42)

Per

iod

(α2)

1.08

6***

0.72

1**

0.99

5***

0.67

2*-0

.007

70-0

.153

-0.2

39-0

.749

(0.1

44)

(0.3

27)

(0.2

51)

(0.3

64)

(0.2

49)

(0.4

50)

(0.5

36)

(0.8

70)

Win

ner

xP

erio

d(α

3)

0.02

69-0

.380

-0.1

52-0

.498

0.11

8-0

.859

-0.0

937

-1.8

85(0

.189

)(0

.459

)(0

.309

)(0

.513

)(0

.321

)(0

.770

)(0

.595

)(1

.846

)P

ost

mer

ger

(α4)

-1.0

521.

338

7.97

123

.18

-7.5

91-0

.804

7.29

3-0

.247

(7.8

30)

(7.5

24)

(8.7

22)

(18.

76)

(7.0

92)

(7.4

73)

(10.

13)

(15.

66)

Win

ner

xP

ost

mer

ger

(α5)

0.20

0-2

.383

-8.7

47-5

.147

3.08

72.

781

-8.1

764.

296

(8.6

66)

(7.9

23)

(7.7

71)

(16.

45)

(7.3

77)

(8.2

30)

(7.2

39)

(22.

82)

Per

iod

xP

ost

mer

ger

(α6)

-0.6

120.

735

0.49

50.

465

-0.5

870.

185

0.40

80.

921

(0.4

41)

(0.5

49)

(0.6

44)

(0.6

61)

(0.3

69)

(0.6

15)

(0.6

02)

(0.8

42)

Win

ner

xP

ost

mer

ger

xP

erio

d(α

7)

0.80

30.

398

0.15

0-0

.546

0.43

70.

846

0.11

91.

361

(0.5

63)

(0.6

57)

(0.4

71)

(0.8

28)

(0.4

84)

(0.8

26)

(0.6

92)

(1.9

01)

Conte

stfi

xed

effec

tsX

XX

XX

XX

X

Iden

tifi

cati

on:α1

+36α2

3.11

6-1

2.83

-3.7

73-2

3.17

6.86

3-3

2.73

-3.0

83-8

4.24

Iden

tifi

cati

on:

P-v

alu

eofα1

+36α2

0.69

10.

506

0.77

30.

403

0.53

40.

321

0.89

10.

352

Mer

ger

effec

t:α1

+α5

+36(α3

+α7)

32.2

3-0

.906

-7.1

32-4

7.97

25.6

80.

520

-6.9

76-3

0.96

Mer

ger

effec

t:P

-val

ue

ofα1

+α5

+36

(α3

+α7)

0.07

40*

0.96

60.

582

0.04

40**

0.01

1***

0.97

10.

557

0.04

7**

Mer

ger

effec

t:D

iffer

ence

lon

g/sh

ort

figh

td

ura

tion

-80.

200

-56.

640

Mer

ger

effec

t:P

-valu

eof

lon

g/sh

ort

figh

td

ura

tion

0.00

4***

0.00

1***

Ob

serv

atio

ns

3240

2808

3168

3168

3240

2808

3168

3168

R-s

qu

ared

0.41

30.

411

0.48

10.

258

0.21

00.

336

0.20

50.

220

Nu

mb

erof

conte

sts

2219

2120

2219

2120

40

Table 6: Announcement Returns of Winning Bids in Merger ContestsThis table reports statistics for 3-day cumulative abnormal returns (CARs) for the winning bid inmerger contests. The CARs are computed from residuals of the model rijt− rft = αij +βij(rmt−rft) + εijt. The parameters are estimated using monthly returns of the 3-year pre-merger period.CAR[−1,+1] is the 3-day cumulative abnormal return around the announcement of the winningbid. Dollar CAR [scaled] is the dollar gain of the winning bid scaled by the transaction value.

Panel A: Winners’ CARs

CAR [-1,+1] Dollar CAR [scaled]Mean -0.0389 -0.2133Median -0.0232 -0.0801N 76 75

Panel B: Winners’ CARs by contest duration

Contest Duration CAR [-1,+1] Dollar CAR [scaled] N1 -0.0339 -0.3286 212 -0.0570 -0.0959 183 -0.0320 -0.1738 204 -0.0340 -0.2391 17

Table 7: Long-Run Winner-Loser Performance Difference and 3-day CARs This tablereports regressions of the winner-loser performance difference 3 years after the merger completion,on the 3-day cumulative abnormal announcement return of the winning bid. The winner-loserperformance difference is measured as the raw price difference at t = 36 months (column 1), andas the cumulative abnormal return (CAR) at t = 36 (column 2). Cumulative abnormal returnsare computed by first fitting the model rit − rft = αi + βi(rmt − rft) + εit, for each bidder,and for the pre- and post-merger period separately, and then cumulating abnormal performance:CARit = (1 + CARit−1) · (1 + ARit) − 1, where ARit = rit − rft − βi(rmt − rft) and CARit isnormalized to zero at t = 0.

Dept. variable: W/L Price Difference at t=36 W/L CAR Difference at t=36

Announcement CAR of winner -43.51 -104.3(208.2) (155.6)

Constant 4.900 -2.004(11.92) (8.905)


41

A Appendix

42

3020

100

1020

40 20 0 20 40Period

Point Estimate, Winner Loser DifferenceUpper Confidence BoundLower Confidence BoundPiecewise Linear Approximation

(a) Winner-loser difference, entire sample.

100

500

5010

0

40 20 0 20 40Period


Note: Number of contests: 21

(b) Winner-loser difference, Q1 of contest duration.

100

500

5010

0

40 20 0 20 40Period



(c) Winner-loser difference, Q4 of contest duration.

100

500

5010

0

40 20 0 20 40Period

Point Estimate, Winner Industry DifferenceUpper Confidence BoundLower Confidence BoundPiecewise Linear Approximation


(d) Winner-industry difference, Q1 of contest dura-tion.

100

500

5010

0

40 20 0 20 40Period

Point Estimate, Winner Industry DifferenceUpper Confidence BoundLower Confidence BoundPiecewise Linear Approximation


(e) Winner-industry difference, Q4 of contest dura-tion.

Figure A-1: Pre- and Post-merger Performance This figure illustrates the winner-loser andwinner-industry performance difference around merger contests. The top figure is for the entiresample. The two figures in the second row illustrate winner-loser differences for quartile one andfour of contest duration. The bottom row shows winner-industry differences for the same quartilesone and four. The dots connected by the solid black line correspond to the point estimates of theπ-coefficients of regression (1). The dots connected by the solid gray lines mark the respective95% confidence intervals. The piecewise-linear approximation given by the straight, solid line isobtained from regression (2).

43

-100

-100

-100-75

-75

-75-50

-50

-50-25

-25

-250

0

025

25

2550

50

5075

75

75100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period

PeriodPoint Estimate, Avg. Winner Price [Market-adjusted]

Point Estimate, Avg. Winner Price [Market-adjusted]

Point Estimate, Avg. Winner Price [Market-adjusted]Point Estimate, Avg. Loser Price [Market-adjusted]

Point Estimate, Avg. Loser Price [Market-adjusted]


(a) Full sample

-100

-100

-100-75

-75

-75-50

-50

-50-25

-25

-250

0

025

25

2550

50

5075

75

75100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-100-1

00-100-75

-75

-75-50-5

0-50-25

-25

-2500

02525

255050

507575

7510010

0100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-100

-100

-100-75

-75

-75-50

-50

-50-25

-25

-250

0

025

25

2550

50

5075

75

75100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-100

-100

-100-75

-75

-75-50

-50

-50-25

-25

-250

0

025

25

2550

50

5075

75

75100

100

100-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







Figure A-2: Cumulative Abnormal Performance around Merger Contests This figureillustrates the cumulative abnormal returns for winners and losers for a three-year window aroundmerger contests. Cumulative abnormal returns are computed by first fitting the model rit−rft =αi + βi(rmt − rft) + εit, for each bidder, and for the pre- and post-merger period separately,and then cumulating abnormal performance: CARit = (1 + CARit−1) · (1 + ARit) − 1, whereARit = rit − rft − βi(rmt − rft) and CARit is normalized to zero at t = 0. The five figurescorrespond respectively to the entire sample (top figure) and the different quartiles of contestduration (bottom figures). The dots connected by the solid black line correspond to the pointestimates of the average winner performance, the solid gray lines to the average loser erformance.

44

-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period






(a) Full sample

-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-50-5

0-50-25

-25

-2500

02525

255050

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







Figure A-3: Market-adjusted Pre- and Post-merger Performance This figure illustratesthe pre- and post-merger stock price performance for winners and losers adjusted by marketperformance. The five figures correspond respectively to the entire sample (top figure) and thedifferent quartiles of contest duration (bottom figures). Winner and loser prices are normalizedto zero in the month preceding the start of the contest, and are adjusted by market performance,i.e. we plot pt = pt−1(1 + rit− rmt). The dots connected by the solid black line correspond to thepoint estimates of the average winner prices, the solid gray lines to the average loser prices.

45

-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period






(a) Full sample

-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-50-5

0-50-25

-25

-2500

02525

255050

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







-50

-50

-50-25

-25

-250

0

025

25

2550

50

50-40

-40

-40-20

-20

-200

0

020

20

2040

40

40Period

Period







Figure A-4: Industry-adjusted Pre- and Post-merger Performance This figure illustratesthe pre- and post-merger stock price performance for winners and losers adjusted by industryperformance. The five figures correspond, respectively, to the entire sample (top figure) and thedifferent quartiles of contest duration (bottom figures). Winner and loser prices are normalized tozero in the month preceding the start of the contest, and are adjusted by industry performance,i.e. we plot pt = pt−1(1 + rit − rkt), where k indexes the bidder’s Fama-French (12) industryclassification. The dots connected by the solid black line correspond to the point estimates of theaverage winner prices, the solid gray lines to the average loser prices.

46

6040

200

2040

40 20 0 20 40Period


Note: Method of payment: Cash; Contests: 27

(a) Method of payment: All cash.

4020

020

40

40 20 0 20 40Period


Note: Method of payment: Cash and stock; Contests: 36

(b) Method of payment: Cash and stock.

500

50

40 20 0 20 40Period


Note: Method of payment: Stock; Contests: 19

(c) Method of payment: All stock

Figure A-5: Winner-Loser Performance Difference by Method of Payment This figureillustrates the winner-loser difference in pre- and post-merger performance for all-cash, mixed,and stock mergers. The dots connected by the solid black line correspond to the point estimates ofthe π-coefficients of regression (1). The dots connected by the solid gray lines mark the respective95% confidence intervals. The piecewise-linear approximation given by the straight, solid line isobtained from regression (2).

47

150

100

500

50

40 20 0 20 40Period


Note: Attitude: Hostile; Contests: 8

(a) Attitude: Hostile.

4020

020

40

40 20 0 20 40Period


Note: Attitude: Friendly; Contests: 70

(b) Attitude: Friendly.

Figure A-6: Winner-Loser Performance Difference by Bidder Attitude and DealType This figure illustrates the winner-loser difference in pre- and post-merger performance fortakeover contests in which the prevailing bid was hostile versus friendly (top panel), and wherethe winning bid was a tender offer versus a merger (bottom panel). The dots connected by thesolid black line correspond to the point estimates of the π-coefficients of regression (1). The dotsconnected by the solid gray lines mark the respective 95% confidence intervals. The piecewise-linear approximation given by the straight, solid line is obtained from regression (2).

48

4020

020

40

40 20 0 20 40Period


Note: Quartile of acquiror Tobin’s Q = 1; Contests: 20

(a) Quartile 4 of Tobin’s Q: High Q.

500

5010

0

40 20 0 20 40Period



(b) Quartile 3 of Tobin’s Q.

100

500

50

40 20 0 20 40Period



(c) Quartile 2 of Tobin’s Q.

100

500

50

40 20 0 20 40Period



(d) Quartile 1 of Tobin’s Q: Low Q.

Figure A-7: Winner-Loser Performance Difference by Tobin’s Q of the WinningBidder This figure illustrates the winner-loser difference in pre- and post-merger performancefor contests sorted by the Tobin’s Q of the acquiring, i.e. the winning firm. The sorting is basedon the pre-contest Tobin’s Q, based on data from the fiscal year ending prior to the start of thebidding contest. The dots connected by the solid black line correspond to the point estimates ofthe π-coefficients of regression (1). The dots connected by the solid gray lines mark the respective95% confidence intervals. The piecewise-linear approximation given by the straight, solid line isobtained from regression (2).

