Rumors of Mergers and Acquisitions: Market Efficiency and

Markup Pricing

Hsin-I Chou,a Gloria Y. Tian,b Xiangkang Yin,a∗

a La Trobe University, b

November 2010

University of New South Wales

Abstract: Rumors can be classified into two types, according to whether they can

credibly predict impending events or not. The analysis of takeover rumors of publicly

traded US companies from 1990 to 2008 shows that these two types of rumors can be

statistically distinguished by returns of rumored takeover targets before rumor

publication. However, market responses to the rumors on the rumor day and the day after

are statistically indifferent. Trading on such rumors can be profitable. Moreover,

takeover premiums of sampled targets cannot be explained by markup pricing hypothesis

although the hypothesis is supported by the view of efficient markets.

Keywords: Financial rumor, merger and acquisition, market efficiency, runup, markup,

takeover premium

JEL Classification: G14, G34

∗ Corresponding author, School of Economics and Finance, La Trobe University, Bundoora, Victoria 3086,

Australia. Tel: 61-3-9479 2312, Fax: 61-3-9479 1654, Email: x.yin@latrobe.edu.au. We are grateful for

the comments of Buly Cadak, Stephen Easton, David Feldman, John Handley, Mark Humphery, Russell

Jame, Ron Masulis, Lily Nguyen, Jerry Parwada, David Prentice, Abul Shamsuddin, Jianfeng Shen, Neal

Stoughton, Peter Swan, and seminar participants at La Trobe University, the University of Newcastle and

the University of New South Wales. We also thank Reza Makouei for his excellent research assistance.

Rumors of Mergers and Acquisitions: Market Efficiency and

Markup Pricing

November 2010

Abstract: Rumors can be classified into two types, according to whether they can

credibly predict impending events or not. The analysis of takeover rumors of publicly

traded US companies from 1990 to 2008 shows that these two types of rumors can be

statistically distinguished by returns of rumored takeover targets before rumor publication.

However, market responses to the rumors on the rumor day and the day after are

statistically indifferent. Trading on such rumors can be profitable. Moreover, takeover

premiums of sampled targets cannot be explained by markup pricing hypothesis although

the hypothesis is supported by the view of efficient markets.

Keywords: Financial rumor, merger and acquisition, market efficiency, runup, markup,

takeover premium

JEL Classification: G14, G34

1

1. Introduction

A financial rumor is an imprecise and unconfirmed message about an impending

financial event. Financial markets are full of such rumors. Rumors can be spread through

word of mouth or newsletters by insiders such as the senior managers or directors of a

company or by outsiders such as investment gurus, professional speculators or financial

journalists. In recent years, the internet provides a new forum for rumors, where investors

can easily exchange information in chat-rooms, by newsgroups and on message boards.

Because of its nature, not all rumors in financial markets are informative. While some

rumormogers might be honest in disseminating their private information, more often

rumors are deliberately added noise (Admati and Pfleiderer, 1986, 1990) because most

rumormongers intend to mislead or manipulate the market by spreading rumors. Of

course, rumors can also be market speculations or predictions based on publicly available

information. Thus, it is extremely important for an investor who receives a financial

rumor to determine whether the rumor conveys a genuine piece of truthful information of

the impending event or it is just a false message intending to manipulate the market. It is

also vital to know how and to what extent a rumor affects the development of the event

and the value of associated financial assets.

This paper is motivated to address these issues by analyzing the rumors of mergers

and acquisitions (M&A) in the US market from 1990 to 2008. Our first goal is to assess

whether stock markets are efficient at responding to published takeover rumors. Through

the data we have collected, we find that although it is impossible to verify whether the

context of each rumor is true or false with certainty at the time when the rumor is

published, investors in the market can statistically distinguish rumors which correctly

2

predict impending takeover events from those which are simply false alarms by analyzing

publicly available information such as historical Cumulative Abnormal Returns (CARs)

of the rumored takeover targets. This suggests that the market prices of a target stock can,

at least partially, reveal the true information of the potential takeover. However, on the

day when a rumor is published and the day after, the abnormal returns of target firms in

the group where an M&A rumor is followed by a formal takeover bid announcement are

statistically indifferent from those of target firms in the group where an M&A rumor does

not lead to any announcement. 1 Moreover, investors can trade on rumors to reap

abnormal returns. A simple investment strategy is to buy the stock of the rumored target

on the rumor publication day, if the firm’s CAR in past 42 or 21 trading days is larger

than a threshold and then to hold the position for one calendar year or until a takeover bid

for the firm is announced, whichever comes first. Our findings show that there is a quite

wide range of the threshold, ranging from zero to 12%, and the investment in the equally-

weighted portfolio of selected firms from the sample can earn an annualized excess return

of up to 100% or more. These findings are in sharp contrast to the efficient markets

hypothesis, even in its semi-strong-form, which states that trading only on public

information cannot earn excess returns (Fama, 1970). Pound and Zeckhauser (1990)

argue that the “market is efficient at responding to published takeover rumor” as they find

that trading on rumors cannot make excess returns. Their trading strategy is buying at the

closing price on the day the rumor is published and selling in the open market at the

closing price on the ending day which is the earlier of the first formal bid announcement

day or one calendar year after the rumor day. A key difference between their strategy and

1 We call the first group rumor-announced group and the second rumor-only group.

3

ours is that they do not distinguish “winners” –– rumors more likely to be followed by a

formal bid from “losers” –– rumors less likely to lead to a bid. More importantly, even

following Pound and Zeckhauser’s (1990) investment strategy to hold a long position in

all rumored targets in our sample, it still yields an excess return between 42% and 55%

per annum. Since selecting winners and/or trading on rumored firms are based on public

information, our empirical evidence does not support the efficient markets hypothesis.

The second goal of this paper is to examine the validity of markup pricing in

M&A. A well-documented observation in corporate control markets is that bidder firms

have to pay substantial premiums to acquire control. A target’s stock price usually has an

abnormal runup before the first takeover bid announcement and thus markup is defined by

the difference between takeover premium and price runup before the first bid. As pointed

out by Schwert (1996), how price runup before the announcement affects the takeover

premium can test two competing views of capital markets. The efficient markets view

predicts that markup should be independent of runup, since the target firm’s stock price

rise before takeover bidding reflects the good news about the value of the firm and such a

rise should make the bidder to increase the takeover premium by an equal amount. On

the other hand, the substitution hypothesis assumes that the bidder’s private information is

not reflected in the market price before the price runup. Thus, runup and markup are

negatively correlated, keeping takeover premium independent of runup. Schwert (1996)

is the first study systematically examining the relationship between pre-bid runup of a

target’s stock price and its post-bid price makeup or the takeover premium. He finds that

a 1% increase in the runup of the target’s CAR leads to approximately a 1% rise in the

total offer premium, supporting the efficient markets view. Because of potential

4

competition among bidders, Betton, Eckbo and Thorburn (2008) use initial offer price to

measure the initial markup. They find that a 1% rise in runup yields an average increase

of 0.8% in the takeover premium implied in the initial offer. It is well recognized that the

runup of a target’s stock price is likely to be driven by the leaked private information

from insiders or legitimate market anticipations (Jarrell and Poulsen, 1989; Schwert,

1996). Thus, the takeover rumor is largely responsible for the price runup before a bid is

formally announced. To pinpoint the impact of rumors, we decompose the runup of a

target’s abnormal return into two parts: the runup before the takeover rumor is published

(pre-runup) and the runup between rumor publication and the announcement of the first

bid (post-runup). Consistent with previous studies, our findings show that both pre-runup

and post-runup have a significantly positive impact on the takeover premium. Different

from these studies, our findings suggest that the impact is much larger in magnitude. In

particular, a 1% increase in the pre-runup (post-runup) of a target’s CAR results in about

a 1.6% (1.2%) increase in takeover premium. There are two potential reasons making our

findings different from the previous findings. First, our sample is biased as it only

includes takeovers preceded by rumors while the samples of Schwert (1996) and Betton,

Eckbo and Thorburn (2008) are much larger and unbiased, including all takeovers. The

second difference is that our runup period (pre-runup period plus post-runup period)

varies across takeover targets and on average it is much longer than 42 trading days

(about two calendar months), adopted by the aforementioned studies for the runup period.

Schwert (1996) finds that CAR starts to rise from 42 trading days before the first bid

announcement. However, we find that runups of rumored target stocks have a quite

different pattern. Runups in 42 or 21 trading days before the first rumor publication (i.e.,

5

pre-runups) are larger than runups between the first rumor publication and first bid

announcement (i.e., post-runups, which are on average of 58.6 trading days). Since pre-

runup also has a more significant marginal effect on takeover premium than post-runup

(i.e., 1.6% vs. 1.2% as reported above), we conclude that pre-runup dominates post-runup

in their role of increasing takeover premium.

In our analysis, event day is the day when the first takeover rumour is published so

that runup period is much longer on average. To be more comparable with previous

studies, particularly with Schwert (1996), this paper also uses the day of the first bid

announcement as the event day and adopts the same estimation window and event

window as Schwert (1996) to test the markup pricing theory in the competition for

corporate control. The results show that markup pricing hypothesis is not consistent with

our empirical findings, at least for the successful takeovers. Thus, our findings not only

differ from Schwert (1996) as markup pricing does not prevail but also differ from the

findings of Betton, Eckbo and Thorburn (2008) based on initial offer prices.2

In both theoretical and empirical analyses, the influence of takeover rumors on the

stock prices of target firms is well recognized. Jarrell and Poulsen (1989) find that the

“presence of rumors in the news media about an impending bid is the strongest

explanatory variable in accounting for unanticipated premiums and prebid runup” for 172

tender offers they have studied. Pound and Zeckhauser (1990) examine the effects of

takeover rumors on the prices of target stocks using a sample of 42 rumors published in

2 However, our findings are, to a certain extent, consistent with the testing results of the markup pricing

hypothesis by Betton, Eckbo and Thorburn (2008) using CAR over the whole takeover contest period rather

than the initial offer price.

6

the Heard on the Street column of the Wall Street Journal (WSJ) from January 1983 to

December 1985. Although target stocks, on average, display significant positive excess

returns in 20 trading days before rumor publication, they find that the market reacts to

rumors efficiently as trading on rumors is not profitable. Zivney, Bertin and Torabzadeh

(1996) extend the study of Pound and Zeckhauser (1990) by documenting rumors appear

not only on the Heard on the Street column but also on the Abreast of the Market column

of the WSJ. More recently, Clarkson, Joyce and Tutticci (2006) examine the market

reaction to takeover rumor postings in the Hotcopper Internet Discussion Site. Their

findings show that a rumor is often associated with an abnormal return and trading

volume during the 10-minute posting interval and an abnormal trading volume in the 10-

minute immediately preceding its posting. Theoretical analyses of rumors are usually

normative. Benabou and Laroque (1992) develop a model, where a rumormonger with

access to private information and incentive to manipulate the market strategically sends a

message of the value of a risky asset to the public. The receivers update their beliefs

based on the message received and their beliefs of the rumormonger’s honesty, and then

take actions which determine the asset price in the market. The model demonstrates that

the rumormonger can manipulate public information and the asset price and such

manipulation exists in the long-run under certain conditions. Using a Kyle (1985) model

with private information diffusion, Bommel (2003) shows that an informed investor with

limited investment capacity can benefit by spreading imprecise rumors of stock prices to

an audience of followers. Following rumors is also beneficial in equilibrium but

uninformed liquidity traders make a loss because of the rumor. In a similar model, Eren

7

and Ozsoylev (2008) illustrate that the very existence of naïve investors makes the hype

and dump manipulation an equilibrium outcome.

The rest of the paper is organized as follows. Section 2 describes the sample used

in our analysis. Sections 3 and 4 demonstrate asset prices before rumor publication can

statistically identify the type of a rumor but price reactions to the rumor on the day of and

the days after rumor publication do not reflect the difference in rumor type. Section 5

tests markup pricing hypothesis and substitution hypothesis based on a sample of rumored

takeover bids. Section 6 examines the robustness of our analysis, focusing on the effects

of rumor sampling and return selection. The final section concludes the paper.

2. Data

The stylized timeline of M&A events is illustrated by Figure 1. Different from

previous studies, we decompose runup period into pre-runup period and post-runup period.

We use a window of 42 or 21 trading days before the first rumor publication for the pre-

runup period while the post-runup period is determined by the observed dates of the first

rumor publication and the first bid announcement. The markup period is standard, which

is the time between the first bid announcement and delisting or 126 trading days (about a

half of a calendar year), whichever comes first.

INSERT FIGURE 1 HERE

Our main data sources include Thomson Financial SDC Platinum database, the

Wall Street Journal, Zephyr, the Centre for Research in Security Prices (CRSP) and

8

Compustat. We first select takeover targets, which must be publicly listed US firms and

identified by SDC of having takeover rumors. The sample period is from January 1, 1990

to December 31, 2008. There are 517 firms satisfy these criteria. Among them, 258 firms

have attracted at least one formal takeover bid (classified as the rumor-announced group)

while the other 259 firms do not have a followed bid offer according to SDC (classified as

the rumor-only group). The rumor information provided by SDC for the rumor-

announced group is quite rough, since it only indicates whether or not a takeover activity

is accompanied with rumors by a “flag” in the rumor column. For our study, the date

when the first rumor is published is vital. Therefore, for all M&A announcements with a

rumor flag in SDC database, we manually search the WSJ and Zephyr to determine the

rumor date.3 We find that there are 75 target firms, of which the date of the first rumor

published in the WSJ or recorded by Zephyr is the same as the first bid announcement

date recorded by SDC. We delete these 75 firms from the rumor-announced sample.

There are 10 and 12 rumors, which have been followed by a formal takeover bid

announcement within one and two days, respectively. The impacts of these rumor

publications are very likely to be intertwined with the impacts of corresponding takeover

announcements. To insure the effect of rumor is not contaminated by takeover

announcement, we require that the first rumor publication is separated from the first bid

announcement by at least two days. Therefore, these 22 rumors are excluded from the

rumor-announced group in our analysis presented in Sections 3-5. However, we have

also conducted a parallel analysis with a rumor-announced group which includes these

rumors in Section 6 and found not substantial variation to the main results reported in

3 Zephyr only provides M&A information from year 2000 onwards.

9

Sections 3-5. For the remaining 161 targets in the group, we drop 53 firms because we

cannot find their rumor dates from the WSJ or Zephyr and further 34 firms because of the

lack of stock return data in CRSP or the insufficient length of estimation window (less

than 100 trading days) for event-study calculation. Thus, the final sample size of the

rumor-announced group is 74. On the other hand, SDC Platinum does provide the rumor

dates for rumors in the rumor-only sample. Thus, we only drop 72 firms from the initial

rumor-only group due to the lack of stock return data in CRSP or the insufficient length of

estimation period. The final sample size of the rumor-only group is 187.4

Table 1 reports the distribution of trading days between the first rumor publication

and the first takeover announcement if the rumor is eventually materialized. The time

period between these events varies significantly. There are 36 target firms which received

a bid within 21 trading days (roughly one calendar month) of rumor publication,

comprising 48.6% of the sample. This result is quite different from Pound and

Zeckhauser (1990) who find that there are only two out of 18 firms which received bids

within 50 calendar days. As Table 1 shows, most sample firms receive bids within one

calendar year. Only two firms receive bids later than one calendar year.5 Since our runup

period is the sum of a certain period before rumor publication (42 days or 21 days) and

the period between rumor publication and bid announcement, the runup period also varies

from one takeover to another. This is quite different from previous studies.

4 The sample size is 188 for the rumor-only group when a 21-day pre-runup period is adopted.

5 We drop these two companies in our analysis below: one has 463 trading days and the other has 680

trading days between rumor publication and announcement. Thus, the final sample size for the rumor-

announced group in the analysis below is 72 firms.

10

INSERT TABLE 1 HERE

There are two competing hypotheses about the sources of takeover rumors. The

public speculation hypothesis believes that rumors derive from research by outside

experts and monitors, representing market predictions of upcoming events. The

alternative hypothesis is that rumors are sourced mainly by leaks from intermediaries

involved in M&A negotiation or insiders of associated firms. Pound and Zeckhauser

(1990) suggest using the duration between the first rumor publication and the first bid

announcement to examine these hypotheses. The intuition behind this examination is that

insider trading or leaks are often accompanied by immediate takeover announcements

while takeovers speculated by the market take a longer period to be materialized. Table 1

can shed some light on this issue and it seems to suggest that a considerable portion of

rumors in the rumor-announced sample derive from insider leaks as 36 rumors out of 74

(48.6%) are followed by a formal M&A announcement within only 21 trading days and

the median duration from the first rumor publication to the first takeover announcement is

23 trading days.

3. Can false rumors be picked up?

The focus of this paper is the abnormal return of target firms in M&A. We apply

the following market model for each target firm in our sample:6

6 To ensure the robustness of our results, we also considered other asset pricing models, including Fama-

French three-factor model (Fama and French, 1993). The results are very similar to these derived from the

single-factor model of Equation (1). Since our main references such as Pound and Zeckhauser (1990),

Schwert (1996), and Betton, Ecobo and Thorburn (2008) use one-factor model, we report our results based

11

itmtiiit RR , (1)

where Rit is the return to the stock of firm i and Rmt is the return to the market index at

date t, αi and βi are regression coefficients. Daily return data of target firms are obtained

from CRSP and the market index uses the data of the CRSP value-weighted market

portfolio. Coefficients αi and βi are first estimated over an estimation window, and then

used to calculate the abnormal return, it , over the pre-runup period leading to the event

or an extended period after the event. Unless otherwise indicated, the event day in this

study is the day when rumor is published for the first time and we consider both 42-day

and 21-day pre-runup periods in our analysis.7 Thus, the estimation window used to

estimate the coefficients in regression (1) is either t = -242 to -43 or -221 to -22. The

cumulative abnormal return of firm i between dates t1 and t2 is given by:

2

1

),( 21

t

ttiti ttCAR . (2)

In Table 2a, the first column reports the average CARs of 72 rumor-announced targets in

the periods of (-42, -1) to (-3, -1). It also reports the average abnormal return on the event

day, CAR(0, 0), and the average CAR of event day and the first day after the event day,

CAR(0, 1). The second column reports their counterparts for 187 rumor-only targets and

the third column documents the mean differences of CARs between the two groups. It is

obvious that the average CARs of the rumor-announced group are consistently larger than

those of rumor-only group for periods from (-42, -1) to (-3, -1). The result of T-test and

on Equation (1) to make our analyses directly comparable to these existing studies. For our analysis using

raw return, see Section 6.

7 We also consider other durations of pre-runup periods and achieve qualitatively similar results.

12

Wilcoxon rank-sum test of mean difference in the forth and fifth columns of the table

show that the differences are statistically significant for most CARs, particularly for CARs

in periods from (-30, -1) to (-14, -1). Thus, although at an individual rumor level an

uninformed investor cannot be sure whether or not a takeover rumor will lead to a formal

bid at the time when he/she receives the rumor, he/she still can statistically determine the

credibility of the rumor by examining the potential target’s historical CARs before the

rumor publication day. The fact that an investor can use market prices or returns to

distinguish rumors suggests the efficiency of the capital market. Accompanying the

publication of a takeover rumor, no matter whether it is credible or not, there must be

some private or public information about the underlying target. The stock prices of the

target seem to be able to correctly incorporate this information and in turn predict the

truthfulness of the rumor, at least in statistical sense.

INSERT TABLES 2A AND 2B HERE

To examine the robustness of this finding, we have repeated our analysis by using

other time windows. The results are qualitatively similar and we document the results

based on the estimation window of (-221, -22) in Table 2b.8 As can be seen from the

table, the most significant mean differences in CARs of the two groups are now in periods

from (-21, -1) to (-14, -1), (-12, -1) to (-11, -1) and (-8, 1) to (-6, -1). In sum, market

8 As a comparison, Pound and Zeckhauser (1990) report a CAR(-21, -1) of 7.78%, based on their full sample

that includes both materialized and unmaterialized rumors. This figure is between our CAR(-21, -1) of

9.25% for the rumor-announced group and 3.06% for the rumor-only group.

13

prices and their movements before the rumor day have the ability to statistically identify a

credible rumor from a false one.

The significantly high CARs of stocks in the rumor-announced subsample are not

at the expense on bearing higher total risk than stocks in the rumor-only group. To show

this point, we have computed each stock’s variance of daily returns using daily return

observations from day -42 to day -1. The mean of 72 variances from the rumor-

announced group is 0.00193 while the mean of 187 variances from the rumor-only group

is 0.00198. The T-statistic and Wilcoxon rank-sum statistic for the test of the mean

difference are -0.0834 and -0.204, respectively, which confirm that the difference in

variances between the two groups is statistically insignificant.9 Consequently, investment

in rumour-announced stocks on average does not bear higher risk than investment in

rumor-only stocks.

Because CAR before rumor publication is an important indicator to detect the type

of a rumor, it is interesting to investigate the relationship between the CAR of a target

firm and the firm’s characteristics. If the price movement reflects the market perspectives

of the potential deal based on public information such as firm characteristics, CAR should

be correlated with these characteristics. A financially distressed firm is more vulnerable

and less likely to survive in the competition and thus is more likely to be a takeover target.