49

4020

020

40

40 20 0 20 40Period


Note: Quartile of acquiror market capitalization = 4; Contests: 19

(a) Quartile 4 of market capitalization: Large firms.

500

50

40 20 0 20 40Period



(b) Quartile 3 of market capitalization.

4020

020

4060

40 20 0 20 40Period



(c) Quartile 2 of market capitalization.

100

500

50

40 20 0 20 40Period



(d) Quartile 1 of market capitalization: Small firms.

Figure A-8: Winner-Loser Performance Difference by Market Capitalization of theWinning Bidder This figure illustrates the winner-loser difference in pre- and post-mergerperformance for contests sorted by the market capitalization of the acquiring, i.e. the winningfirm. The sorting is based on the pre-contest market capitalization, based on data from the fiscalyear ending prior to the start of the bidding contest. The dots connected by the solid black linecorrespond to the point estimates of the π-coefficients of regression (1). The dots connected by thesolid gray lines mark the respective 95% confidence intervals. The piecewise-linear approximationgiven by the straight, solid line is obtained from regression (2).

50

B Online Appendix60

8010

012

014

016

0

40 20 0 20 40Period

Point Estimate, Avg. Loser PricePoint Estimate, Avg. Winner Price

(a) Winner vs loser

6080

100

120

140

160

40 20 0 20 40Period

Point Estimate, Avg. Winner PricePoint Estimate, Avg. Industry PricePoint Estimate, Avg. Market Price

(b) Winner vs market and industry

Figure B-1: Pre- and Post-merger Stock Price Performance of Winners against Vari-ous Benchmarks This figure illustrates pre- and post-merger stock price performance of winnersin takeover contests against various benchmarks. The left figure plots winner and loser prices.The right figure plots winner prices along with industry and market index levels. Prices and indexlevels are normalized to 100 in the month preceding the start of the contest. The dots connectedby the black (gray) lines correspond to the point estimates of the average prices (index levels).

51

100

500

50

40 20 0 20 40Period


Note: Quartile of transaction value / acquiror assets = 4; Contests: 19

(a) Quartile 4 of deal value to acquiror market cap-italization.

500

50

40 20 0 20 40Period



(b) Quartile 3 of deal value to acquiror market cap-italization.

4020

020

4060

40 20 0 20 40Period



(c) Quartile 2 of deal value to acquiror market cap-italization.

6040

200

20

40 20 0 20 40Period



(d) Quartile 1 of deal value to acquiror market cap-italization.

Figure B-2: Winner-loser Performance Difference by Quartile of Relative Deal SizeThis figure illustrates the winner-loser difference in pre- and post-merger performance for contestssorted by the ratio of deal value to market capitalization of the acquiror. Market capitalizationof the winning bidder is computed for the fiscal year ending prior to the start of the biddingcontest. The deal value is the dollar amount of the final effective offer. The dots connectedby the solid black line correspond to the point estimates of the π-coefficients of regression (1).The dots connected by the solid gray lines mark the respective 95% confidence intervals. Thepiecewise-linear approximation given by the straight, solid line is obtained from regression (2).

52

4020

020

40 20 0 20 40Period


Note: Number of bidders = 2; Contests: 55

(a) Two bidders.

4020

020

40

40 20 0 20 40Period


Note: Number of bidders > 2; Contests: 27

(b) More than two bidders.

Figure B-3: Winner-loser Performance Difference for Contests with Two and MoreBidders This figure illustrates the winner-loser difference in pre- and post-merger performancefor contests, separately for contest with two bidders and those with more than two bidders. Thedots connected by the solid black line correspond to the point estimates of the π-coefficients ofregression (1). The dots connected by the solid gray lines mark the respective 95% confidenceintervals. The piecewise-linear approximation given by the straight, solid line is obtained fromregression (2).

53

100

500

50

40 20 0 20 40Period


Note: Diversifying Mergers; Contests: 17

(a) Diversifying mergers.

2010

010

2030

40 20 0 20 40Period


Note: Non Diversifying Mergers; Contests: 65

(b) Concentrating mergers.

Figure B-4: Winner-loser Performance Difference for Diversifying and ConcentratingMergers This figure illustrates the winner-loser difference in pre- and post-merger performancefor contests, separately for diversifying and concentrating mergers. A merger is classified asdiversifying if the winning bidder is in a Fama-French 12-industry group that is different from thetarget’s, and concentrating otherwise. The dots connected by the solid black line correspond tothe point estimates of the π-coefficients of regression (1). The dots connected by the solid graylines mark the respective 95% confidence intervals. The piecewise-linear approximation given bythe straight, solid line is obtained from regression (2).

54

References

Asquith, P., R. F. Bruner, and D. W. J. Mullins (1987): “Merger Returns and the Form

of Financing,” Proceedings of the Seminar on the Analysis of Security Prices, 64, 115–146.

Betton, S., B. E. Eckbo, and K. Thorburn (2008): “Corporate Takeovers,” .

Boone, A. L. and J. H. Mulherin (2007): “How are Firms Sold?” The Journal of Finance,

62, 847–875.

Brennan, M. J., T. Chordia, and A. Subrahmanyam (1998): “Alternative factor specifi-

cations, security characteristics, and the cross-section of expected stock returns,” Journal of

Financial Economics, 49, 345373.

Greenstone, M., R. Hornbeck, and E. Moretti (2011): “Identifying Agglomeration

Spillovers: Evidence from Winners and Losers of Large Plant Openings,” Journal of Politi-

cal Economy.

Jensen, M. (1986): “Agency Cost of Free Cash Flow, Corporate Finance, and Takeovers,”

American Economic Review Proceedings, 76, 323–329.

Loughran, T. and A. M. Vijh (1997): “Do Long-Term Shareholders Benefit from Corporate

Acquisitions?” Journal of Finance, 52, 1765–1790.

Malmendier, U. and G. Tate (2008): “Who Makes Acquisitions? CEO Overconfidence and

the Market’s Reaction,” Journal of Financial Economics, 89, 20–43.

Mitchell, M., T. Pulvino, and E. Stafford (2004): “Price Pressure around Mergers,”

Journal of Finance, 59.

Moeller, S. B., F. P. Schlingemann, and R. M. Stulz (2005): “Wealth Destruction on a

Massive Scale? A Study of Acquiring-Firm Returns in the Recent Merger Wave,” Journal of

Finance, 60.

55

Morck, R., A. Shleifer, and R. Vishny (1990): “Do Managerial Objectives Drive Bad

Acquisitions,” The Journal of Finance, 45, 31–48.

Rau, P. R. and T. Vermaelen (1998): “Glamour, Value, and the Post-Acquisition Perfor-

mance of Acquiring Firms,” Journal of Financial Economics, 49, 223–253.

Rhodes-Kropf, M. and S. Viswanathan (2004): “Market Valuation and Merger Waves,”

The Journal of Finance, 59, 2685–2718.

Roll, R. (1986): “The Hubris Hypothesis of Corporate Takeovers,” Journal of Business, 59,

197–216.

Shleifer, A. and R. W. Vishny (2003): “Stock Market Driven Acquisitions,” Journal of

Financial Economics.

56

Cash is KingWhat the Market Learns about Targets through Merger Bids�

PRELIMINARY AND INCOMPLETE. COMMENTS WELCOME.

Ulrike MalmendierUC Berkeley

[email protected]

Marcus Matthias OppUC Berkeley

[email protected]

Farzad SaidiNew York [email protected]

March 31, 2011

Abstract

Returns to merger announcements are commonly used to measure the expected value createdby mergers. We provide evidence that a signi�cant portion re�ects, instead, a revaluationof the target. Using a sample of withdrawn deals from 1980 to 2008, we show that tar-gets of cash o¤ers are revalued by +15% post-withdrawal. Stock bids, on the other hand,do not seem to provide target information: targets with equity o¤ers fall back to theirpre-announcement levels after withdrawal. The results are not driven by future takeoveractivity since cash targets are insigni�cantly less likely to receive future merger bids. Theresults are also independent of the speci�c type of withdrawal reason. Our �ndings, as wellas the observed value changes in acquirers, are consistent with cash bids indicating targetundervaluation while stock bids signal acquirer overvaluation.

JEL classi�cation: G14, G34, D03, D82

Keywords: mergers & acquisitions, private information, misvaluation, cash o¤er, stock o¤er

�We are extremely grateful to comments and suggestions by Malcolm Baker, Alberto Bisin, Slava Fos, XavierGabaix, Alessandro Lizzeri, Gustavo Manso, Atif Mian, Terrance Odean, Antoinette Schoar, Andrei Shleifer,David Sraer, Jeremy Stein, Je¤rey Wurgler, as well as seminar participants at Princeton, LSE, and NYU Stern.Research assistance by Javed Ahmed and Marlena Lee on an earlier draft of the paper was greatly appreciated.

1

1 Introduction

Much of the research on mergers and acquisitions revolves around the question: do mergerscreate value? With mergers being among the most important and, possibly, most disruptiveevents in a corporation�s lifetime, this question has been of foremost interest to policymakersand researchers alike, including a recent debate about �massive wealth destruction� throughmergers in the late 90s and at the beginning of this century (Moeller, Schlingemann, and Stulz(2005)). Empirically, the measurement of value creation is challenging: long-run studies mayconfound the causal e¤ect of the merger with other return-relevant events. Short-run studiescan, under e¢ cient market assumptions, measure the expected value creation at announcement,and are statistically most precise (see Barber and Lyon (1997) or Fama (1998)). In fact, variousstudies document a small positive combined announcement return of targets and bidders, andinterpret this �nding as evidence of value creation.1

The validity of this assessment hinges on the assumption that the o¤er reveals little privateinformation about the stand-alone value of the two entities. In this paper, we document thatthis is not the case. We provide evidence that a bid reveals economically and statisticallyimportant information about the target. While the negative information e¤ect of stock bidson the bidder�s stand-alone value is well understood at least since Myers and Majluf (1984),we model, theoretically, information revelation about the target (under-)valuation and show,empirically, that there is a large revaluation e¤ect for cash targets (+15%) but no e¤ect forstock targets.

Our identi�cation strategy uses the sample of withdrawn deals to measure informationrevelation: we compare the value of targets prior to the announcement and after withdrawal.To see the intuition for our approach, consider the following thought experiment. Supposea target company is currently trading at $100 and obtains a takeover o¤er in cash for $125.Moreover, assume that the deal fails within a short time period. After the failure of the deal, thevalue implications of the takeover o¤er and the confounding e¤ect of the takeover probabilityare no longer incorporated in the target stock price. If the stock price trades post-withdrawalat $115, we can attribute $15 to the revision of beliefs about the stand-alone value of the targetas a result of the o¤er.

Our empirical analysis is motivated by a signi�cant body of existing theories that predictthat the bidder�s private information will be re�ected in his choice of acquisition currency,i.e., cash or stock.2 We provide a simple model that extends Shleifer and Vishny (2003) byallowing the market to draw inferences about misvaluations revealed through takeover bids andthe choice of acquisition currency. The model predicts that the use of stock is more likely ifthe (hypothetical) combined entity of the target and the bidder is overvalued, and the use ofcash if it is undervalued. A bidder willing to buy an overvalued target with stock as long asthe target is less overvalued than the bidder�s own stock. And a bidder prefers to use cashfor undervalued targets because it enables him to fully extract the long-run price appreciation.By receiving a �xed payment, target shareholders in cash deals do not bene�t from a positivelong-run revaluation. Upon announcement of the deal and the method of exchange, the market

1 See Jensen and Ruback (1983), Jarrell, Brickley, and Netter (1988), or Andrade, Mitchell, and Sta¤ord (2001).2 In Fishman (1989), bidders use cash o¤ers for valuable targets to preempt competing o¤ers. If the �rst biddersignals a su¢ ciently high valuation for the target by using cash, bidder 2 does not �nd it worthwhile to starta bidding war. In Hansen (1987) as well as the extension by Eckbo, Giammarino, and Heinkel (1990), thebidder�s choice of the medium of exchange re�ects private information about his own value.