We take financial distress as the focus of our investigation. Of course, other firm

characteristics can also affect market speculation of takeover and the firm’s abnormal

9 For daily returns on day -21 through day -1, the average variances of the two groups are 0.0024 and

0.0023, respectively. T-statistic and Wilcoxon rank-sum statistic for the mean difference tests are 0.0887

and 0.561. Thus, we obtain the same result.

14

return but we include them as control variables in the analysis. More specifically, we

examine the effects of firm characteristics by the following model:

iir

R

rrii eControlDistressCAR

,

1

, (3)

where dependent variable CARi measures target i’s cumulative abnormal return of 42 days

or 21 days before the rumor publication, Distressi measure the degree of the firm’s

financial risk and Controlr,i represents a set of firm-level control variables.

We use two measures for a firm’s financial distress level and bankruptcy

probability in the short run: Altman’s Z-score 10 (Altman, 1968) and multiple-choice

Zmijewski probit model11 (Zmijewski, 1984). A higher Z-score or a lower Zmijewski

probability means that the company is financially healthier in comparison to a lower score

or a higher probability. Distress variable in (3) uses annual data, which is the Z-score or

Zmijewski probability of a target firm in the year immediately before the event date. In

control variables, firm leverage (Leverage) is measured as total debt divided by the sum

of market value of equity and total debt, and firm size (Firm Size) is measured as the

logarithm of market value of equity. Since there are quarterly data on these variables,

their observations in the quarter immediately before the event date are used in regression

(3). The third control variable is sales growth (Sales Growth) of the target firm, which is

10 The measure of Altman's Z-score is: Z = 1.2 (Working capital / Total assets) + 1.4 (Retained earnings

/ Total assets) + 3.3 (Earnings before interest and taxes /Total assets) + 0.6 (Market value of equity /Book

value of total liabilities) + 0.999 (Sales / Total assets).

11 In Zmijewski's model, the probability of bankruptcy is measured by the cumulative probability of the

standard normal distribution at point X, where X = −4.3 – 4.5 (Net income/ Total assets) + 5.7 (Total

liabilities / Total assets) − 0.004 (Current assets /Current liabilities).

15

the average growth rate for the four-quarter period just before the rumor publication.

Industry dummies, including manufacturing (SIC codes 30-39), communication (SIC

codes 48 and 49), finance (SIC codes 60-99) and service (SIC codes 72-82), are also

included as control variables. Data used to calculate these financial variables are obtained

from Compustat database. We test rumor-announced firms and rumor-only firms

separately to see whether there is any difference between the two groups.

INSERT TABLE 3 HERE

Table 3 documents the descriptive statistics of dependent variable and explanatory

variables of (3). The two subsamples of rumor-announced and rumor-only firms are quite

similar in terms of financial leverage, firm size and sales growth. Of our main interest—

financial distress, the two groups are quite similar in the measure of Zmijewski

probability. However, the sample mean of Altman’s Z-score of the rumor-announced

group is much larger than the rumor-only group. This is due to a couple of outliers in the

rumor-announced group. These firms have a Z-score of more than 100. Indeed, the

median Z-scores of the two subsamples have no substantial difference.

The results of the OLS regression of model (3) are reported in Table 4. The

dependent variable are CARi(-42, -1) and CARi(-21, -1) in Panels A and B, respectively.

In the table, each model comprises two regression results for the subsamples of rumor-

announced firms and rumor-only firms. Model 1 and Model 2 are simple OLS

regressions for the relationship between CARi and Distressi variables using Altman’s Z-

score and Zmijewski probability for Distressi, respectively, while Model 3 and Model 4

16

include all control variables in the regression. The first observation from Table 4 is that

the impact of financial distress on CAR is consistent across all models, sample groups and

dependent variable choices. The coefficients of Distressi show that for firms in the

rumor-announced group, an increase in financial distress (a fall in Z-score or a rise in

Zmijewski probability) leads to a greater CAR. However, the signs of the coefficients of

Distressi for the firms in the rumor-only group are just opposite to their counterparts of

rumor-announced group, which implies that an increase in financial distress of a rumor-

only firm leads to a lower CAR. This evidence is consistent with the conjecture that

although there are various sources and transmission networks of takeover rumors,

investors in the market can still utilize some publicly observable variables such as

financial distress in the pre-runup period to judge a firm’s possibility of being a takeover

target. A financially sound firm is less likely to become a takeover target, so that a

marginal deterioration of its financial position does not trigger a stock price rise resulted

from takeover speculation. In contrast, for a financially distressed firm, a similar

marginal deterioration is likely to increase the likelihood of takeover and, in turn, its stock

price rises rather than falls. The impact of Distressi is significant in most scenarios as

evidenced that only 4 out 16 of its coefficients are statistically indifferent from zero based

on heteroskedasticity-consistent standard errors (White, 1980). However, the effects of

control variables are mostly insignificant. An exception is Firm Size. When the

dependent variable is CARi(-21, -1), it is negatively and significantly affects CARi of both

rumor-announced and rumor-only targets.

INSERT TABLE 4 HERE

17

4. Does the market respond to takeover rumors efficiently?

Although market prices are able to statistically predict whether a takeover rumor is

true or false, the puzzle is that such a prediction has not been utilized by investors in the

market. Takeover is usually considered as good news for investing in the target firm and

the market usually responds positively to an M&A announcement.12 Such a positive

effect is also reflected in the surge of stock prices in our sampled targets on the rumor day

and the day after the rumor publication, as shown by CAR(0, 0) and CAR(0, 1) in the

Tables 2a and 2b,13 because the investors anticipate the impending takeover bids. Since

the takeover rumors of the targets in the rumor-only group are false and the market can

statistically identify such rumors, their stock price increases should be smaller than their

counterparts in the rumor-announced group if the market is efficient. Although the

sample mean of rumor day abnormal return, CAR(0, 0), of the rumor-announced group is

greater than its counterpart of the rumor-only group (i.e., 0.0538 vs. 0.0495 in Table 2a),

the two-day cumulative abnormal return, CAR(0, 1), of the rumor-announced group is

actually smaller than its counterpart of the rumor-only group (i.e., 0.0635 vs. 0.0669 in

Table 2a). Moreover, both T-test and Wilcoxon rank-sum test show that the differences

12 Substantial increases in target firm’s stock prices before and after takeover announcements have been

well documented. See, for example, Andrade, Mitchell and Stafford (2001), Jensen and Ruback (1983), and

Keown and Pinkerton (1981).

13 This observation is noticeably different from Pound and Zeckhauser (1990) who find that no significant

excess returns occur on the rumor publication day while the volatility of excess returns on that day is high.

More specifically, they report a mean excess return and standard deviation of 0.07% and 4.19%,

respectively.

18

of these sample means are statistically indifferent from zero. In other words, investors

respond indifferently, on the event day and the day after, to these statistically

distinguishable rumors. This casts a serious doubt on the efficiency of stock markets.

The market prices seem not to efficiently reflect the information available to the public

when takeover rumors appear in the WSJ and/or other media. This conclusion is robust as

we have estimated CAR(0, 0) and CAR(0, 1) with other estimation windows but obtained

qualitatively similar results. For instance, Table 2b reports the estimations based on the

window of (-221, -22). The CAR(0, 0) of the rumor-announced group is 0.0538, which is

greater than the CAR(0, 0) of the rumor-only group, 0.0488. For CAR(0, 1), it is 0.0655

vs. 0.0659. The differences between these two sets of sample means are statistically

indifferent.

To demonstrate further market inefficiency, we show that there is a simple

investment strategy which ensures an investor a statistically significant excess return by

trading on takeover rumors. This strategy is to buy rumored stocks which are more likely

to be followed by a takeover bid. It involves picking a certain threshold such that an

investor buys a dollar worth of the rumored stock at the closing price on the rumor day if

the firm’s CAR in past 42 or 21 days is greater than the threshold and then holds the

position until the first takeover bid is announced or for 252 trading days (one calendar

year), whichever comes first. But the investor takes no action if the firm’s CAR is below

the chosen threshold. The outcome of the investment strategy is documented in Table 5

below. There are 36 rumor-only firms delisted within 252 trading days. Therefore, we

divide tables into two panels: Panel A uses a sample of 72 rumor-announced firms14 and

14 All rumor-announced firms have less than 252 trading days between their rumor day and delisting.

19

151 rumor-only firms which have at least 252 trading days record; while Panel B includes

all rumor-announced and rumor-only firms. Since winner stocks can be any stocks whose

CAR is greater than a specified threshold, Table 5 selectively reports the outcome of the

investment strategy by setting the threshold equal to 0%, 2%, 4%, up to 12% for both

CARi(-42, -1) and CARi(-21, -1). For comparison purpose, it also reports the investment

outcomes if the investor chooses losers to invest, i.e., investing exclusively in stocks with

CARi(-42, -1) and CARi(-21, -1) below the specified thresholds. The last row of each

panel documents the excess returns of longing all rumored stocks.

INSERT TABLE 5 HERE

As Panel A of the table shows, the investment strategy yields annualized excess

returns ranging from 72.2% to 114% if an investor chooses to long winner stocks with a

CARi(-42, -1) greater than 0% to 12%. All these excess returns are statistically significant

at the 1% level. For Panel B, the results are similar to Panel A’s—the investment strategy

yields annualized excess returns from 71.8% to 107.2% with a statistical significance at

the 1% level. An interesting observation is that investing in losers only, i.e., longing

stocks with a CARi(-42, -1) which is smaller than the threshold, still yields a positive

excess return, ranging from 29.1% to 33.6% in Panel A. However, Panel B shows that

investing in losers only can yield negative excess returns and the range of excess returns

is from -7.3% to 11.8%. Moreover, all excess returns from investing in losers are

statistically indifferent from zero. Table 5 also reports the results of using the CARs of

21 days before rumor publication to select stocks for investment. Clearly, the results are

20

qualitatively equivalent to using CARi(-42, -1) to select stocks although excess returns are

quantitatively different.15

Pound and Zeckhauser (1990) suggest another trading strategy of buying all

rumored target stocks on the day of rumor publication and holding them until the first bid

or for a calendar year, whichever comes first. Given that such trading strategy cannot

yield statistically significant excess returns for their sampled firms, they conclude that the

market is efficient because trading on rumor is not profitable. There are two substantial

differences between our approach/findings and those of Pound and Zeckhauser’s. First,

their investment strategy does not utilize publicly available price information to determine

which rumored targets to buy. Our investment strategy uses this price information but not

other private or public information at all and it is still profitable. Second, perhaps more

surprisingly, even following Pound and Zeckhauser’s investment strategy and long all

rumored target firms in our sample, investors can still earn excessive profits. As shown in

the last rows of Panels A and B of Table 5, such an investment strategy yields annualized

excess returns ranging from 41.9% to 54.5%. They are statistically significant at least at

the 5% level. There are only 42 target firms in the Pound-Zeckhauser sample, which is

much smaller than our samples of 223 to 260 firms. Moreover, Pound and Zeckhauser’s

(1990) sample period is from January 1, 1983 through December 31, 1985 while ours

covers a more recent and longer period, from January 1, 1990 through December 31, 2008.

The divergence in sample size and sample period is likely to be the reasons for the

disparity in investment outcomes.

15We have also tried other CARs, for instance, CARi(-50, -1), to select stocks and found no substantial

difference to those reported in Table 5.

21

It is also interested to compare the risk of investing in the winner stocks with that

of loser stocks. The column, Average variance, in Table 5 compiles the average of

annualized variances of the daily returns of invested stocks over the investment periods.

Obviously, for each pair of winner set and loser set, the former has a much smaller

average variance. This is not surprising. Most winner stocks experience a small and

stable gain between rumor day and delisting while most loser stocks experience a large

loss over a period a few days after rumor publication. It is also worth noting that

investing in winner stocks does not bear very high total risk as the average variances falls

in a range of 0.018 to 0.050. In comparison, the average annualized variance of the CRSP

market portfolio is 0.0308 over the period of 1990-2008.

5. Do bidders markup price?

An implication of efficient markets is that the stock price movement of a target

firm, before takeover bidding, reveals its value changes unknown to bidders. Since the

markup of a takeover bid represents the bidder’s willingness to pay for the impending

takeover, a one-dollar increase in the target’s stock price in the runup period should on

average result in a one-dollar rise in the takeover premium according to the view of

efficient markets. A contrasting view is that runup is not caused by new information.

Because the runup merely reflects the anticipation of a planned takeover premium, the

offer premium should be independent of runup. By definition, premium equals the sum of

runup and markup, this view implies a perfect substitution hypothesis that any increase in

runup will be eased by a decline in markup leaving the premium unchanged. We revisit

this issue of relationship between runup and takeover premium using our newly collected

22

data. But departing from the conventional analysis, we decompose runup into pre-runup

and post-runup to better understand the effects of rumor on takeover premium. The pre-

runup of target i is calculated using its CAR between day -42 (or -21) to day -1, i.e.:

1

2142 or titirunupPre . (4)

The post-runup is the CAR in the post-runup period so that it can be calculated by:

1

0

T

titirunupPost , (5)

where T is the date of the first bid announcement. The definition of markup is

conventional, which is the CAR from the day of the first bid announcement through

delisting or 126 trading days, whichever comes first:

}126min{ delisting ,T

TtitiMarkup . (6)

The total premium, Premiumi, paid by a successful bidder is the sum of Pre-runupi, Post-

runupi and Markupi.16 There are 60 out of 72 rumor-announced firms which have been

taken over in the end of the takeover process and we classify them into the successful

sample. To distinguish from it, we call the sample of all 72 rumor-announced firms the

rumor-announced sample. Panels A and B in Table 6 report the descriptive statistics of

pre-runup, post-runup, runup, markup and premium calculated based on an estimation

window of (-242, -42) of these two samples, respectively, while Panels C and D report the

same items based on the (-221, -22) window.

16 To be comparable with the existing literature, the “premiums” of unsuccessful takeover are also included,

and it is defined as by the sum of Pre-runup, Post-runup and Markup too.

23

INSERT TABLE 6 HERE

Previous studies find that runup is usually more substantial at the time when it is

closer to the first bid announcement.17 Table 6 shows a quite different runup pattern. For

both 42- or 21-day pre-runup periods, pre-runup on average dominates post-runup. Since

the average length of post-runup periods is around 58.6 trading days, such dominance is

not due to a shorter post-runup period. Moreover, we cannot reject the hypothesis that the

mean of post-runup is zero. Therefore, the main runup of the target’s stock price occurs

in the pre-runup period rather than post-runup period although the latter is closer to the

takeover bid announcement than the former.

The average markups documented in the table are negative under various

scenarios although they are not significantly different from zero. The negative markup is

consistent with an initial overreaction of the market to the takeover rumors and then the

market and/or bidders correct the overreaction in bidding or consecutive trading. In other

words, instead of marking up, bidders actually mark down from runups in their offers on

average. This is particularly obvious by comparing the successful subsample with the

rumor-announced sample, in that the former has a larger markdown than the latter. To

investigate the effect of a change of runup on the changes of premium and markup we,

following Schwert (1996), consider a regression model that:

Premiumi = a + b1 Pre-runupi + b2 Post-runupi + ui. (7)

17 “The CARs start to rise around day -42 (about two months before the first bid announcement), with the

largest pre-bid rise occurring from days -21 to - 1.” (Schwert, 1996, page 162). Note that day 0, the event

day, in Schwert’s analysis is the day of the first bid announcement.

24

This regression difference from Schwert (1996) by decomposing Runup into Pre-runup

and Post-runup.18 The results of the regressions are reported in Table 7 below. As we

can see from the table, both Pre-runup and Post-runup contribute significantly to

takeover Premium. We can also strongly reject the hypotheses that b1 ≤ 1 for the

successful sample. For instance, the heteroskedasticity-consistent T-statistic for the null

hypothesis of b1 ≤ 1 of the successful sample with pre-runup measure of CAR(-42, -1) is

3.359, and we can reject the null hypothesis at the 1% level. More importantly, the

marginal effect of Pre-runup is considerably greater than its counterpart of Post-runup

(i.e., 21 bb ). Table 6 shows that the size of Pre-runup is much larger than that of Post-

runup on average. Thus, pre-runup not only dominates post-runup in terms of marginal

effect on takeover premium since 21 bb , but also in terms of the overall impact on

takeover premium.

INSERT TABLES 7 AND 8 HERE

To examine markup pricing hypothesis further, we study a regression model that

is exactly the same as Schwert (1996):

Premiumi = a + b Runupi + ui. (8)

The results in Panels A and B of Table 8 are the regression outcomes using the rumor-

announced sample and the successful subsample, respectively. They demonstrate that the

null hypothesis of b = 1 can be strongly rejected in 3 out of 4 scenarios using the

heteroskedasticity-consistent T-statistics. Thus, we have strong empirical evidence

18 A test shows that the correlation between pre-runup and post-runup is statistically insignificant.

25

against markup pricing hypothesis for the takeovers in our successful sample. Note that

Premiumi = Runupi + Markupi by definition. Therefore, Markupi = a + (b – 1) Runupi,

and Markupi and Runupi are perfectly and negatively correlated if and only if b = 0. Our

result of b > 1 also strongly rejects the substitution hypothesis because it requires b = 0.

The estimates of b in Panels A and B are larger than Schwet’s (1996) estimation. The

estimates of b based on his Main and Successful samples are 1.075 and 1.018,

respectively, and the null hypothesis of markup pricing (b = 1) cannot be rejected. In a

slightly different fashion, Betton, Eckbo and Thorburn (2008) obtain an estimate around

0.8 when markup is measured by the natural logarithm of the ratio of initial offer price to

the target stock price on the day before the first bid.19

There are two potential reasons making our results different from those of

previous studies. First, we focus on M&A deals preceded by relevant rumors while the

samples of Schwert (1996) and Betton, Eckbo and Thorburn (2008) include all target

firms which are publicly traded in the U.S. Their sample sizes are much larger than ours.

As a result the impact of takeover rumors discovered by this paper might have been eased

by price dynamics of other target firms in their samples over the runup period. The

second reason relates to the duration of runup period. Our runup period is much longer

than a 42-day period used by Schwert (1996) and Betton, Eckbo and Thorburn (2008).

Our runup period equal to 42 or 21 days of pre-runup period plus an average post-runup

period of 58.6 days, leading to an average runup period of 100.6 or 79.6 days. 19 Their runup is measured in a similar way so that market risk has not been removed from markup and

runup. However, they also use a market model similar to equation (1) to estimate CAR in the runup period

and markup period, which implies an estimate of b of 1.595 based on their full sample and 1.493 based on

their successful sample. These estimates are closer to these reported in Panels A and B of Table 8.

26

To address the second difference, we re-run model (8) with a runup period of 42

days, irrespective of the different lengths of actual post-runup periods. More specifically,

we choose the first bid announcement day as the event day, similar to Schwert (1996),

and Betton, Eckbo and Thorburn (2008). The estimation period starts from days -379 and

ends on day -127 and the runup is the CAR over day -42 to day -1 (before the first bid

announcement). The markup is the CAR from the first bid announcement through

delisting or 126 trading days after the first bid, whichever comes first. We deliberately

choose estimation window, runup period and markup period in this way so that they are

the same as those adopted by Schwert (1996). The descriptive statistics of runup, markup

and takeover premium under this specification are presented in Table 9, where Panel A

includes all rumor-announced firms and Panel B includes only targets which have been

successfully taken over.20 The average CARs of all rumor-announced firms from day -

126 to day 252 are illustrated in Figure 2. A notable difference of Table 9 from the

statistics in Table 6 is that the mean markups are now positive rather than negative,

although they are still statistically insignificant from zero. Comparing Figure 2 with

Figure 2 of Schwert (1996), we find substantially larger early runups in the rumored

M&A processes. The average CAR starts to move up as early as around day -120 while

the average CAR of the Schwert sample does not show substantial upward movement

until day -42. This difference reconfirms the impact of rumor on runups, as pointed

earlier.

20 The samples are two firms smaller than their counterparts in Panels A and B of Table 6 because of data

unavailability.

27

INSERT TABLE 9 AND FIGURE 2 HERE

The regression results of using takeover bid announcement as the event are

reported in Panel C of Table 8. First note that the heteroskedasticity-consistent T-statistic

testing b ≤ 1 tends to decline from Panels A and B to C. This implies the role played by

early runups before and/or around rumor publication has been partially excluded in Panel

C. The estimate of b based on the successful sample in Panel C is 1.466, which is still

substantially larger than Schwert’s estimates. It is however consistent with the estimate

of Betton, Eckbo and Thorburn (2008) based on their successful sample with runups and

markups measured by CARs. But our estimate based on the rumor-announced sample is

considerably smaller than their estimate based on their full sample (1.021 vs. 1.493).

More importantly, the hypothesis of b ≤ 1 can be rejected for our successful sample whilst

the hypothesis of b = 1 cannot be rejected for our rumor-announced sample. In sum, our

findings reject substitution hypothesis and cannot be fully explained by the markup

pricing theory, at least for the successful sample. The difference between our finding and

those of previous studies, Schwert’s (1996) in particular, indicates that the very existence

of M&A rumors can have some material impact on the bidders’ pricing strategies and the

final realizations of takeover premiums.