2

price partially incorporates this strategic behavior of the bidder into the stock price. If adeal fails, the information revealed through the choice of the medium of exchange is no longerconfounded with merger e¤ects, including the split up of surplus between target and acquirerand match-speci�c synergies. This enables us to identify information revelation in the sampleof withdrawn deals.

Figure 1: Announcement and Withdrawal E¤ectsFigure 1: CARs from -25 (before announcement) to +25 (after withdrawal), normalized

Note (Figures 1-3): 95 cash and 131 stock deals.

Figure 2: BHARs from -25 (before announcement) to +25 (after withdrawal), normalized

‐0.15

‐0.1

‐0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

‐30 ‐20 ‐10 0 10 20 30 40 50 60 70 80

Cash target CAR

Stock target CAR

Cash acquirer CAR

Stock acquirer CAR

Announcement

Withdrawal

‐0.15

‐0.1

‐0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

‐30 ‐20 ‐10 0 10 20 30 40 50 60 70 80

Cash target BHAR

Stock target BHAR

Cash acquirer BHAR

Stock acquirer BHAR

Announcement

Withdrawal

Notes: Cumulative Abnormal Returns (CARs) from 25 trading days pre-announcement to 25trading days post-withdrawal. The sample consists of 95 cash and 131 stock deals. See Table1 for the construction of the sample.

Figure 1 previews our key empirical results. It plots the evolution of cumulative returnsin the sample of withdrawn all-cash and all-stock deals between 1980 and 2008, separatelyfor acquirers and targets. The pre-announcement and post-withdrawal di¤erences are striking.Starting 25 trading days prior to the announcement, we observe a run-up among targets of stocko¤ers and targets of cash o¤ers, yielding announcement e¤ects of 15% (stock) and 25% (cash).At the withdrawal date, however, which is normalized to occur 50 (synthetic) trading days laterin the graph,3 the value of stock targets falls below the pre-announcement level, to which itultimately returns, while the value of cash targets remains signi�cantly higher than prior tothe bid: cash targets earn 15% cumulative abnormal returns relative to the pre-announcementlevel. Hence, despite the small upward trend in both cash and stock targets post-withdrawal,stock target returns remain more than 15% below cash target returns. In addition, we �nd thatstock acquirers trade at signi�cantly lower prices post-withdrawal (�10%) while cash acquirers3 For deals where the withdrawal occurs later than 50 days after announcement, we cumulated returns to matchthe proportion of elapsed trading days (e.g., the �rst trading day in the normalized period is equal to thecumulative return of the �rst two trading days for a deal withdrawn after 100 trading days). For deals wherethe withdrawal occurs earlier than 50 days after announcement, we use a linear approximation to match theproportions of elapsed trading days (e.g., the �rst two trading days in the normalized period are each equalto half the return of the �rst trading day for a deal withdrawn after 25 trading days).

3

return to their pre-announcement level, consistent with �ndings in previous studies.4

Our results indicate that cash bids induce the market to positively revaluate the target.One explanation for the revaluation is positive private information of the acquirer about thetarget, which is revealed in the cash bid. A related explanation is limited attention, i.e., bidsdrawing investors�attention to the target company and inducing them to process informationabout the target that was already available. Since, in both cases, the market learns from thebidder�s action, we will dub both interpretations information revelation in a broad sense. Analternative explanation is that the value increase in cash targets does not re�ect a revaluationof the targets themselves but of them being lucrative merger opportunities relative to stocktargets. However, we �nd that failed cash deals are insigni�cantly less likely than failed stockdeals to be followed by another takeover attempt over the next two years.

We address several concerns about our identi�cation strategy. A �rst possible concernis that transactions are not canceled randomly or, more speci�cally, that withdrawal is notexogenous with respect to target value. However, such exogeneity is also not required. Ouridenti�cation of a stock versus cash e¤ect solely requires that post-announcement news about anentity do not di¤erentially a¤ect the likelihood of completion in the stock and cash samples. Forexample, if positive news about targets makes deals more likely to be completed, the endogeneityof completion does not a¤ect the observed di¤erences between failed stock and failed cashdeals. One important channel for di¤erential e¤ects are �nancial constraints of the acquirer:intermediate positive news about the target may require the acquirer to adjust the bid upwards,and this may be harder for cash than for stock deals if the acquirer is �nancially constrained.Using several measures of �nancial constraints, we �nd no signi�cant e¤ect of the latter. Whilewe are able to dismiss a pure sample selection e¤ect based on the �nancial constraints measuresemployed in this paper, we cannot completely rule out the �nancial constraints channel due tothe lack of availability of more precise measures. In addition, it is worth emphasizing that theestimated revaluation e¤ects are, of course, speci�c to the sample of withdrawn mergers. Evenif there are no systematic di¤erences between the withdrawal of stock and cash bids, in termsof their relation to target value, we still cannot generalize the speci�c revaluation estimatebeyond the sample of withdrawn mergers. Only exogenous cancelations (with respect to targetvalues) allow us to tackle this identi�cation issue.5 Despite these limitations, the magnitudesof the observed valuation di¤erences recipients of cash and of stock bids indicate that mergervaluations cannot neglect the impact of information revelation on announcement returns andmake it worthwhile investigating the economic mechanism behind the sample selection.

Beyond providing new empirical facts about the informational content of merger bids, our�ndings have important implications for the e¢ ciency of mergers and acquisitions. From apolicymaker�s perspective, it is crucial to know whether these deals quantitatively a¤ect totalsurplus (allocative role), as in the Q-theory model of Jovanovic and Rousseau (2002), or whetherthey merely have distributional consequences, such as in the model of Shleifer and Vishny(2003), including value adjustment due to the revelation of information. While total valuecreation might be the most relevant measure of the desirability of merger activities, the (ex-postirrelevant) split-up of surplus in�uences the ex-ante incentive to engage in value-maximizingtransactions. If target shareholders can extract all the surplus from a transaction, a value-

4 For example, Savor and Lu (2009) compare the stock price performance of stock bidders in the withdrawnsample to that of stock bidders in the completed sample, and �nd that bidders in the withdrawn sample doworse. Further evidence on stock market driven acquisitions is provided in Friedman (2006).

5 We are currently working on identifying a subsample of deal withdrawals unrelated to target valuations.

4

maximizing bidder would have no incentive to start a takeover attempt. This is the powerfullogic of Grossman and Hart (1980). The rationale applies regardless of the source of thesurplus, that is, also if the bidder simply has private information about the value of the target:if a bidder�s private information about the target is fully revealed to the market and targetshareholders would, once the bidder has announced his bid, demand the full surplus, biddersmay abstain from bidding in the �rst place. An initial stake in the target company (see Shleiferand Vishny (1986) and Hirshleifer and Titman (1990)) dampens this mechanism as the bidderwould bene�t from a revaluation of his initial stake. In summary, the division of surplus,including a valuation increase due to information revelation, a¤ects ex-ante incentives and thusthe e¢ ciency of the takeover market mechanism.

In addition to the research cited above, our �ndings relate to several strands of the priorliterature. A large literature evaluates the returns to mergers and acquisitions. Most studiesbased on announcement returns �nd that tender o¤ers generate small overall value with virtuallyall the gains accruing to the target (see overview papers by Jensen and Ruback (1983), Andrade,Mitchell, and Sta¤ord (2001), and Betton, Eckbo, and Thorburn (2008)). Using probabilityscaling methods, Bhagat, Dong, Hirshleifer, and Noah (2005) �nd larger estimates of combinedvalue creation (7:3%) than the conventional CAR estimate (5:3%) by Bradley, Desai, and Kim(1988). Following Travlos (1987), studies of announcement returns distinguish between cashand stock transactions. Consistent with the pecking order model of Myers and Majluf (1984),bidders using stock reveal negative private information about themselves, which manifests itselfin negative announcement returns for the bidder.6

Long-run post-takeover performance studies by Rau and Vermaelen (1998) and Loughranand Vijh (1997) document strong negative abnormal returns for bidders of stock transactionsand positive abnormal returns for bidders using cash. Apart from facing statistical issues incomputing long-run abnormal returns (see Barber and Lyon (1997), Fama (1998), and Brav(2000)), these post-takeover studies face a fundamental logical problem: if markets are e¢ cient,even a value-destructive merger should be correctly incorporated in the price after completionof the deal. Thus, stock bidders should not exhibit a negative CAR. In an e¢ cient market,long-run empirical studies would solely pick up the extent of asset pricing model misspeci�ca-tion. If stock markets are not e¢ cient, it is not clear how to use stock market data to makesurplus assessments. Malmendier, Moretti, and Peters (2010) address this identi�cation issueby comparing the long-run returns of competing bidders in contested mergers.

Most closely related to our analysis are earlier papers by Dodd (1980), Bradley, Desai, andKim (1983), and Davidson, Dutia, and Cheng (1989), which examine failed acquisitions to studyinformation revelation about the target. For example, Davidson, Dutia, and Cheng (1989) �ndthat targets of withdrawn deals trade higher than before the announcement only because theyare more likely to become future targets, whereas targets that do not obtain a future o¤errevert to pre-announcement levels. They interpret this result as evidence against informationrevelation about the stand-alone value of the target. In contrast to our analysis, their studydoes not distinguish between cash and stock transactions. This conditioning information allowsus to separate targets for which (almost) no information is revealed (stock targets) and targetsthat are revalued (cash targets). Our analysis indicates that this e¤ect is not driven by cashtargets being more prone to becoming takeover targets in the future. Moreover, our results arerobust to controlling for hostile deals even though these deals are more likely to be �nanced

6 Jovanovic and Braguinsky (2004) generate bidder discounts in a Q-theory framework without resorting to theinformational content revealed through the medium of exchange.

5

with cash. Thus, the disciplinary channel of hostile bids (see Franks and Mayer (1996)) doesnot drive our results either.7

The remainder of the paper is organized as follows. In Section 2, we motivate our empiricalanalysis with a simple model of mergers building on the framework of Shleifer and Vishny (2003),but allowing investors to process the information contained in merger bids. That section alsodiscusses the assumptions underlying our identi�cation strategy. In Section 3, we describethe selection and composition of our data set. Section 4 contains the results of our empiricalanalysis, and Section 5 concludes.

2 Theoretical Framework

2.1 Model

We consider an asymmetric information environment in which a bidder, B, possesses privateinformation about his own value, the value of a potential target, T , and potential synergiesbetween the two �rms. Our setting allows both for the possibility that stocks are (on average)fairly valued and for systematic misvaluation. In either case, the assumption of asymmetricinformation only requires that some �rms have access to information that is not fully re�ectedby the stock price. For ease of exposition, we assume that there are no information asymmetries,instead, between the bidder and the target. Intuitively, these assumptions could be motivatedby industry-speci�c knowledge that is shared by the bidder and the target share but is notknown to the market (yet).

Building on the modeling framework and notation of Shleifer and Vishny (2003), we studyhow market misvaluations a¤ect the probability of merger bids and the form of payment, i.e.,the choice between a cash and a stock acquisition. However, in contrast to Shleifer and Vishny(2003), we allow market participants to learn from actions that are (potentially) motivatedby misvaluations, including the form of payment o¤ered by the bidder. This learning channelgenerates testable predictions for our empirical examination of stock market reactions to cashand stock o¤ers.

For expositional reasons, we express market valuations of �rm i, Vi, in terms of valuationmultiples, qi, and the capital stock in place, Ki:

Vi = qiKi (1)

where i 2 fB; T;Mg and M refers to the merged �rm. The capital stock is observable tothe market, so that potential initial misvaluations must be caused by false market assessmentsof the valuation multiple. The fair long-run multiple of the �rms is denoted by qi whereasthe initial time-0 market valuation is given by q0i . The true long-run synergies are given bysKM = s (KB +KT ) whereas the market estimates synergies of sKM .

The timeline (see Figure 2) is as follows:

7 Bhagat, Shleifer, Vishny, Jarrel, and Summers (1990) interpret the hostile takeover activity in the 1980s mostlyas a return to corporate specialization and deconglomeration.

6

Figure 2: Timeline of the Model

- - -t = 0 t = 1 t = 2 t = 3

Players learn qi

��

QQQQQQQQQQQQ

a = S

a = C

a = N

!!!!

!!!

aaaaaaa

!!!!