The super-markup pricing (b > 1), found in this study, is consistent with the hubris

hypothesis (Roll, 1986), in that bidders are interested in winning a takeover contest

irrespective of costs. However, it can also be explained by reevaluating the synergy of

the merger and/or redistributing the gain of the merger in the takeover contest between the

28

target and bidder firms. Testing these hypotheses is beyond the scope of this paper and

remains for future studies.

6. Further robustness examinations

To ensure our findings are robust, we have conducted various robustness checks.

In addition to what have reported in the above analysis, we briefly report and discuss in

this section the results of adopting alternative sampling of rumor-announced group and

using raw returns rather than abnormal returns for our analysis.

6.1. Alternative sampling of rumor-announced group

As mentioned in Section 2, we required in the above analysis that the first rumor

publication and the first bid announcement are separated by two days to isolate rumor

publication effects from bid announcement effects. Thus, we have dropped 22 targets

from the rumor-announced group as 10 of them have attracted a takeover bid within one

day of rumor publication and the other 12 targets attracted a bid within two days. Adding

these 12 or 22 targets back into the rumor-announced group does not change the main

findings reported in the previous sections. We summarize the results of using this

alternative sampling below.

First and obviously, if the 12 targets (or 22 targets) are included in the sample, the

number of observed targets with less than 21 trading days between rumor publication and

bid announcement increases from 36 in Table 1 to 48 (or 58). In terms of percentage, the

increase is from 48.6% to 55.8% (or 60.4%). The average length and median length of

the time interval between rumor publication and bid announcement are reduced from 58.6

29

and 23 days to 50.7 and 13.5 (or 45.5 and 10.0) days, respectively. This provides stronger

evidence supporting the hypothesis that rumors in the rumor-announced group are

sourced from insider leaks.

Second, the average CARs of the rumor-announced in the pre-runup period are

still greater than the average CARs of the rumor-only group. Corresponding to Table 2a,

the mean differences of CAR(-30, -1) to CAR(-14, -1) between the two groups are in the

range of 0.0166 (or 0.0208) to 0.0472 (or 0.0543) when the 12 (or 22) targets are included

in the rumor-announced group. Most of them are significantly different from zero though

they are smaller than their counterparts obtained from the original sampling.

Corresponding to Table 2b, the mean differences of CAR(-21, -1) to (-14, -1) between the

two groups are in the range of 0.0158 (or 0.0196) to 0.043 (or 0.0483) and most of them

are also significantly different from zero.

Third, the differences of abnormal returns on the rumor day between the two

groups and the CARs over the rumor day and the day after are still statistically

indistinguishable.21 The mean difference of CAR(0, 0) and CAR(0, 1) between the two

groups are every small and their values, corresponding to Table 2a, are equal to 0.0094

(or 0.0145) and 0.0055 if the 12 (or 22) targets are included. Their counterparts for Table

2b are 0.0102 (or 0.0154) and 0.0082. All these mean differences are statistically

insignificant.

21 Note, CAR(0, 1) cannot be used to measure the market response to rumor if the rumor-announced group

includes targets whose bid is announced within one day of rumor publication because CAR(0, 1) includes

abnormal return on announcement day, which is the major part of markup.

30

Forth, the investment strategy specified in Section 4 still yields substantial excess

returns. For instance, corresponding to Panel A of Table 5, we find that selecting winner

stocks from a sample including the 12 (or 22) new targets with a threshold of 0% to 12%

for the CAR over 42 days before rumor publication leads to an annualized excess return

from 101.8% (or 145.8%) to 171.1% (or 230.8%). Using the CAR over 21 days before

rumor publication to select, the range is from 132.4% (or 175.7%) to 194.3% (or 224.8%).

Investing in all rumored targets yields an annualized excess return of 93% (or 122%).

Obviously, these results are stronger than what are documented in Panel A of Table 5. A

similar conclusion can be drawn for Panel B. The reason for higher excess returns of the

investment strategy by adding these targets into the sample is that rumor publication day

and the day after usually have a very high excess return. Thus, adding these targets not

only enlarges the amount of winner stocks but also provides an opportunity for investors

to hold very high returns for a very short period (i.e., one or two days).

Fifth, the regression outcomes of models (7) and (8) have no qualitative changes

after more targets are added back into the sample. For instance, the estimates of b1,

corresponding to the successful sample in Panels A and B of Table 7, are 1.569 (or 1.484)

and 1.716 (or 1.616) when 12 (or 22) new targets are added into the sample. The

estimates of b, corresponding to the successful sample in Panels A and B of Table 8, are

1.352 (or 1.319) and 1.416 (or 1.384) when 12 (or 22) new targets are added into the

sample. All of these eight estimates are strongly and significantly different from one.

In sum, the main findings of this paper are robust to the sample selection of the

rumor-announced group.

31

6.2. Analysis with raw returns

The choice of return is critical to our analysis. Instead of abnormal returns, we

have also used raw return data to repeat the analysis conducted in Section 3 through

Section 5.22 More specifically, we define cumulative raw return as 2

1),( 21

t

tt iti RttCRR

and conduct the analysis generating Tables 2a-b and 5-8 by replacing CAR by CRR. With

raw returns, the difference of mean returns between the rumor-announced group and

rumor-only group is larger and statistically more significant than what are reported in

Tables 2a and 2b. For instance, corresponding to Table 2a, almost all differences of

CRR(–42, –1) through CRR(–14,–1) are significant at the 1% level by both T-test and

Wilcoxon rank-sum test, and corresponding to Table 2b almost all differences of CRR(–

21, –1) through CRR(–14,–1) are significant at the 1% level. Based on CRR(41, 1) or

CRR(21, 1) to select winner stocks, the winner stocks on average perform better than

the loser stocks gauged by raw returns after rumor publication. The results similar to

Table 5 are obtained when CAR is replaced by CRR. However, using CRRi to run

regressions (7) and (8) lead to somewhat different estimates than using CARs. For

instance, the estimates of b1 are smaller than their counterparts in Table 7 while the

estimates of b2 are larger. Nevertheless, the estimates of b, based on the successful

subsample, are quite consistent with their counterparts in Table 8 and are significantly

greater than one.

Following the process of Subsection 6.1, we then add back the 22 targets, whose

first merger announcement day are within two days of the first rumor publication day,

22 See, for instance, McConnell and Sanger (1987), and Jaffe and Mahoney (1999) for the adoption of raw

returns.

32

into the sample and repeat the above analysis process of testing CRRs. Moreover, there

are 5 rumor-announced targets and 14 rumor-only targets, which have no sufficient CAR

data but complete data of CRR. We add these 19 firms into the sample further and

duplicate the analysis process again. All three different samples yield quite consistent

results.

7. Conclusions

Using the data of M&A rumors of publicly traded US target firms, we find quite

contradicting evidences regarding capital markets efficiency. On the one hand, stock

prices of these rumored firms before rumor publication can be used to statistically

distinguish a genuine prediction of takeover from a false alarm, indicating that prices

largely assemble and reflect market information. On the other hand, the market

participants do not seem to fully utilize the information and leaves some profitable

opportunities unexplored. In addition, price runups in our sample seem to appear earlier

than what are reported in previous studies. Pre-rumor runups dominate post-rumor

runups not only in their magnitudes but also in the marginal effect on takeover premiums.

Markup pricing hypothesis can be rejected for rumored M&A deals, if the runup period is

extended to 42 days before rumor publication. Even with a standard runup period of 42

days before the first bid announcement, markup pricing hypothesis still cannot fully

explain the empirical evidence documented in this paper. The competing substitution

hypothesis of takeover premiums is also strongly rejected by our empirical evidence,

irrespective of the choice of the length of runup period. How to resolve the puzzling

findings of this paper , both theoretically and empirically, remains for future research.

33

References

Admati, A., Pfleiderer P., 1986. A monopolistic market for information. Journal of

Economic Theory 39, 400-438.

Admati, A., Pfleiderer P., 1990. Direct and indirect sale of information. Econometrica 58,

901-928.

Altman, E.I., 1968. Financial ratios, discriminate analysis and the prediction of corporate

bankruptcy. Journal of Finance 23, 589-609.

Andrade, G., Mitchell, M., Stafford, E., 2001. New evidence and perspectives on mergers.

Journal of Economic Perspectives 15, 103-120.

Benabou, R., Larogue, G., 1992. Using privileged information to manipulate markets:

Insiders, gurus, and Credibility. Quarterly Journal of Economics 107, 921-958.

Betton, S., Eckbo, B.E., Thorburn, K., 2008, Markup pricing revisited.

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1094946.

Clarkson, P.M., Joyce, D., Tutticci, I., 2006. Market reaction to takeover rumor in Internet

Discussion Sites. Accounting and Finance 46, 31–52.

Eren, N., Ozsoylev, H.N., 2008. Hype and Dump Manipulation.

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=948814.

Fama, E.F., 1970. Efficient capital markets: A review of theory and empirical work.

Journal of Finance 25, 393-417.

Fama, E.F., French, K.R., 1993. Common risk factors in the returns on stocks and bonds.

Journal of Financial Economics 33, 3–56.

Jaffe, J.F., Mahoney, J.M., 1999. The performance of investment newsletters. Journal of

Financial Economics 53, 289-307.

34

Jarrell, G., Poulsen, A., 1989. Stock trading before the announcement of tender offers:

Insider trading or market anticipation? Journal of Law, Economics, and

Organization 50, 225-248.

Jensen, M.C., Ruback, R.S., 1983. The market for corporate control: the scientific

evidence. Journal of Financial Economics 11, 5-50.

Keown, A.J., Pinkerton, J.M., 1981. Merger announcements and insider trading activity:

an empirical investigation. Journal of Finance 34, 855-869.

Kyle, A.S., 1985. Continuous auctions and insider trading. Econometrica 53, 1315-1335.

McConnell, J.J., Sanger, G.C., 1987. The puzzle in post-listing common stock returns.

Journal of Finance 42, 119-140.

Pound, J., Zeckhauser, R., 1990. Clearly heard on the street: The effect of takeover

rumors on stock prices. Journal of Business 63, 291-308.

Roll, R., 1986. The hubris hypothesis of corporate takeovers. Journal of Business 59, 197-

216.

Schwert, W., 1996. Markup pricing in mergers and acquisitions. Journal of Financial

Economics 41, 153-192.

Van Bommel, J., 2003. Rumors. Journal of Finance 58, 1499-1519.

White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct

test for heteroskedasticity. Econometrica 48, 817-838.

Zivney, T.L., Bertin, W.J., Torabzadeh, K.M., 1996. Overreaction to takeover speculation.

Quarterly Review of Economics and Finance 36, 89-115.