!!!

aaaaaaa

t

tt

t

t

�VM = (qM + s)KM

�Vi

�VM = (qM + s)KM

�Vi

�Vi

t

t

t

t

qi(C)

qi(S)

p

1� p

p

1� p

t

tt

t

0. The bidder and the target learn the fair values of their �rms, qi, including the value ofthe merged �rm.

1. The bidder can choose between three possible actions, a. The market updates the valua-tion multiple and synergies estimate after observing a, denoted by qi (a) and s (a):

(a) All-Cash acquisition, a = C

(b) All-Stock acquisition, a = S

(c) No acquisition, a = N .

2. The deal fails with probability 1� p.

3. Market prices adjust back to long-run values (indicated by �Vi).

Since periods 1 and 2 are close events, we do not incorporate an additional learning stageafter the withdrawal. Eventually, long after the withdrawal (in period 3); market prices revertto fundamentals. As in Shleifer and Vishny (2003) (henceforth SV), we assume that the tar-get management is only concerned about short-run valuations whereas bidder management isconcerned about long-run valuations.8 Consequently, target shareholders will accept any o¤erprice PKT as long as P > qT (a). Assuming that the target o¤er premium, P , is identical forcash and stock o¤ers, a cash payment of PKT corresponds to a stock deal with an exchangeratio such that a fraction x of the merged company is owned by target shareholders:

x =P

qM (S) + s (S)

KTKM

: (2)

8 Our model yields the original SV model for the following parameters: p = 1, qi (C) = qi (S) = q0i , qB = qT = q,s (C) = s (S) > 0, and s = 0.

7

Intuitively, the target share is the ratio of the takeover premium multiple, P , relative to thecombined �rm valuation multiple, s (S) + qM (S), times the fraction of the capital stock thatthe target contributes to the merged �rm. Note that this speci�cation is consistent with ourdata (which will be discussed further in Section 3): the premia do not vary signi�cantly be-tween the cash and stock deal samples (comprising successful and withdrawn deals).9 Further-more, we know that the updated valuation multiple of the merged company qM (S) is equal toqB (S)

KBKB+KT

+ qT (S)KT

KB+KT. Deal failure is modeled exogenously with respect to the cash

or stock choice. Section 2.2 critically discusses this important assumption.

With a slight abuse of notation, we denote the long-run value of shares owned by theacquiring company�s shareholders as �VB, also in the cases where the deal goes through:

Lemma 1 The long-run market value of shares owned by the bidder�s shareholders is:

�VB (S) = (1� x) (qM + s)KM (3)�VB (C) = (qM + s)KM � PKT (4)�VB (N) = qBKB: (5)

Note that if the transaction fails, the long-run market valuation of the bidder is simply thelong-run stand-alone value, �VB (N).

Proposition 1 Given the updating functions of the market, qM (S) and s (S), a bidder choosesstock over cash if:

qM (S) + s (S)| {z }Post-Stock O¤er Combined Firm Valuation

> qM + s| {z } :Long-run Combined Firm Valuation

(6)

A transaction is made at all as long as:

sKM �((P � qT )KT for qM + s > qM (S) + s (S)�P qM+sqM (S)+s(S)

� qT�KT otherwise

: (7)

Proof: Since the payo¤s in case of failure (and the probabilities of failure) are the same forstock and cash transactions, the bidder only needs to compare the valuations upon a successfuldeal (see Lemma 1). The bidder chooses stock if:�

1� PKT(qM (S) + s (S))KM

�(qM + s)KM > (qM + s)KM � PKT : (8)

Simple algebra yields the desired result.

Thus, stock is more likely to be used if both the target and the bidder are overvalued or if themarket is too optimistic about synergies, i.e., s (S) > s. Intuitively, a transaction is made ifeither the true synergies, sKM , are high or if the target can be purchased at a discount (P < qT ).For stock bidders, the price P is shaded by the long-run devaluation ratio qM+s

qM (S)+s(S)< 1.

As long as the combined �rm is currently overvalued (relative to market expectations), i.e.,

9 Theoretically, the e¤ective value of the o¤er could be a function of the medium of exchange.

8

qM (S) + s (S) > qM + s, the long-run e¤ective price of a stock o¤er is lower. Note that bothstock and cash bidders would prefer to buy the target at a discount (P < qT ). In that case,the deal adds value for the bidder regardless of the method of payment. However, the choicebetween cash and stock depends on the combined overvaluation of the bidder and the target. Ifthe bidder is undervalued, he would not like to let the target shareholders bene�t from his ownundervaluation. Therefore, he uses cash. Bidders in stock deals are willing to overpay targets(P > qT > qT ) if and only if their own overvaluation is even higher (or synergies are high).

Since the bidder takes the updating functions of the market, qi (a) and s (a), as given,these functions were treated as exogenous for the decisionmaking behavior of the bidder. If theupdating functions are rational, they must be consistent with the optimal policy function of thebidder. As a result, if our model fully captured all payo¤-relevant dimensions for the bidder�sdecision problem, his using stock would signal that the market valuation of the combined �rm(including synergies) is too high. This could not be consistent with an equilibrium (in thePerfect Bayesian sense). Assuming partial updating by the market, the market will update allterms in the same direction (but not fully) upon seeing a stock o¤er. De�ne the overvaluationof a �rm as:

vi (a) = qi (a)� qi: (9)

Then, under the assumption that all priors are independent, the learning component of ourmodel predicts:

Corollary 1 Partial updating implies:

vi (S) � vi (C) (10)

s (S)� s � s (C)� s: (11)

Note that this information e¤ect cannot be cleanly identi�ed in the sample of successful deals.Targets that receive a takeover price P , whether in stock or cash, will simply converge to theo¤ered takeover price (until the deal is sealed). In contrast, the sample of withdrawn dealsallows us to separately identify the information e¤ect on the stand-alone valuation multiples.At the time of the withdrawal, match-speci�c synergies should no longer be priced, so that thedi¤erence between the post-withdrawal price and the pre-announcement price should re�ect theinformational content revealed through the medium of exchange.

2.2 Identi�cation

The model (see Corollary 1) predicts that the expected overvaluation of a target, vT , conditionalon a stock o¤er is greater than the expected target overvaluation conditional on a cash o¤er:

E (vT (S)) > E (vT (C)) : (12)

In order to isolate the information revealed about the target, we examine the subsample ofwithdrawn takeovers (W = 1). For now, assume that deals fail for exogenous reasons (as in themodel). After a deal is withdrawn, the deal-related synergies or overpayment by this particularbidder no longer a¤ect the target value. Our data (cf. Figure 1) are consistent with the followingconditional hypothesis:

E (vT (S)jW = 1) > E (vT (C)jW = 1) : (13)

9

If deals fail for exogenous reasons (unrelated to vT ), then the conditional statement in thewithdrawn sample automatically implies the unconditional statement in the full sample.

Suppose that deals fail endogenously (related to vT ), then our results are biased if thereis a di¤erential impact of target information about vT on the success probability of the dealdepending on whether the deal is in cash or stock. To see this, it is useful to decompose theexpected target overvaluation as:

E (vT (a)) =Pr (W = 1j a)E (vT (a)jW = 1) (14)

+Pr (W = 0j a)E (vT (a)jW = 0) :

Our identi�cation assumption is that the withdrawal selection does not mechanically generatethe following relationship:

E (vT (S)� vT (C)jW = 1) > 0 > E (vT (S)� vT (C)jW = 0) : (15)

Put di¤erently, it would be problematic if the most undervalued targets (vT < 0) within thecash sample (C) were more likely to be withdrawn (W = 1). Likewise, it would be problematicif the most undervalued targets within the stock sample (S) were more likely to be completed(W = 0). Note that it is (in theory) irrelevant whether the probability of withdrawal is di¤erentgiven a cash or a stock o¤er.10

Thus, any story that threatens our identi�cation requires di¤erential e¤ects of new informa-tion on the success probability for cash vs. stock deals. One such hypothetical story would bebased on �nancial constraints of the acquirer. We will revisit our identi�cation after we presentthe main empirical results.

3 Data Description

Our main sources are CRSP, Compustat, and SDC. We match CRSP market data with targetsand acquirers in the SDC database using the six-digit CUSIP provided in the SDC database.In determining which CUSIP identi�er to use from the CRSP database, we always choose theCUSIP with the lowest possible 7th digit (typically 1). Our underlying assumption is that ear-lier issuances make for greater proportions of market capitalization (note that, however, hardlyany entity in our �nal regression sample is a¤ected by this assumption). Regarding the dealsin the SDC database, we drop those for which the announcement and/or execution/withdrawaldates are missing. Furthermore, we also drop deals which were withdrawn on the day of theirannouncement, as well as deals with announcement dates after their execution/withdrawal (thiscriterion is labeled as "valid deal dates" in Table 1).

Sample selection. We focus on deals that were executed/withdrawn within at least �veand at most 250 trading days. While economically sensible, this criterion for the deal window isnot very restrictive. Out of 12; 025 deals that ful�ll the criteria listed so far, only 657 deals aredropped due to the deal window restriction. Note that, for e¤ective deals, targets sometimesstop trading well before the announcement date, so we use the last trading day to calculatethe number of trading days between deal announcement and execution. Lastly, we restrict our

10 It turns out empirically that cash deals are not more likely to be withdrawn than stock deals after controllingfor other observables.

10

analysis to deals for which no competing o¤ers were outstanding. The conventional rationalefor this is to avoid capturing new deal announcements when calculating returns after an o¤erfor the same target was withdrawn.

Table 1 summarizes the sample construction outlined above, and displays the compositionof the �nal regression sample of successful and withdrawn deals.

Variable de�nitions. In our analysis, we use the following return measures:

CARit =tXj=1

(rij � rmj) (16)

1 +BHRit =

tYj=1

(1 + rij) (17)

1 +BHARit =

tYj=1

(1 + rij)�tYj=1

(1 + rmj) (18)

where rij and rmj denote �rm i�s equity return and the CRSP value-weighted market returnat time j, respectively.

As can be seen in Figures 1, 3, and 4, our main �nding is robust to whether we choosecumulative abnormal returns (CARs), buy-and-hold returns (BHRs), or buy-and-hold abnormalreturns (BHARs). Therefore, we will henceforth use cumulative abnormal returns, but can atrequest provide results for the other return measures which we do not report in this paper.11

Other important variables in our analysis are the �rms�q ratios and Kaplan and Zingales(1997) indices. The former is de�ned as the market value of equity plus assets minus the bookvalue of equity all over assets. As a measure of �nancial constraints, we use the four-variableversion of the KZ index given in Lamont, Polk, and Saa-Requejo (2001) and Baker, Stein, andWurgler (2003), namely:

KZit = �1:002CFitAi;t�1

� 39:368DIVitAi;t�1

� 1:315 CitAi;t�1

+ 3:139LEVit (19)

where CFit, DIVit, Cit, and LEVit denote cash �ows, cash dividends, cash balances, andleverage, respectively, and Ai;t�1 is the �rm�s lagged assets.

Summary statistics. The summary statistics are in Tables 2a and 2b. We summarize thecharacteristics of successful and withdrawn deals separately for the entire regression sample inTable 2a, and then focus on the subsample of withdrawn all-cash and all-stock deals in Table2b.

In general, successful and withdrawn deals are similar along many dimensions. Naturally,they di¤er (and signi�cantly so at the 1% level) in their time to execution/withdrawal, theproportion of hostile deals, and the ratio between the target�s and acquirer�s market values ofequity: deals take longer to be executed than to be withdrawn, withdrawn deals are more likelyto be hostile, and � compared to the acquirers�market capitalization � the targets�market

11 Figure 5 also plots the CARs for 100 days pre- and post-announcement. The results are robust.

11

values of equity are higher in withdrawn than in successful deals. Most importantly, the �rms�q ratios and KZ indices do not vary signi�cantly between the successful and withdrawn samples.While, at face value, withdrawn deals seem more likely to be �nanced with stock rather thancash, the regression analysis in Table 3 will reveal that this di¤erence can be explained by dealcharacteristics (most notably the log of the relative deal size).

The all-cash and all-stock subsamples add up to more than two-thirds of the total regressionsample, revealing that the majority of deals do not involve hybrid �nancing structures. As canbe seen in Table 2b, among withdrawn deals,12 all-cash deals di¤er from all-stock ones in thatthe former yield higher premia (signi�cant at the 5% level), and involve acquirers and targetswith lower q ratios (signi�cant at the 1% level). Lastly, withdrawn cash deals are also morelikely to be hostile (signi�cant at the 1% level).

We conclude that the withdrawn sample is not systematically di¤erent from the sampleof successful deals, and that �conditional on other deal characteristics � stock deals are notdisproportionately more often withdrawn than cash deals, for which we will provide furtherevidence in the following section.

4 Empirical Results

Before measuring the impact of cash bids on post-withdrawal returns, we �rst investigatewhether the medium of exchange has explanatory power for the withdrawal of a deal. Tothis end, we estimate a linear probability model, and regress a binary withdrawal variable ona continuous cash variable, which indicates what fraction of the total payment was o¤ered incash. The results are presented in Table 3. Without any controls, cash deals are less likely tobe withdrawn. However, after controlling for the relative deal size, the impact of the medium ofexchange on withdrawal becomes insigni�cant. Note furthermore that, in the last two columns,the absolute coe¢ cient on cash drops by two-thirds once we control for announcement returns,which backs up the idea that any signi�cance of the cash variable is likely due to other deal-related covariates.13 The announcement return of the target, CAR (A� 25; A+ 1), shouldcontrol for market expectations of deal failure that are based on publicly available informationat the time of the announcement (but are unavailable to the econometrician).14 Our results arealso robust to reducing our sample to the subset of pure cash and stock deals only (cf. TableA.3). In conjunction with the summary statistics, we can now rest assured that we may focuson withdrawn deals in the remainder of the analysis.

Table 4 presents our main result: bids with higher cash fractions are associated with higherpost-withdrawal target CARs compared to their pre-announcement levels.15 Note that cumula-tive abnormal returns can be meaningfully compared across deals with di¤erent window lengths

12 The characteristics of successful all-cash and all-stock deals can be found in Table A.1.13 Further correlates of cash deals are investigated in Table A.2.14 The announcement return should approximately be: CART (a) � p�Premium+(1� p) �Learning(a). Since wecontrol for the premium and learning encoded in the choice of medium of exchange, a, variations in CARshould capture variations in the deal probability.

15 In regressions unreported in this paper, we �nd that our results are robust to using industry rather thanmarket returns for the speci�cation of abnormal returns.

12

as long as the underlying equilibrium asset pricing model is correctly speci�ed.16 The e¤ect iseven stronger than suggested in Figure 1, which can be inferred from the estimates in the �rsttwo columns without cash interaction e¤ects.

In the second and third columns �besides the q ratios, the ratio of the target�s to acquirer�smarket value of equity, and the relative deal size � we control for deal characteristics thatare correlated with the medium of exchange and potentially re�ect the target�s stand-alonevalue. First, we control for the o¤er premium (over the target�s share price one month priorto the bid). Moreover, we include leveraged buyouts, some of which are management buyouts,for these deals are naturally in cash and involve well-informed bidders. However, due to theseldom occurrence of LBOs in our data, the respective coe¢ cient �although it has the predictedpositive sign �is imprecisely estimated. We also control for the disciplinary channel of hostilebids, which are more likely to be in cash (see summary statistics in Table 2b), but their positiveimpact on target revaluation is not robust.

In the fourth column of Table 4, we add the KZ index to control for �nancial constraints. Wedo so because post-announcement news about the target might di¤erentially a¤ect �nanciallyconstrained acquirers o¤ering cash rather than stock in that cash o¤ers are more likely to bewithdrawn upon positive target news that may require the bidder to upgrade his bid. If �nancialconstraints of the acquirer were driving our results, one would expect that the interaction of themedium of exchange with the KZ index shows up signi�cantly. However, this is not the case.Despite a rich variety of deal- and entity-speci�c controls, only the medium of exchange andthe deal premium consistently matter for target CARs. Intuitively, these two variables re�ectthe value of the target.

Our results are robust in the subsample of pure cash and stock deals (see last two columnsof Table 4), where the cash e¤ect seems stronger for targets with low q ratios. This canbe interpreted as re�ecting our theoretical prediction that the revaluation process is moreemphasized for potentially undervalued targets once cash is o¤ered for them.

4.1 Robustness Checks

Generalizing our cash-versus-stock result to all mergers hinges on the assumption that thereare no systematic di¤erences between cash and stock deals in the withdrawn as opposed to thesuccessful sample. As alluded to above, �nancial constraints on the acquirer�s side could leadto such di¤erences. While we �nd no such e¤ect in Table 4, it might be insightful to decomposecash and stock deals by their withdrawal reasons. For this purpose, we collected deal synopsesas provided in the SDC database, categorized the withdrawal reasons for the subsample ofpure cash and stock deals, and report their relative frequencies in Table 5. The �rst threecategories �acquirer withdrawal, target rejection, and mutual consent �summarize the mostcommon reason for deal withdrawal, namely failure to �nd an agreement. Note that acquirersmay decide to withdraw deals due to (anticipated) target rejection, so the classi�cation is notunambiguous/exclusive in these cases.

Deals that are withdrawn due to target defense represent cases in which the board and/or

16 Due to the relatively short time length of our event window (see summary statistics in Table 2a), the mis-speci�cation of the asset pricing model to compute "normal" returns is a second-order concern. Detaileddiscussions of the statistical issues with calculating long-run returns are given by Barber and Lyon (1997),Brav (2000), and Fama (1998).

13

the shareholders actively fought o¤ a merger attempt, which suggests that the target mayhave believed it was undervalued. According to our discussion in Section 2.2, the inclusion ofsuch deals would bias our estimates if they were more likely to be in cash than in stock. Onthe other hand, self-selection in the stock sample seems innocuous. Prima facie, stock dealsappear more prone to withdrawals as a reaction to market revaluations (here, news may relateto the acquirer as well as the target as, from the theory, we know that overvaluation of thecombined �rm makes a stock deal more likely). That is, overvalued targets (and acquirers)may be relatively overrepresented in the subsample of successful rather than withdrawn stockdeals. This would in turn imply that the di¤erence in target overvaluation between stock andcash deals that we capture in the withdrawn sample should actually be a lower bound (in theabsence of any further self-selection in the cash sample).

Overall, only two withdrawal reasons are signi�cantly more frequent for cash rather thanstock deals, namely deals withdrawn due to target rejection and due to active target defense.As already argued, these are potentially deals for which our identi�cation assumption may nothold. In Table A.4, we re-run the regressions of the �fth and sixth columns of Table 4 fordi¤erent subsets of pure cash and stock deals, and �nd that our results are robust to droppingdeals that are withdrawn due to target rejection, active target defense, or both.

In Table 6, we present further robustness checks of our main �nding. We include a toeholdvariable to examine whether our results are driven by deals in which the acquirer purchasedtarget shares before launching the bid. We �nd that stock bidders with an initial toeholdin the company signal positive information about the target. This e¤ect is not present forcash bidders.17 The rationale for this is as follows: if the toehold variable captures the bidder�spositive valuation of the target irrespective of the medium of exchange (as also found by Betton,Eckbo, and Thorburn (2009) in bidding contests with no winner), then the positive e¤ect on thepost-withdrawal target CAR should only hold for stock deals because �unlike the informationrevealed by cash o¤ers �there is no target revaluation in stock o¤ers per se (as witnessed inFigure 1). Other than that, our results are neither driven by the reputation of the acquirer�sinvestment bank (as a proxy for the quality of due diligence) nor by within-industry (horizontal)mergers. Both variables could be reasonably related to a lower degree of information asymmetrybetween acquirer and target, potentially magnifying the scope of the information e¤ect in cashdeals. While the respective interaction e¤ects do have the expected positive sign, they fail tobe signi�cant.

All results are robust in the subsample of pure cash and stock deals, as can be seen inTable A.5. Our �ndings are consistent with a framework in which the method of exchangeis a su¢ cient statistic for private information. If the acquisition currency perfectly measuresinformation, any additional variable that is correlated with information should not matter initself or through its interaction with cash.

As the most important robustness check, we scrutinize whether post-withdrawal CARsare higher for cash targets merely because the latter may be more likely to become futuretargets. That is, we test whether the positive cash e¤ect in Table 4 was driven by (expected)announcement returns of future deals for cash targets. For this purpose, we consider the sampleof withdrawn deals, and run probit regressions of a dummy indicating another merger bid within

17 We cannot reject the hypothesis that the sum of the coe¢ cients for Toehold and Toehold � Cash is 0.

14

the next two years (but after a grace period of half a year to avoid capturing bidding wars).18

The results are summarized in Table 7. We �nd that cash targets are insigni�cantly less likelyto receive future merger bids. More than that, targets with low q ratios are signi�cantly morelikely to become targets in the future, which shows that we have in part controlled for futuremerger bids in Table 4 by including the target�s q. For similar reasons as in Table 3, we includethe target CAR (from announcement to withdrawal) to directly check whether the cumulativeabnormal returns predict future merger activity. We cannot �nd evidence for this. Once again,our results are also robust to considering the subset of pure cash and stock deals only (cf. TableA.6). Lastly, we replace the dependent variable by the actual takeover premium in future deals,and consider OLS regressions with the same set of control variables as before. The resultsin Table A.7, and in Table A.8 for the subsample of pure cash and stock deals, demonstratethat cash targets do not receive higher future takeover premia. We therefore conclude thatthe positive impact of cash o¤ers on post-withdrawal target CARs is not an artifact of futuremerger activity.

5 Conclusion

In this paper, we have presented robust evidence that the medium of exchange used by the bidderin merger transactions provides economically and statistically signi�cant information about thestand-alone value of the target: targets of cash o¤ers trade 15% above pre-announcement levelswhereas no signi�cant information is revealed about the targets of stock o¤ers. The previousliterature has primarily focused on the information content of the method of payment relatedto the bidder, and thus ignored this channel. Our analysis has important implications for thequantitative assessment of value creation in merger transactions.

Building on the �ndings of this paper, an important next step would be the generalization ofour results to completed deals. Exploiting exogenous variation in the probability of withdrawalin a subsample of our cases and other sources of identi�cation, it might be possible to build acredible structural model that allows to jointly estimate the endogenous withdrawal selection,information revelation, and value creation (or destruction) for cash and stock transactions.Such an analysis would deepen our understanding of these important company decisions, andwould help quantify the economic bene�ts of mergers.

18 Our results are robust to other speci�cations of the grace period (such as one month, six weeks, or threemonths).

15

References

Andrade, G., M. Mitchell, and E. Stafford (2001): �New Evidence and Perspectives onMergers,�Journal of Economic Perspectives, 15(2), 103�120.

Baker, M., J. C. Stein, and J. Wurgler (2003): �When Does The Market Matter? StockPrices And The Investment Of Equity-Dependent Firms,� The Quarterly Journal of Eco-nomics, 118(3), 969�1005.

Barber, B. M., and J. D. Lyon (1997): �Detecting long-run abnormal stock returns: Theempirical power and speci�cation of test statistics,�Journal of Financial Economics, 43(3),341�372.

Betton, S., B. E. Eckbo, and K. S. Thorburn (2008): �Corporate Takeovers,�in Hand-book of Corporate Finance: Empirical Corporate Finance, ed. by B. E. Eckbo, vol. 2, chap. 15,pp. 291�430. Elsevier/North-Holland.

(2009): �Merger Negotiations and the Toehold Puzzle,� Journal of Financial Eco-nomics, 91(2), 158�178.

Bhagat, S., M. Dong, D. Hirshleifer, and R. Noah (2005): �Do tender o¤ers createvalue? New methods and evidence,�Journal of Financial Economics, 76(1), 3 �60.

Bhagat, S., A. Shleifer, R. W. Vishny, G. Jarrel, and L. Summers (1990): �Hos-tile Takeovers in the 1980s: The Return to Corporate Specialization,�Brookings Papers onEconomic Activity. Microeconomics, 1990, 1�84.

Bradley, M., A. Desai, and E. H. Kim (1983): �The rationale behind inter�rm tendero¤ers: Information or Synergy?,�Journal of Financial Economics, 11(1-4), 183�206.

(1988): �Synergistic gains from corporate acquisitions and their division between thestockholders of target and acquiring �rms,�Journal of Financial Economics, 21(1), 3�40.

Brav, A. (2000): �Inference in Long-Horizon Event Studies: A Bayesian Approach with Ap-plication to Initial Public O¤erings,�Journal of Finance, 55(5), 1979�2016.