Zmijewski, M.E., 1984. Methodological issues related to the estimation of financial

distress prediction models. Journal of Accounting Research 22, 59-82.

35

Fig. 1. Stylized timeline of merger and acquisition events.

0

0.05

0.1

0.15

0.2

0.25

0.3

-126

-120

-114

-108

-102 -96

-90

-84

-78

-72

-66

-60

-54

-48

-42

-36

-30

-24

-18

-12 -6 0 6 12 18 24 30 36 42 48 54 60 66 72 78 84 90 96 102

108

114

120

126

132

138

144

150

156

162

168

174

180

186

192

198

204

210

216

222

228

234

240

246

252

Trading day

Ave

rag

e cu

mu

lati

ve a

bn

orm

al r

etu

rn

Fig.2. Average cumulative abnormal return. The average of cumulative abnormal return to target firms in the rumor-announced group from trading day -126 to 252 is estimated based on the market model itmtiiit RR , where market index is the CRSP value-

weighted market portfolio. The regression of the market model uses daily returns from day -379 to day -127. The event day, day 0, is the day when the first formal bid is announced.

First rumor publication

Final outcome

Pre-runup period Post-runup period

Runup period Markup period

First bid announcement

36

Table 1 Distribution of trading days between the first rumor publication and the first takeover announcement.

The calculation is based on the original data from SDC Platinum database, the Wall Street Journal and Zephyr database. The sample period is from January 1, 1990 to December 31, 2008.

Trading days between rumor and bid Number of observations Less than 21 days 36 22-42 days 10 43-100 days 14 101-200 days 11 201-300 days 1 More than 300 days 2 Average length: 58.6 days 74 Median length: 23.0 days 74

37

Table 2a Average CARs of rumor-announced and rumor-only targets and their differences, estimation based on time series between (-242,-43).

Event day (day 0) is the day when the takeover rumor is published for the first time. Using the daily return data of target i from CRSP and daily returns of the CRSP value-weighted market portfolio over (-242,-43), coefficients αi and βi are estimated by the following market model: itmtiiit RR .

Then the estimated αi and βi are substituted into the model to calculate the abnormal return, it , over (-42, -

1), using the observations of daily returns of target i and the CRSP market portfolio over this period. The CAR of firm i between dates t1 and t2 is given by:

2

1

),( 21

t

tt

iti ttCAR .

There are 72 rumor-announced firms and 187 rumor-only firms in the sample. The first column reports the average CARs of rumor-announced firms in the periods of (-42, -1) to (-3, -1), the average abnormal return on the event day, CAR(0, 0), and the average CAR of event day and the first day after the event day, CAR(0, 1). The second column reports their counterparts for the rumor-only group and the third column documents the differences between the two groups. The results of T-test and Wilcoxon rank-sum test of mean difference are reported in the forth and fifth columns, respectively. Standard deviations are reported in parentheses. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively.

CAR Rumor-

announced Rumor-only

Mean difference

T-statistic Wilcoxon rank-

sum

CAR(-42,-1) 0.0738***

(0.236) 0.029* (0.229)

0.0448 1.397 1.342

CAR(-41,-1) 0.0753** (0.243)

0.029* (0.225)

0.0463 1.451 1.339

CAR(-40,-1) 0.0784***

(0.234) 0.0304* (0.224)

0.048 1.521 1.472

CAR(-39,-1) 0.074*** (0.237)

0.0312* (0.221)

0.0428 1.373 1.300

CAR(-38,-1) 0.074*** (0.229)

0.034** (0.221)

0.040 1.293 1.196

CAR(-37,-1) 0.078*** (0.209)

0.0302* (0.219)

0.0478 1.597 1.429

CAR(-36,-1) 0.0805***

(0.219) 0.028* (0.213)

0.0525 1.76* 1.503

CAR(-35,-1) 0.0776***

(0.216) 0.027* (0.213)

0.0506 1.704* 1.542

CAR(-34,-1) 0.0804***

(0.202) 0.0219 (0.211)

0.0585 2.021** 1.737*

CAR(-33,-1) 0.084*** (0.209)

0.0184 (0.208)

0.0656 2.273** 1.942*

CAR(-32,-1) 0.0798***

(0.203) 0.0233 (0.208)

0.0565 1.974** 1.602

CAR(-31,-1) 0.083*** (0.207)

0.0224 (0.204)

0.0606 2.132** 1.807*

CAR(-30,-1) 0.0914***

(0.21) 0.0211 (0.201)

0.0703 2.483** 2.057**

CAR(-29,-1) 0.0916***

(0.211) 0.0245* (0.197)

0.0671 2.40** 2.157**

CAR(-28,-1) 0.093*** (0.206)

0.025* (0.189)

0.068 2.54** 2.301**

38

CAR(-27,-1) 0.0924***

(0.198) 0.0227* (0.178)

0.0697 2.728*** 2.524**

CAR(-26,-1) 0.0907***

(0.193) 0.0231* (0.175)

0.0676 2.702*** 2.492**

CAR(-25,-1) 0.083***

(0.19) 0.0232* (0.171)

0.0598 2.435** 2.174**

CAR(-24,-1) 0.0892***

(0.193) 0.0258** (0.169)

0.0634 2.60*** 2.285**

CAR(-23,-1) 0.0904***

(0.181) 0.024* (0.171)

0.0664 2.747*** 2.485**

CAR(-22,-1) 0.0911***

(0.178) 0.0283** (0.165)

0.0628 2.675*** 2.651***

CAR(-21,-1) 0.0922***

(0.19) 0.029** (0.166)

0.0632 2.634*** 2.718***

CAR(-20,-1) 0.0933***

(0.184) 0.0294** (0.158)

0.0639 2.774*** 2.507***

CAR(-19,-1) 0.0931***

(0.184) 0.031*** (0.154)

0.0621 2.748*** 2.466**

CAR(-18,-1) 0.0966***

(0.194) 0.0304***

(0.156) 0.0662 2.855*** 2.74***

CAR(-17,-1) 0.0974***

(0.186) 0.0336***

(0.162) 0.0638 2.718*** 2.82***

CAR(-16,-1) 0.0869***

(0.185) 0.0359***

(0.143) 0.051 2.354** 2.20**

CAR(-15,-1) 0.0877***

(0.189) 0.0353***

(0.136) 0.0524 2.47** 2.235**

CAR(-14,-1) 0.0807***

(0.181) 0.0372***

(0.132) 0.0435 2.128** 2.059**

CAR(-13,-1) 0.0782***

(0.177) 0.0375***

(0.130) 0.0407 2.035** 1.563

CAR(-12,-1) 0.0817***

(0.180) 0.0314***

(0.124) 0.0503 2.551** 1.735*

CAR(-11,-1) 0.0842***

(0.176) 0.0319***

(0.125) 0.0523 2.677*** 1.94*

CAR(-10,-1) 0.080*** (0.178)

0.033*** (0.122)

0.047 2.438** 1.333

CAR(-9,-1) 0.0729***

(0.163) 0.0285***

(0.118) 0.0444 2.42** 1.422

CAR(-8,-1) 0.0642***

(0.147) 0.0228***

(0.115) 0.0414 2.38** 1.881*

CAR(-7,-1) 0.0626***

(0.144) 0.018** (0.101)

0.0446 2.81*** 2.107**

CAR(-6,-1) 0.0565***

(0.133) 0.019** (0.102)

0.0375 2.42** 1.87*

CAR(-5,-1) 0.0543***

(0.14) 0.0243***

(0.108) 0.030 1.84* 1.287

CAR(-4,-1) 0.0509***

(0.138) 0.0269***

(0.089) 0.024 1.645 1.12

CAR(-3,-1) 0.0474***

(0.126) 0.0205***

(0.084) 0.0269 1.985* 1.181

CAR(0,0) 0.0538***

(0.141) 0.0495***

(0.128) 0.0043 0.236 1.176

CAR(0,1) 0.0635***

(0.136) 0.0669***

(0.241) -0.0034 -0.111 0.859

39

Table 2b Average CARs of rumor-announced and rumor-only targets and their differences, estimation based on time series between (-221,-22).

The estimation process for CAR is the same as Table 2a except for using an estimation window of (-221,-22). There are 72 rumor-announced firms and 188 rumor-only firms in the sample. The first column reports the average CARs of rumor-announced targets in the periods of (-21, -1) to (-3, -1), (0, 0) and (0, 1). The second column reports their counterparts for the rumor-only group and the third column documents the differences between the two groups. The results of T-test and Wilcoxon rank-sum test of mean difference are reported in the forth and fifth columns, respectively. Standard deviations are reported in parentheses. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively.

CAR Rumor-announced Rumor-only Mean difference T-statistic Wilcoxon rank-sum

CAR(-21,-1) 0.0925***

(0.195) 0.0306** (0.168)

0.0619 2.54** 2.577**

CAR(-20,-1) 0.0946***

(0.190) 0.0310***

(0.160) 0.0636 2.716*** 2.479**

CAR(-19,-1) 0.0943***

(0.190) 0.0320***

(0.156) 0.0623 2.703*** 2.49**

CAR(-18,-1) 0.0969***

(0.199) 0.0317***

(0.158) 0.0652 2.763*** 2.639***

CAR(-17,-1) 0.0979***

(0.192) 0.0347***

(0.165) 0.0632 2.638*** 2.737***

CAR(-16,-1) 0.0871***

(0.191) 0.0371***

(0.145) 0.050 2.274** 2.048**

CAR(-15,-1) 0.0878***

(0.194) 0.0366***

(0.137) 0.0512 2.387** 2.121**

CAR(-14,-1) 0.0807***

(0.185) 0.0388***

(0.134) 0.0419 2.018** 1.979**

CAR(-13,-1) 0.0781***

(0.180) 0.0391***

(0.130) 0.0390 1.931* 1.515

CAR(-12,-1) 0.0812***

(0.182) 0.0328***

(0.126) 0.0484 2.424** 1.756*

CAR(-11,-1) 0.0836***

(0.177) 0.0334***

(0.127) 0.0502 2.541** 1.913*

CAR(-10,-1) 0.0801***

(0.179) 0.0339***

(0.123) 0.0462 2.363** 1.366

CAR(-9,-1) 0.0727***

(0.164) 0.0299***

(0.121) 0.0428 2.30** 1.485

CAR(-8,-1) 0.0634***

(0.147) 0.0246***

(0.118) 0.0388 2.207** 1.808*

CAR(-7,-1) 0.0633***

(0.143) 0.0198** (0.104)

0.0435 2.704*** 2.232**

CAR(-6,-1) 0.055*** (0.131)

0.0207*** (0.105)

0.0343 2.199** 1.843*

CAR(-5,-1) 0.0545***

(0.137) 0.0251***

(0.108) 0.0294 1.819* 1.513

CAR(-4,-1) 0.0508***

(0.138) 0.0277***

(0.090) 0.231 1.583 1.122

CAR(-3,-1) 0.0471

(0.125)*** 0.0214

(0.086)*** 0.0257 1.888* 1.176

CAR(0,0) 0.0538***

(0.141) 0.0488***

(0.128) 0.050 0.275 1.227

CAR(0,1) 0.0655***

(0.137) 0.0659***

(0.241) -0.0004 -0.014 1.004

40

Table 3 Descriptive statistics of CARs and firm characteristics.

The sample of 259 US publicly listed target firms is split into two subsamples: the rumour-announced group in Panel A and the rumour-only group in Panel B. The descriptive statistics of CARi are based on data used in Tables 2a and 2b. The statistics of firm characteristics are calculated using data from Compustat from 1990 to 2008. Altman’s Z-score is calculated by formula Z = 1.2 (Working capital / Total assets) + 1.4 (Retained earnings / Total assets) + 3.3 (Earnings before interest and taxes /Total assets) + 0.6 (Market value of equity /Book value of total liabilities) + 0.999 (Sales / Total assets). Zmijewski Probability of bankruptcy is measured by the cumulative probability of the standard normal distribution at point X, where X = −4.3 – 4.5 (Net income/ Total assets) + 5.7 (Total liabilities / Total assets) − 0.004 (Current assets /Current liabilities). Both Altman Z-score and Zmijewski Probability use the annual data of the year immediately before rumor publication. Leverage is measured as total debt divided by the sum of market value of equity and total debt, Firm Size is the logarithm of market value of equity. These two variables use quarterly observations in the quarter just before the event date. Sales Growth is the average growth rate of sales of the four quarters just before the rumor publication.

Number of firm

Mean Standard deviation

Min 25% Median 75% Max

Panel A: Rumour-announced group CARi(-42,-1) 70 0.065 0.231 -0.698 -0.048 0.046 0.189 1.115 CARi(-21,-1) 71 0.091 0.196 -0.240 -0.005 0.059 0.142 1.069 Altman’s Z-score

49 7.401 28.045 -6.956 1.241 2.498 3.253 195.604

Zmijewski Probability

50 0.297 0.315 0.0001 0.018 0.188 0.511 0.999

Leverage 63 0.359 0.282 0 0.142 0.268 0.601 0.98 Firm size 67 7.18 2.14 1.016 6.152 7.562 8.776 10.632 Sales growth 56 0.086 0.113 0.0008 0.017 0.040 0.102 0.446

Panel B: Rumour-only group CARi(-42,-1) 187 0.029 0.229 -0.751 -0.073 0.027 0.142 0.777 CARi(-21,-1) 187 0.030 0.168 -0.680 -0.059 0.019 0.098 0.749 Altman Z 127 4.388 7.910 -3.505 1.467 2.659 4.694 83.78 Zmijewski Probability

134 0.284 0.319 0.0001 0.017 0.160 0.472 1.00

Leverage 156 0.345 0.285 0 0.094 0.281 0.533 0.979 Firm size 175 7.965 1.833 2.165 6.830 8.093 9.272 12.536 Sales growth 158 0.066 0.134 0.0003 0.0144 0.036 0.077 1.492

41

Table 4 OLS regression of cumulative abnormal return against firm characteristics.

The dependent variables in Panels A and in panel B are CAR of a target from day -42 to day -1 and day -21 to day -1, respectively. There are four models and each model comprises two sets of results reporting the regressions of the rumor-announced firms and rumor-only firms, respectively. Model 1 and Model 2 are simple OLS regressions for the relationship between CARi and Distressi variables using Altman’s Z-score and Zmijewski probability for Distressi, respectively, while Model 3 and Model 4 include all control variables in the regression. Control variables also include four industry dummies: manufacturing (SIC codes 30-39), communication (SIC codes 48 and 49), finance (SIC codes 60-99) and service (SIC codes 72-82). Heteroskedasticity-consistent standard errors are reported in parentheses. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively. Model 1 Model 2 Model 3 Model 4

Rumor-

announcedRumor-

only Rumor-

announcedRumor-

only Rumor-

announcedRumor-only

Rumor-announced

Rumor-only

Panel A: Dependent variable is CARi(-42,-1) Altman Z-score

-0.0014*** (0.0004)

0.0586*** (0.0013)

-0.0004 (0.0008)

0.072*** (0.0016)

Zmijewski Probability

0.364** (0.166)

-0.141** (0.066)

0.347** (0.165)

-0.23** (0.094)

Leverage 0.152

(0.184) 0.087

(0.117) -0.073 (0.166)

0.198 (0.136)

Firm Size -0.037* (0.019)

-0.0068 (0.017)

-0.025 (0.016)

-0.008 (0.016)

Sales growth

-0.105 (0.305)

-0.197 (0.339)

-0.228 (0.292)

-0.064 (0.092)

Constant 0.0994** (0.0391)

-0.007 (0.023)

-0.018 (0.050)

0.061** (0.024)

0.38** (0.17)

0.064 (0.175)

0.249* (0.145)

0.125 (0.159)

Industry dummy

No No No No Yes Yes Yes Yes

F statistics 11.73*** 19.86*** 4.77*** 4.60*** 5.56*** 3.58** 2.00* 1.19 R square 0.027 0.037 0.194 0.036 0.23 0.068 0.347 0.078 Number of firms

46 127 46 134 43 117 43 124

Panel B: Dependent variable is CARi(-21,-1) Altman Z-score

-0.0014*** (0.0004)

0.0044*** (0.0011)

-0.0005 (0.0005)

0.0056*** (0.0009)

Zmijewski Probability

0.291** (0.152)

-0.035 (0.054)

0.237* (0.126)

-0.042 (0.056)

Leverage 0.197

(0.159) -0.044 (0.087)

0.047 (0.155)

-0.055 (0.089)

Firm Size -0.051** (0.022)

-0.032* (0.173)

-0.043** (0.017)

-0.033* (0.017)

Sales growth

-0.096 (0.217)

-0.271 (0.212)

-0.217 (0.210)

-0.052 (0.083)

Constant 0.126*** (0.035)

0.005 (0.016)

0.027 (0.035)

0.036* (0.018)

0.045** (0.213)

0.290* (0.161)

0.374** (0.164)

0.327** (0.157)

Industry dummy

No No No No Yes Yes Yes Yes

F statistics 8.98*** 16.39*** 3.64* 0.43 10.21*** 7.55*** 3.31*** 1.13 R square 0.030 0.044 0.168 0.005 0.416 0.1671 0.477 0.106 Number of firms

47 127 47 134 44 117 44 124

42

Table 5 Investment outcomes of buying and holding the selected target stocks.

This table reports the outcomes of investment in the selected target stocks whose CARs in day -42 to day -1 or in day -21 to day -1 are greater than (i.e., winner stocks) or smaller than (i.e., loser stocks) a particular threshold. The outcomes of investing in all rumored targets are also reported in the end of each panel. In Panel A, stocks are selected from a sample of rumor-announced firms and rumor-only firms which have data for at least 252 trading days after rumor publication. Stocks in Panel B are selected from the sample of all rumor-announced firms and rumor-only firms. Selected stocks are bought at the closing price on the rumor day and then are sold in the open market on the day when the first takeover bid is announced or 252 trading days after the rumor day, whichever comes first. The column of Excess return shows the annualized excess returns of the investment profiles with their standard deviations in parentheses. The significance levels of excess returns are reported by statistics of T-test and Wilcoxon signed-rank test. The column of Average variance documents the average of the annualized variances of daily returns of invested stocks. Standard deviations are reported in parentheses. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively.

Investment strategy Observation Excess return

T-statistic Wilcoxon

signed-rank Average variance

Investment strategy Observation Excess return

T-statistic Wilcoxon

signed-rank Average variance

Panel A: Stocks are selected from a sample of all rumor-announced firms and rumor-only firms with one year return data

CARi(-42,-1)>0% 134 0.722

(2.475) 3.375*** 3.699*** 0.019 CARi(-21,-1)>0% 141

0.654 (2.822)

2.754*** 2.016** 0.026

CARi(-42,-1)<0% 89 0.291

(3.121) 0.881 -1.139 0.039 CARi(-21,-1)<0% 83

0.360 (2.631)

1.248 0.631 0.029

CARi(-42,-1)>2% 121 0.738

(2.545) 3.190*** 3.731*** 0.018 CARi(-21,-1)>2% 120

0.657 (2.692)

2.673*** 2.344** 0.023

CARi(-42,-1)<2% 102 0.327

(2.977) 1.109 -0.960 0.038 CARi(-21,-1)<2% 104

0.417 (2.824)

1.505 0.269 0.032

CARi(-42,-1)>4% 106 0.786

(2.665) 3.035*** 3.604*** 0.018 CARi(-21,-1)>4% 102

0.696 (2.887)

2.435** 1.995** 0.026

CARi(-42,-1)<4% 117 0.336

(2.823) 1.289 -0.695 0.036 CARi(-21,-1)<4% 122

0.419 (2.636)

1.758* 0.758 0.028

CARi(-42,-1)>6% 92 0.869 (2.82)

2.957*** 3.450*** 0.018 CARi(-21,-1)>6% 91 0.761

(3.037) 2.392** 2.058** 0.028

CARi(-42,-1)<6% 131 0.326

(2.692) 1.384 -0.349 0.034 CARi(-21,-1)<6% 133

0.398 (2.537)

1.807* 0.713 0.026

CARi(-42,-1)>8% 83 1.007

(2.909) 3.152*** 3.918*** 0.020 CARi(-21,-1)>8% 77

1.027 (3.177)

2.837*** 2.729*** 0.028

CARi(-42,-1)<8% 140 0.279

(2.628) 1.257 -0.599 0.032 CARi(-21,-1)<8% 147

0.293 (2.473)

1.438 0.257 0.026

CARi(-42,-1)>10% 74 1.053

(3.061) 2.960*** 3.521*** 0.022 CARi(-21,-1)>10% 58

0.893 (3.045)

2.234** 2.056** 0.028

43

CARi(-42,-1)<10% 149 0.300

(2.560) 1.431 -0.134 0.030 CARi(-21,-1)<10% 166

0.424 (2.639)

2.070** 0.928 0.027

CARi(-42,-1)>12% 68 1.140

(3.147) 2.987*** 3.691*** 0.022 CARi(-21,-1)>12% 47

1.124 (3.283)

2.346** 2.381** 0.033

CARi(-42,-1)<12% 155 0.291

(2.528) 1.434 -0.123 0.030 CARi(-21,-1)<12% 177

0.392 (2.580)