Davidson, Wallace N, I., D. Dutia, and L. Cheng (1989): �A Re-examination of theMarket Reaction to Failed Mergers,�Journal of Finance, 44(4), 1077�83.

Dodd, P. (1980): �Merger proposals, management discretion and stockholder wealth,�Journalof Financial Economics, 8(2), 105 �137.

Eckbo, B. E., R. M. Giammarino, and R. L. Heinkel (1990): �Asymmetric Informationand the Medium of Exchange in Takeovers: Theory and Tests,�Review of Financial Studies,3(4), 651�75.

Fama, E. F. (1998): �Market e¢ ciency, long-term returns, and behavioral �nance,� Journalof Financial Economics, 49(3), 283�306.

Fishman, M. J. (1989): �Preemptive Bidding and the Role of the Medium of Exchange inAcquisitions,�Journal of Finance, 44(1), 41�57.

16

Franks, J., and C. Mayer (1996): �Hostile takeovers and the correction of managerialfailure,�Journal of Financial Economics, 40(1), 163�181.

Friedman, J. (2006): �Stock Market Driven Acquisitions: Theory and Evidence,� Mimeo,Harvard University.

Grossman, S. J., and O. D. Hart (1980): �Takeover Bids, The Free-Rider Problem, andthe Theory of the Corporation,�The Bell Journal of Economics, 11(1), 42�64.

Hansen, R. G. (1987): �A Theory for the Choice of Exchange Medium in Mergers and Ac-quisitions,�Journal of Business, 60(1), 75�95.

Hirshleifer, D., and S. Titman (1990): �Share Tendering Strategies and the Success ofHostile Takeover Bids,�Journal of Political Economy, 98(2), 295�324.

Jarrell, G. A., J. A. Brickley, and J. M. Netter (1988): �The Market for CorporateControl: The Empirical Evidence Since 1980,�Journal of Economic Perspectives, 2(1), 49�68.

Jensen, M. C., and R. S. Ruback (1983): �The Market for Corporate Control : The Scienti�cEvidence,�Journal of Financial Economics, 11(1-4), 5�50.

Jovanovic, B., and S. Braguinsky (2004): �Bidder Discounts and Target Premia inTakeovers,�American Economic Review, 94(1), 46�56.

Jovanovic, B., and P. L. Rousseau (2002): �The Q-Theory of Mergers,�American Eco-nomic Review, 92(2), 198�204.

Kaplan, S. N., and L. Zingales (1997): �Do Investment-Cash Flow Sensitivities ProvideUseful Measures of Financing Constraints?,�The Quarterly Journal of Economics, 112(1),169�215.

Lamont, O., C. Polk, and J. Saa-Requejo (2001): �Financial Constraints and StockReturns,�Review of Financial Studies, 14(2), 529�54.

Loughran, T., and A. M. Vijh (1997): �Do Long-Term Shareholders Bene�t from CorporateAcquisitions?,�Journal of Finance, 52(5), 1765�90.

Malmendier, U., E. Moretti, and F. Peters (2010): �Winning by Losing: Evidence onOverbidding in Mergers,�Mimeo, University of California, Berkeley.

Moeller, S. B., F. P. Schlingemann, and R. M. Stulz (2005): �Wealth Destruction ona Massive Scale? A Study of Acquiring-Firm Returns in the Recent Merger Wave,�Journalof Finance, 60(2), 757�782.

Myers, S. C., and N. S. Majluf (1984): �Corporate �nancing and investment decisionswhen �rms have information that investors do not have,� Journal of Financial Economics,13(2), 187�221.

Rau, P. R., and T. Vermaelen (1998): �Glamour, Value and the Post-Acquisition Perfor-mance of Acquiring Firms,�Journal of Financial Economics, 49(2), 223�253.

Savor, P. G., and Q. Lu (2009): �Do Stock Mergers Create Value for Acquirers?,�Journalof Finance, 64(3), 1061�1097.

17

Shleifer, A., and R. W. Vishny (1986): �Large Shareholders and Corporate Control,�Journal of Political Economy, 94(3), 461�88.

Shleifer, A., and R. W. Vishny (2003): �Stock market driven acquisitions,� Journal ofFinancial Economics, 70(3), 295�311.

Travlos, N. G. (1987): �Corporate Takeover Bids, Methods of Payment, and Bidding Firms�Stock Returns,�Journal of Finance, 42(4), 943�963.

18

6 Figures

Figure 3: Announcement and Withdrawal E¤ects

Figure 1: CARs from -25 (before announcement) to +25 (after withdrawal), normalized

Note (Figures 1-3): 95 cash and 131 stock deals.

Figure 2: BHARs from -25 (before announcement) to +25 (after withdrawal), normalized

‐0.15

‐0.1

‐0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

‐30 ‐20 ‐10 0 10 20 30 40 50 60 70 80

Cash target CAR

Stock target CAR

Cash acquirer CAR

Stock acquirer CAR

Announcement

Withdrawal

‐0.15

‐0.1

‐0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

‐30 ‐20 ‐10 0 10 20 30 40 50 60 70 80

Cash target BHAR

Stock target BHAR

Cash acquirer BHAR

Stock acquirer BHAR

Announcement

Withdrawal

Notes: Buy-and-Hold Abnormal Returns (BHARs) from 25 trading days pre-announcement to25 trading days post-withdrawal. 95 cash and 131 stock deals. The market value of equityof the target relative to the acquirer is strictly greater than 5%. See Table 1 for the exactconstruction of the sample.

19

Figure 4: Announcement and Withdrawal E¤ectsFigure 3: BHRs from -25 (before announcement) to +25 (after withdrawal), normalized

‐0.15

‐0.1

‐0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

‐30 ‐20 ‐10 0 10 20 30 40 50 60 70 80

Cash target BHR

Stock target BHR

Cash acquirer BHR

Stock acquirer BHR

Announcement

Withdrawal

Notes: Buy-and-Hold Returns (BHRs) from 25 trading days pre-announcement to 25 tradingdays post-withdrawal. 95 cash and 131 stock deals. The market value of equity of the targetrelative to the acquirer is strictly greater than 5%. See Table 1 for the exact construction ofthe sample.

Figure 5: Announcement and Withdrawal E¤ectsFigure 4: CARs from -100 (before announcement) to +100 (after withdrawal), normalized

Note: 76 cash and 114 stock deals.

‐0.15

‐0.1

‐0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

‐110 ‐90 ‐70 ‐50 ‐30 ‐10 10 30 50 70 90 110 130 150

Cash target CAR

Stock target CAR

Cash acquirer CAR

Stock acquirer CAR

Announcement

Withdrawal

Notes: Cumulative Abnormal Returns (CARs) from 100 trading days pre-announcement to 100trading days post-withdrawal. 76 cash and 114 stock deals. The market value of equity of thetarget relative to the acquirer is strictly greater than 5%. See Table 1 for the exact constructionof the sample.

20

7 Tables

Table 1: Sample Construction

Criterion # DealsAvailability of CRSP data for acquirer and/or target 15,502

Availability of valid deal dates 13,490

Availability of stock data 12,025(at least 25 trading days prior to bid announcement)

Deal window of 5 to 250 trading days 11,368(between announcement and completion/withdrawal)

No competing o¤ers 10,285

Subsamplesof which

successful withdrawn7,071 3,214

Availability of all variables used in any regression 1,936of which

successful withdrawn1,716 220

Graph sample of withdrawn pure cash and stock deals 236of which

cash stockAcquirer and target 95 131

Acquirer only 4 8Target only 3 0

Notes: The availability of valid deal dates requires that announcement and withdrawal orcompletion dates are not missing and consistent, e.g., completion not prior to announcement oforiginal bid. The graph sample contains all deals used in Figure 1, i.e., deals with the additionalrequirement of being a pure cash or a pure stock deal, without competing o¤ers outstanding,and where the target�s market value of equity is at least equal to 5% of the acquirer�s marketvalue of equity. Furthermore, we require CRSP data availability for at least 25 days beforeannouncement and at least 25 days after withdrawal.

21

Table2a:SummaryStatistics�AllDeals

Successfuldeals

Withdrawndeals

Variable

Mean

Std.dev.

Min

Max

Mean

Std.dev.

Min

Max

p-value

Cashin%

48.76

45.89

0100

40.23

45.96

0100

0.011

Stockin%

44.47

45.65

0100

51.75

46.28

0100

0.024

Otherpaymentin%

6.77

17.23

0100

8.01

22.97

0100

0.470

Timetocompletion/withdrawal

75.85

43.82

5245

65.67

51.49

5238

0.002

LBOin%

0.12

3.41

0100

0.91

9.51

0100

0.014

Premium(to1monthprior)in%

45.57

40.24

-48.30

202.10

42.23

45.16

-48.30

202.10

0.414

Hostiledealin%

1.86

13.53

0100

15.45

36.22

0100

0.000

Transactionvaluein2010$bn

1.39

4.13

0.00

70.51

1.73

7.27

0.00

77.04

0.257

Relativedealsize

1.11

17.04

0.00

682.90

1.10

2.21

0.00

21.86

0.996

%oftargetsought

95.27

16.98

1.80

100

93.67

16.68

15.80

100

0.217

Newdealannouncedwithin2yearsin%

n/a

n/a

n/a

n/a

20.00

40.09

0100

n/a

Experiencedacquirerin%

82.05

38.39

0100

71.82

45.09

0100

0.000

Equityratio

0.35

0.87

0.00

8.93

0.79

1.32

0.00

8.93

0.000

qofacquirer

2.50

2.39

0.51

15.20

2.36

2.34

0.51

15.20

0.434

qoftarget

2.06

1.71

0.50

9.91

1.94

1.72

0.50

9.91

0.395

qofacquirer>qoftargetin%

62.30

48.48

0100

58.64

49.36

0100

0.259

KZindexofacquirer

-0.09

1.47

-10.46

3.73

0.06

1.49

-5.94

3.73

0.114

KZindexoftarget

0.10

1.68

-10.05

5.22

0.19

1.58

-10.73

4.28

0.560

Sameindustry(1digitSIC)in%

74.30

43.71

0100

69.55

46.13

0100

0.126

Sameindustry(2digitsSIC)in%

60.26

48.95

0100

53.18

50.01

0100

0.035

N1,716

220

Notes(Tables2a

and2b):

Timetocompletion/withdrawalisintradingdays.The

historicaltransactionvaluewasconverted

usingConsumerPriceIndex(CPI)ConversionFactors.Relativedealsizeisthetransactionvalueovermarketvalueofequityofthe

acquirer.New

dealannouncedwithin2yearsisconditionalonthedealbeingannouncedatleasthalfayearafterthepreviousone.

Experiencedacquirersappearatleast�vetimesinthedataset.Equityratioisde�nedastheratiooftarget�stoacquirer�smarketvalue

ofequity.KZindexisthefour-variableversioninLamont,Polk,andSaa-Requejo(2001).Allnon-deal-relatedvariables(e.g.,qratios)

areyear-endmeasuresoneyearpriortothedealannouncement.Premium,Equityratio,allqandKZvariablesarewinsorizedatthe

1stand99thpercentiles.Thesampleisrestrictedtodealsthatincludeallvariablesusedinanyofthespeci�cationsinTables3-7.

22

Table2b:SummaryStatistics�WithdrawnDeals

Cashdeals

Stockdeals

Variable

Mean

Std.dev.

Min

Max

Mean

Std.dev.