2.021** 0.887 0.025

All rumored firms 223 0.550

(2.752) 2.984*** 2.159** 0.027 All rumored firms 224

0.545 (2.751)

2.968*** 2.030** 0.027

Panel B: Stocks are selected from a sample of all rumor-announced firms and rumor-only firms

CARi(-42,-1)>0% 158 0.736

(2.913) 3.177*** 3.839*** 0.0363 CARi(-21,-1)>0% 161

0.775 (3.124)

3.149*** 2.349** 0.036

CARi(-42,-1)<0% 101 -0.073 (4.439)

-0.166 -1.001 0.117 CARi(-21,-1)<0% 99 -0.160 (4.262)

-0.373 0.223 0.118

CARi(-42,-1)>2% 141 0.718

(2.982) 2.857*** 3.757*** 0.0348 CARi(-21,-1)>2% 139

0.801 (3.079)

3.069*** 2.670*** 0.035

CARi(-42,-1)<2% 118 0.065

(4.207) 0.169 -0.657 0.107 CARi(-21,-1)<2% 121

-0.019 (4.127)

0.052 -0.053 0.105

CARi(-42,-1)>4% 126 0.755

(3.117) 2.721*** 3.617*** 0.037 CARi(-21,-1)>4% 119

0.873 (3.298)

2.888*** 2.445*** 0.040

CARi(-42,-1)<4% 133 0.103

(3.099) 0.298 -0.423 0.097 CARi(-21,-1)<4% 141

0.036 (3.843)

0.112 0.361 0.091

CARi(-42,-1)>6% 111 0.823

(3.293) 2.634*** 3.381*** 0.040 CARi(-21,-1)>6% 107

0.954 (3.46)

2.851*** 2.493*** 0.044

CARi(-42,-1)<6% 148 0.118

(3.797) 0.379 -0.034 0.089 CARi(-21,-1)<6% 153

0.046 (3.695)

0.153 0.374 0.084

CARi(-42,-1)>8% 100 0.967

(3.419) 2.829*** 4.071*** 0.044 CARi(-21,-1)>8% 92

1.218 (3.624)

3.224*** 3.177*** 0.047

CARi(-42,-1)<8% 159 0.077

(3.678) 0.263 -0.456 0.083 CARi(-21,-1)<8% 168

-0.018 (3.555)

-0.065 -0.078 0.079

CARi(-42,-1)>10% 90 1.005

(3.591) 2.654*** 3.619*** 0.048 CARi(-21,-1)>10% 73

1.162 (3.651)

2.718*** 2.631*** 0.052

CARi(-42,-1)<10% 169 0.109

(3.576) 0.398 0.057 0.078 CARi(-21,-1)<10% 187

0.130 (3.577)

0.496 0.575 0.074

CARi(-42,-1)>12% 84 1.072

(3.686) 2.664*** 3.769*** 0.050 CARi(-21,-1)>12% 60

1.434 (3.937)

2.821*** 3.044*** 0.061

CARi(-42,-1)<12% 175 0.108

(3.525) 0.405 0.061 0.076 CARi(-21,-1)<12% 200

0.115 (3.473)

0.468 0.505 0.069

All rumored firms 259 0.421

(3.600) 1.88* 2.355** 0.067 All rumored firms 260

0.419 (3.621)

1.868* 2.030** 0.067

44

Table 6 Descriptive statistics of pre-runup, post-runup, runup, markup and takeover premium. Pre-runupi is equal to the CAR of target i’s stock from day -42 to day -1 (or day -21 to day -1). Post-runupi is target i’s CAR from the day of the first rumor publication through the day before the first bid announcement. Runupi is the sum of Pre-runupi and Post-runupi. Markupi is the CAR from the day of the first bid announcement through delisting or 126 trading days, whichever comes first. Premiumi is the sum of Runupi and Markupi. Panels A and B report the descriptive statistics of the rumor-announced sample and the successful sample, respectively, using an estimation window of (-242, -43) to estimate excess returns and using CARi(-42, -1) for Pre-runupi. Panels C and D report the same items but using an estimation window of (-221, -22) to estimate excess returns and using CARi(-21, -1) for Pre-runupi. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively.

Mean

Standard deviation

Min 25% Median 75% Max

Panel A: rumor-announced sample with (-242, -43) estimation window Pre-runupi 0.0764*** 0.226 -0.698 -0.043 0.046 0.196 1.115 Post-runupi 0.036 0.292 -1.438 -0.029 0.018 0.124 0.932 Runupi 0.113** 0.413 -2.255 -0.107 0.068 0.357 1.308 Markupi -0.0089 0.329 -1.003 -0.159 -0.0008 0.131 1.081 Premiumi 0.104 0.634 -2.255 -0.107 0.068 0.357 2.017

Panel B: successful sample with (-242, -43) estimation window Pre-runupi 0.0795** 0.234 -0.698 -0.046 0.049 0.203 1.115 Post-runupi 0.022 0.284 -1.438 -0.024 0.017 0.115 0.659 Runupi 0.101* 0.413 -1.469 -0.031 0.092 0.294 1.308 Markupi -0.034 0.279 -1.004 -0.129 -0.009 0.096 0.433 Premiumi 0.068 0.603 -2.255 -0.079 0.062 0.348 1.741

Panel C: rumor-announced sample with (-221, -22) estimation window Pre-runupi 0.090*** 0.190 -0.24 -0.006 0.057 0.157 1.069 Post-runupi 0.035 0.272 -1.39 -0.187 0.024 0.123 0.856 Runupi 0.125*** 0.343 -1.426 -0.006 0.101 0.255 1.262 Markupi -0.0009 0.321 -0.098 -0.13 0.013 0.127 1.121 Premiumi 0.124* 0.560 -1.801 -0.014 0.111 0.288 1.689

Panel D: successful sample with (-221, -22) estimation window Pre-runupi 0.0908*** 0.199 -0.24 -0.007 0.045 0.164 1.069 Post-runupi 0.022 0.264 -1.39 -0.018 0.021 0.11 0.659 Runupi 0.113** 0.346 -1.426 -0.009 0.098 0.259 1.262 Markupi -0.023 0.272 -0.987 -0.127 0.010 0.116 0.451 Premiumi 0.090 0.533 -1.801 -0.012 0.097 0.265 1.689

45

Table 7 OLS regression of takeover premium against pre-runup and post-runup. The table reports the results of the following regression, Premiumi = a + b1 Pre-runupi + b2 Post-runupi + ui. Dependent variable Premiumi is equal to the sum of Pre-runupi, Post-runupi and Markupi. Pre-runupi is measured by the CAR of target i’s stock from day -42 to day -1 in Panel A and from day -21 to day -1 in Panel B. Post-runupi is the CAR from the day of the first rumor date through the day before the first bid announcement. Markupi is the CAR from the day of the first bid announcement through delisting or 126 trading days, whichever comes first. Each panel includes a sample of all rumor-announced targets and a sample of targets being successfully taken over. Heteroskedasticity-consistent standard errors are reported in parentheses. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively.

Sample Sample

size Pre-runup

T-statistic for b1 ≤ 1

Post-runup

T-statistic for b2 ≤ 1

Intercept R2

Panel A: Pre-runupi = CARi(-42, -1) Rumor-announced

72 1.635*** (0.197)

3.217*** 1.198*** (0.149)

1.326 -0.083 (0.051)

0.5729

Successful 60 1.599*** (0.178)

3.359*** 1.148*** (0.113)

1.315 -0.043 (0.044)

0.7464

Panel B: Pre-runupi = CARi(-21, -1) Rumor-announced

72 1.459*** (0.397)

1.156 1.180*** (0.205)

0.88 -0.068 (0.056)

0.4461

Successful 60 1.760*** (0.190)

3.99*** 1.251*** (0.116)

2.156** -0.061 (0.046)

0.7126

46

Table 8 OLS regression of takeover premium against runup.

This table reports the results of the following regression, Premiumi = a + b Runupi + ui.

Dependent variable Premiumi is equal to the sum of Runupi and Markupi. In Panels A and B, event day (day 0) is the day when the rumor is published first time and Runupi is the sum of Pre-runupi and Post-runupi, where Pre-runupi is the CAR of target i’s stock from day -42 to day -1 in Panel A and from day -21 to day -1 in Panel B, and Post-runupi is the CAR from the event day through the day before the first bid announcement. In panel C, day 0 is the day when the first bid is announced and Runupi is measured by the CAR of target i’s stock from day -42 to day -1. Markupi in all three panels is the CAR from the day of the first bid announcement through delisting or 126 trading days, whichever comes first. Heteroskedasticity-consistent standard errors are reported in parentheses. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively.

Sample Sample size Runup T-statistic for b ≤ 1

Intercept R2

Panel A: Pre-runupi = CARi(-42,-1) with the rumor publication day as the event day Rumor-announced

72 1.387*** (0.139)

2.779*** -0.081 (0.054)

0.5590

Successful 60 1.346*** (0.128)

2.71*** -0.036 (0.046)

0.7264

Panel B: Pre-runupi = CARi(-21,-1) with the rumor publication day as the event day Rumor-announced

72 1.286*** (0.249)

1.148 -0.061 (0.053)

0.4398

Successful 60 1.440*** (0.149)

2.964*** -0.043 (0.049)

0.6844

Panel C: Runupi = CARi(-42,-1) with the first takeover announcement day as the event dayRumor-announced

70 1.021** (0.393)

0.053 0.009

(0.067) 0.2540

Successful 58 1.466*** (0.271)

1.72* -0.023 (0.063)

0.5311

47

Table 9 Descriptive statistics of the runup, markup and takeover premium when event day is the first bid announcement day. Event day is the day when the first bid is announced. Estimation window extends from day -379 to day -127. Runupi is equal to the CAR of target i’s stock from day -42 to day -1. Markupi is the CAR from the event day through delisting or 126 trading days, whichever comes first. Premiumi is the sum of Runupi and Markupi. Panels A and B report the descriptive statistics of the rumor-announced sample and the successful sample, respectively. Symbols ***, ** and * indicate significance at the 0.01, 0.05 and 0.1 levels, respectively.

Mean Standard deviation

Min 25% Median 75% Max

Panel A: Rumor-announced sample (N = 70) Runup 0.163*** 0.229 -0.316 0.035 0.107 0.255 1.054 Markup 0.012 0.401 -1.591 -0.098 -0.020 0.227 0.828 Premium 0.175*** 0.464 -1.401 -0.004 0.173 0.393 1.353

Panel B: Successful sample (N = 58) Runup 0.150*** 0.216 -0.316 0.035 0.106 0.255 1.054 Markup 0.045 0.314 -1.258 -0.086 -0.014 0.227 0.828 Premium 0.197*** 0.434 -1.401 0.010 0.203 0.419 1.065