Min

Max

p-value

Timetowithdrawal

56.27

44.97

5178

62.81

50.19

5232

0.394

LBOin%

2.82

16.66

0100

00

00

0.115


52.97

43.78

-48.30

202.10

37.57

44.03

-44.20

199.90

0.029

Hostiledealin%

23.94

42.98

0100

5.68

23.28

0100

0.001


0.66

1.47

0.00

8.93

1.56

6.20

0.01

55.64

0.239

Relativedealsize

1.09

2.35

0.00

11.77

0.84

1.18

0.01

8.49

0.399

%oftargetsought

92.67

15.20

20.00

100

94.44

17.77

19.10

100

0.507


16.90

37.74

0100

21.59

41.38

0100

0.461


69.01

46.57

0100

72.73

44.79

0100

0.610

Equityratio

0.77

1.70

0.00

8.93

0.78

1.12

0.01

8.58

0.955

qofacquirer

1.84

1.69

0.51

13.72

2.89

2.75

0.58

15.20

0.005

qoftarget

1.51

1.31

0.50

9.91

2.51

2.07

0.50

9.91

0.001


64.79

48.10

0100

54.55

50.08

0100

0.194

KZindexofacquirer

0.12

1.62

-5.94

3.41

0.00

1.43

-4.30

3.73

0.618

KZindexoftarget

0.21

1.06

-2.29

2.90

-0.08

1.94

-10.73

4.28

0.250


73.24

44.59

0100

72.73

44.79

0100

0.943


47.89

50.31

0100

56.82

49.82

0100

0.265

N71

88

23

Table 3: Determinants of Withdrawal (All Deals)

Deal withdrawnCash 2 [0; 1] -0.041** -0.017 -0.005 -0.018 -0.006

(0.02) (0.02) (0.02) (0.02) (0.02)Log(relative deal size) 0.027*** 0.027*** 0.023*** 0.020***

(0.00) (0.00) (0.00) (0.00)LBO 0.318 0.322

(0.31) (0.31)Premium to 1 month prior -0.020 0.029

(0.02) (0.02)Hostile 0.333*** 0.320***

(0.07) (0.07)% of target sought -0.001** -0.001*

(0.00) (0.00)q of acquirer 0.006* 0.005

(0.00) (0.00)q of target -0.001 -0.001

(0.00) (0.00)KZ index of acquirer 0.004 0.004

(0.00) (0.00)KZ index of target -0.000 0.000

(0.00) (0.00)Experienced acquirer -0.029* -0.031*

(0.02) (0.02)Target CAR (A-25, A+1) -0.109***

(0.04)Industry & year FE N N Y Y YN 1,936 1,936 1,924 1,924 1,924

Notes: The table reports marginal e¤ects of probit regressions. All non-deal-related independentvariables are year-end measures one year prior to deal announcement. Relative deal size is thetransaction value over the market value of equity of acquirer. An acquirer is experienced if the�rm appears at least �ve times in the data set. Target CAR is measured on the [-25, 1] windowaround deal announcement. Robust standard errors are in parentheses.

24

Table 4: Target Returns (Withdrawn Deals)

Target CAR (A-25, W+25)Cash 0.336*** 0.207*** 0.460*** 0.357** 0.595*** 0.457**

(0.08) (0.08) (0.16) (0.16) (0.20) (0.19)LBO 0.394 0.363 0.312 0.619 0.495

(0.31) (0.30) (0.42) (0.43) (0.49)Premium to 1 month prior 0.363*** 0.493*** 0.463*** 0.418*** 0.321***

(0.11) (0.09) (0.10) (0.10) (0.10)Premium � Cash -0.371 -0.319 -0.302 -0.185

(0.25) (0.24) (0.26) (0.24)Hostile 0.154** 0.141** 0.120 0.161 0.131

(0.07) (0.06) (0.08) (0.10) (0.12)Equity ratio -0.030 -0.035 -0.044 -0.033 -0.038

(0.03) (0.03) (0.03) (0.04) (0.04)Log(relative deal size) 0.030 0.029 0.034 0.015 0.016

(0.03) (0.03) (0.03) (0.03) (0.04)q of acquirer 0.013 0.012 -0.007 0.004 -0.011

(0.02) (0.02) (0.03) (0.02) (0.03)q of target -0.040 -0.030 -0.025 0.011 0.033

(0.02) (0.03) (0.04) (0.03) (0.03)q of target � Cash -0.041 -0.017 -0.101* -0.082

(0.04) (0.04) (0.05) (0.05)KZ index of acquirer -0.059 -0.031

(0.04) (0.04)KZ index of acquirer � Cash 0.020 -0.012

(0.05) (0.06)KZ index of target 0.030 0.072**

(0.03) (0.03)Industry & year FE Y Y Y Y Y YDeal sample All All All All Pure C/S Pure C/SN 242 242 242 213 176 156

Notes: OLS regressions with Target CAR from 25 days before announcement to 25 days afterwithdrawal as the dependent variable. Cash is expressed as a fraction of the total payment (andhence equal to a dummy for cash in the sample of pure cash and stock deals in the last twocolumns). Equity ratio is de�ned as the ratio of target�s to acquirer�s market value of equity,Relative deal size is equal to transaction value over market value of equity of acquirer. Allnon-deal-related independent variables are year-end measures one year prior to the withdrawndeal�s announcement. Robust standard errors are in parentheses.

25

Table 5: Synopses of Withdrawn Deals

Withdrawal reason Cash deals Stock deals p-valueAcquirer withdrawal 48.19% 58.76% 0.158Target rejection 19.28% 8.25% 0.030Mutual consent 3.61% 8.25% 0.198P

71.08% 75.26% 0.531Target defense 9.64% 2.06% 0.027Financing & other closing problems 2.41% 2.06% 0.876Regulatory & other exogenous reasons 15.66% 17.53% 0.740(Stock) market conditions 1.20% 3.09% 0.395N 83 97

Notes: The sample is restricted to pure cash and stock deals as in the �fth column of Ta-ble 4. Acquirer withdrawals comprise the subset of deals that are withdrawn by the acquirerwithout any mention of the target in the deal synopsis. Similarly, target rejections comprisethe subset of bids that are rejected by the target�s board and/or shareholders (this includescases where insu¢ cient shares are tendered). We classify deals as withdrawn by mutual con-sent if the deal synopsis explicitly states this (as opposed to deals withdrawn primarily dueto one party�s concerns). Deals withdrawn due to target defense include, most notably, theadoption of a shareholder rights plan and other active defense mechanisms. By Financing &other closing problems we denote practical reasons for deal withdrawal, e.g., shortcomings in�nancing commitments, lacking due diligence and customary closing conditions, etc. Regu-latory & other exogenous reasons include deals that were subject to regulatory approval andsudden withdrawals triggered by events such as acquisition of the bidder (which naturally leadsto a withdrawal of the bidder�s o¤er). The last category comprises the subset of deals that arewithdrawn due to shifting market conditions (typically stock market plunges).

26

Table 6: Target Returns (Withdrawn Deals) �Robustness Checks

Target CAR (A-25, W+25)Cash 2 [0; 1] 0.416** 0.359** 0.304* 0.378**

(0.17) (0.16) (0.16) (0.18)LBO 0.364 0.323 0.315 0.387

(0.43) (0.41) (0.44) (0.44)Premium to 1 month prior 0.455*** 0.458*** 0.457*** 0.450***

(0.10) (0.10) (0.10) (0.10)Premium � Cash -0.324 -0.320 -0.319 -0.328

(0.24) (0.24) (0.24) (0.25)Hostile 0.072 0.102 0.109 0.039

(0.09) (0.08) (0.08) (0.10)Equity ratio -0.055 -0.046 -0.044 -0.058*

(0.03) (0.03) (0.03) (0.03)Log(relative deal size) 0.046 0.036 0.035 0.051*

(0.03) (0.03) (0.03) (0.03)q of acquirer -0.002 -0.005 -0.005 0.001

(0.03) (0.03) (0.03) (0.03)q of target -0.023 -0.024 -0.026 -0.025

(0.03) (0.04) (0.04) (0.04)q of target � Cash -0.021 -0.025 -0.019 -0.032

(0.05) (0.04) (0.05) (0.05)KZ index of acquirer -0.063 -0.059 -0.061 -0.065

(0.04) (0.04) (0.04) (0.04)KZ index of acquirer � Cash 0.026 0.020 0.022 0.030

(0.05) (0.05) (0.05) (0.05)KZ index of target 0.034 0.030 0.030 0.033

(0.03) (0.03) (0.03) (0.03)Toehold 0.215** 0.224**

(0.11) (0.11)Toehold � Cash -0.185 -0.186

(0.14) (0.15)Tier-one IB -0.046 -0.007

(0.16) (0.16)Tier-one IB � Cash 0.195 0.182

(0.25) (0.26)Same industry (2 digits SIC) -0.067 -0.057

(0.09) (0.08)Same industry � Cash 0.113 0.094

(0.15) (0.15)Industry & year FE Y Y Y YN 213 213 213 213

Notes: OLS regressions with Target CAR from 25 days before announcement to 25 days afterwithdrawal as the dependent variable. Equity ratio is the ratio of target�s to acquirer�s marketvalue of equity, Relative deal size is the transaction value over the market value of equity of theacquirer. Toehold is an indicator equal to one if the acquirer owned a share of the target beforeannouncement, and Tier-one IB indicates whether the acquirer was advised by either GoldmanSachs, Morgan Stanley, Merrill Lynch, or JPMorgan. All non-deal-related independent variablesare year-end measures one year prior to the withdrawn deal�s announcement. Robust standarderrors are in parentheses.

27

Table 7: Future Takeover Attempts (Withdrawn Deals)

New deal announced 2 years after withdrawalCash 2 [0; 1] -0.665*** -0.055 -0.128 -0.125 -0.127

(0.07) (0.14) (0.16) (0.16) (0.16)Premium to 1 month prior -0.221 -0.217 -0.237

(0.14) (0.14) (0.15)Hostile -0.397* -0.397* -0.404**

(0.20) (0.20) (0.20)Log(transaction value) -0.011 -0.010 -0.008

(0.04) (0.04) (0.04)q of target -0.154** -0.137** -0.134**

(0.06) (0.06) (0.06)KZ index of target 0.059 0.059

(0.05) (0.05)Target CAR (A-25, W+25) 0.056

(0.15)Industry & year FE N Y Y Y YN 562 531 531 531 531

Notes: Probit regressions of a dummy indicating another merger bid within the next two years.We exclude observations with merger bids within half a year after withdrawal since their clas-si�cation as competing bid (in the previous takeover attempt) versus new bid is ambiguous.Transaction value is in 2010 $bn. Target CAR (A-25, W+25) are the cumulative abnormalreturns from 25 days before announcement until 25 days after withdrawal. All non-deal-relatedindependent variables are measured at the end of the year prior to the withdrawn deal�s an-nouncement. Robust standard errors are in parentheses.

28

AAppendixTables

TableA.1:SummaryStatistics(SuccessfulDeals)

Cashdeals

Stockdeals

Variable

Mean

Std.dev.

Min

Max

Mean

Std.dev.

Min

Max

p-value

Timetocompletion

58.34

37.52

5230

83.32

38.64

16239

0.000

LBOin%

0.31

5.55

0100

00

00

0.188


46.13

36.27

-48.30

202.10

46.80

45.89

-48.30

202.10

0.774

Hostiledealin%

2.77

16.43

0100

0.89

9.39

0100

0.016


0.70

1.51

0.00

17.40

1.47

5.37

0.00

70.51

0.001

Relativedealsize

0.70

6.06

0.00

137.03

0.58

3.39

0.00

79.38

0.662

%oftargetsought

91.96

22.29

1.80

100

96.38

14.70

7.70

100

0.000


n/a

n/a

n/a

n/a

n/a

n/a

n/a

n/a

n/a


83.05

37.55

0100

78.86

40.86

0100

0.063

Equityratio

0.29

0.96

0.00

8.93

0.34

0.63

0.00

8.93

0.258

qofacquirer

2.09

1.69

0.51

15.20

3.41

3.29

0.51

15.20

0.000

qoftarget

1.85

1.38

0.50

9.91

2.62

2.27

0.50

9.91

0.000


60.09

49.01

0100

64.48

47.90

0100

0.117

KZindexofacquirer

-0.14

1.29

-6.03

3.73

-0.41

1.69

-10.46

3.01

0.002

KZindexoftarget

-0.04

1.57

-10.05

5.22

-0.28

1.89

-9.57

4.70

0.016


69.65

46.01

0100

75.13

43.26

0100

0.034


53.31

49.93

0100

63.41

48.21

0100

0.000

N649

563

Notes:Timeto

completionisintradingdays.

The

historicaltransactionvaluewas

convertedusingConsumerPrice

Index

(CPI)ConversionFactors.Relativedealsizeisthetransactionvalueovermarketvalueofequityoftheacquirer.New

dealannounced

within2yearsisconditionalonthedealbeingannouncedatleasthalfayearafterthepreviousone.Experiencedacquirersappearat

least�vetimesinthedataset.Equityratioisde�nedastheratiooftarget�stoacquirer�smarketvalueofequity.KZindexisthe

four-variableversioninLamont,Polk,andSaa-Requejo(2001).Allnon-deal-relatedvariables(e.g.,qratios)areyear-endmeasuresone

yearpriortothedealannouncement.Premium,Equityratio,allqandKZvariablesarewinsorizedatthe1stand99thpercentiles.The

sampleisrestrictedtodealsthatincludeallvariablesusedinanyofthespeci�cationsinTables3-7.

29

Table A.2: Determinants of Cash O¤ers (All Deals)

Cash 2 [0; 1] Cash 2 f0; 1gLog(relative deal size) -0.054*** -0.060*** -0.062*** -0.209*** -0.246*** -0.258***

(0.01) (0.01) (0.01) (0.02) (0.03) (0.03)% of target sought -0.002*** -0.002** -0.002*** -0.007*** -0.004 -0.004*

(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)Experienced acquirer -0.018 -0.020 -0.019 -0.090 -0.065 -0.087

(0.03) (0.03) (0.03) (0.10) (0.10) (0.10)q of acquirer -0.028*** -0.031*** -0.120*** -0.126***

(0.00) (0.00) (0.03) (0.03)q of target -0.029*** -0.031*** -0.144*** -0.144***

(0.01) (0.01) (0.03) (0.03)LBO 0.365***

(0.09)Premium to 1 month prior 0.065*** 0.154

(0.02) (0.10)Hostile 0.211*** 0.933***

(0.05) (0.26)KZ index of acquirer -0.001 0.007

(0.01) (0.03)KZ index of target -0.017*** -0.041

(0.01) (0.03)Industry & year FE Y Y Y Y Y YDeal sample All All All Pure C/S Pure C/S Pure C/SN 1,932 1,932 1,932 1,400 1,400 1,396

Notes: OLS regressions in the �rst three columns and probit regressions in the last threecolumns include acquirer and target industry controls. Cash is expressed as a fraction of thetotal payment (and hence equal to a dummy for cash in the sample of pure cash and stock dealsin the last three columns). Relative deal size is equal to transaction value over market value ofthe acquirer�s equity. Experienced acquirers appear at least �ve times in the data set. Robuststandard errors are in parentheses.

30

Table A.3: Determinants of Withdrawal (All Pure Cash & Stock Deals)

Deal withdrawnCash 2 f0; 1g -0.033* -0.005 0.007 -0.010 -0.003

(0.02) (0.02) (0.02) (0.02) (0.02)Log(relative deal size) 0.032*** 0.030*** 0.025*** 0.024***

(0.00) (0.00) (0.00) (0.00)LBO 0.360 0.357

(0.31) (0.31)Premium to 1 month prior 0.001 0.038

(0.02) (0.03)Hostile 0.311*** 0.309***

(0.08) (0.08)% of target sought -0.000 -0.000

(0.00) (0.00)q of acquirer 0.002 0.002

(0.00) (0.00)q of target 0.002 0.002

(0.01) (0.00)KZ index of acquirer 0.007 0.006

(0.01) (0.01)KZ index of target 0.002 0.002

(0.00) (0.00)Experienced acquirer -0.016 -0.018

(0.02) (0.02)Target CAR (A-25, A+1) -0.075*

(0.04)Industry & year FE N N Y Y YN 1,403 1,403 1,396 1,396 1,396

Notes: The table reports marginal e¤ects of probit regressions. All non-deal-related independentvariables are year-end measures one year prior to deal announcement. Relative deal size is thetransaction value over the market value of equity of acquirer. An acquirer is Experienced if the�rm appears at least �ve times in the data set. Target CAR is measured on the [-25, 1] windowaround deal announcement. Robust standard errors are in parentheses.

31

Table A.4: Target Returns (Withdrawn Pure Cash & Stock Deals) �No TargetDefense and/or Rejection

Target CAR (A-25, W+25)Cash 2 f0; 1g 0.794*** 0.683*** 0.871*** 0.680*** 0.550*** 0.759***

(0.21) (0.20) (0.22) (0.21) (0.19) (0.22)LBO 0.503 1.419*** 1.616*** 0.557 1.396*** 1.454***

(0.53) (0.40) (0.45) (0.49) (0.42) (0.46)Premium to 1 month prior 0.448*** 0.422*** 0.427*** 0.352*** 0.356*** 0.363***

(0.11) (0.12) (0.12) (0.12) (0.12) (0.13)Premium � Cash -0.398 -0.360 -0.410 -0.275 -0.258 -0.319

(0.26) (0.26) (0.26) (0.24) (0.24) (0.25)Hostile 0.208* 0.217* 0.260* 0.162 0.238* 0.280*

(0.12) (0.12) (0.15) (0.15) (0.13) (0.15)Equity ratio -0.070* -0.015 -0.059 -0.080* -0.006 -0.053

(0.04) (0.04) (0.04) (0.05) (0.04) (0.04)Log(relative deal size) 0.033 0.017 0.042 0.041 0.008 0.035

(0.04) (0.03) (0.04) (0.04) (0.04) (0.04)q of acquirer 0.013 -0.001 0.009 -0.007 -0.019 -0.012

(0.02) (0.02) (0.02) (0.03) (0.03) (0.04)q of target -0.003 0.020 0.009 0.026 0.039 0.033

(0.03) (0.03) (0.03) (0.04) (0.04) (0.04)q of target � Cash -0.174*** -0.126** -0.203*** -0.175*** -0.115** -0.197***

(0.06) (0.05) (0.07) (0.06) (0.05) (0.06)KZ index of acquirer -0.049 -0.032 -0.041

(0.04) (0.04) (0.04)KZ index of acquirer � Cash 0.008 -0.053 -0.057

(0.06) (0.06) (0.06)KZ index of target 0.082** 0.062* 0.071*

(0.04) (0.03) (0.04)Industry & year FE Y Y Y Y Y YDeal sample (I) (II) (III) (I) (II) (III)N 152 164 142 135 147 128

Notes: OLS regressions with Target CAR from 25 days before announcement to 25 days afterwithdrawal as the dependent variable. Deal samples are as follows: (I) no target rejection;(II) no target defense; (III) no target rejection or defense. Equity ratio is de�ned as the ratioof target�s to acquirer�s market value of equity, Relative deal size is equal to transaction valueover market value of equity of acquirer. All non-deal-related independent variables are year-endmeasures one year prior to the withdrawn deal�s announcement. Robust standard errors are inparentheses.

32

Table A.5: Target Returns (Withdrawn Pure Cash & Stock Deals) �RobustnessChecks

Target CAR (A-25, W+25)Cash 2 f0; 1g 0.510** 0.464** 0.406** 0.482**

(0.21) (0.19) (0.19) (0.22)LBO 0.578 0.503 0.489 0.586

(0.49) (0.47) (0.50) (0.47)Premium to 1 month prior 0.321*** 0.316*** 0.306*** 0.297***

(0.10) (0.10) (0.10) (0.11)Premium � Cash -0.206 -0.183 -0.171 -0.193

(0.25) (0.25) (0.24) (0.27)Hostile 0.112 0.084 0.141 0.075

(0.13) (0.13) (0.12) (0.13)Equity ratio -0.041 -0.040 -0.044 -0.053

(0.04) (0.04) (0.04) (0.04)Log(relative deal size) 0.019 0.018 0.021 0.030

(0.04) (0.04) (0.04) (0.04)q of acquirer -0.007 -0.010 -0.003 0.004

(0.03) (0.03) (0.03) (0.03)q of target 0.034 0.038 0.031 0.035

(0.03) (0.04) (0.03) (0.04)q of target � Cash -0.081 -0.098* -0.073 -0.088*

(0.05) (0.05) (0.05) (0.05)KZ index of acquirer -0.033 -0.032 -0.030 -0.032

(0.04) (0.04) (0.04) (0.04)KZ index of acquirer � Cash -0.007 -0.013 -0.012 -0.005

(0.06) (0.06) (0.06) (0.06)KZ index of target 0.074** 0.074** 0.073** 0.077**

(0.03) (0.03) (0.03) (0.03)Toehold 0.217* 0.222*

(0.12) (0.12)Toehold � Cash -0.240 -0.248

(0.18) (0.18)Tier-one IB -0.127 -0.104

(0.23) (0.23)Tier-one IB � Cash 0.340 0.372

(0.35) (0.36)Same industry (2 digits SIC) -0.096 -0.106

(0.10) (0.11)Same industry � Cash 0.054 0.029

(0.18) (0.19)Industry & year FE Y Y Y YN 156 156 156 156

Notes: OLS regressions with Target CAR from 25 days before announcement to 25 days afterwithdrawal as the dependent variable. Equity ratio is the ratio of target�s to acquirer�s marketvalue of equity, Relative deal size is the transaction value over the market value of equity of theacquirer. Toehold is an indicator equal to one if the acquirer owned a share of the target beforeannouncement. Tier-one IB indicates whether the acquirer was advised by Goldman Sachs,Morgan Stanley, Merrill Lynch, or JPMorgan. All non-deal-related independent variables aremeasured at the end of the year prior to the withdrawn deal�s announcement. Robust standarderrors are in parentheses.

33

Table A.6: Future Takeover Attempts (Withdrawn Pure Cash & Stock Deals)

New deal announced 2 years after withdrawalCash 2 f0; 1g -0.664*** -0.040 -0.114 -0.115 -0.118


(0.17) (0.17) (0.17)Hostile -0.336 -0.341 -0.365

(0.24) (0.24) (0.24)Log(transaction value) -0.002 0.001 0.008

(0.04) (0.04) (0.04)q of target -0.131* -0.120* -0.116*


(0.05) (0.05)Target CAR (A-25, W+25) 0.161


Notes: Probit regressions of a dummy indicating another merger bid within the next two years.We exclude observations with merger bids within half a year after withdrawal since their clas-si�cation as competing bid (in the previous takeover attempt) versus new bid is ambiguous.Transaction value is in 2010 $bn. Target CAR (A-25, W+25) are the cumulative abnormalreturns from 25 days before announcement until 25 days after withdrawal. All non-deal-relatedindependent variables are measured at the end of the year prior to the withdrawn deal�s an-nouncement. Robust standard errors are in parentheses.

34

Table A.7: Future Takeover Premia (Withdrawn Deals)

New o¤er premium 2 years after withdrawalCash 2 [0; 1] 0.086*** -0.000 -0.015 -0.014 -0.015


(0.04) (0.04) (0.04)Hostile -0.044 -0.043 -0.046


(0.01) (0.01) (0.01)q of target -0.015** -0.013** -0.013*


(0.01) (0.01)Target CAR (A-25, W+25) 0.020


Notes: OLS regressions with future o¤er premium (to 1 month prior) in case of another mergerbid within the next two years (and zero otherwise) as the dependent variable. We excludeobservations with merger bids within half a year after withdrawal since their classi�cation ascompeting bid (in the previous takeover attempt) versus new bid is ambiguous. Transactionvalue is in 2010 $bn. Target CAR (A-25, W+25) are the cumulative abnormal returns from25 days before announcement until 25 days after withdrawal. All non-deal-related independentvariables are measured at the end of the year prior to the withdrawn deal�s announcement.Robust standard errors are in parentheses.

35

Table A.8: Future Takeover Premia (Withdrawn Pure Cash & Stock Deals)

New o¤er premium 2 years after withdrawalCash 2 f0; 1g 0.070*** -0.012 -0.032 -0.033 -0.034

(0.01) (0.03) (0.04) (0.04) (0.04)Premium to 1 month prior 0.019 0.021 0.013

(0.04) (0.04) (0.04)Hostile -0.033 -0.034 -0.039


(0.01) (0.01) (0.01)q of target -0.016* -0.013 -0.013


(0.01) (0.01)Target CAR (A-25, W+25) 0.031


Notes: OLS regressions with future o¤er premium (to 1 month prior) in case of another mergerbid within the next two years (and zero otherwise) as the dependent variable. We excludeobservations with merger bids within half a year after withdrawal since their classi�cation ascompeting bid (in the previous takeover attempt) versus new bid is ambiguous. Transactionvalue is in 2010 $bn. Target CAR (A-25, W+25) are the cumulative abnormal returns from25 days before announcement until 25 days after withdrawal. All non-deal-related independentvariables are measured at the end of the year prior to the withdrawn deal�s announcement.Robust standard errors are in parentheses.

36

Documents

Winning by Losing: Evidence on Overbidding in Mergers