Post on 13-Jan-2015
description
Christian Funke
Selected Essays in Empirical Asset Pricing
GABLER EDITION WISSENSCHAFT
EBS ForschungSchriftenreihe der European Business School (EBS)International University · Schloss ReichartshausenHerausgegeben vonUniv.-Prof. Ansgar Richter, PhD
Band 69
Die European Business School (EBS) – gegründet im Jahr 1971– ist Deutschlands älteste private Wissenschaftliche Hoch-schule für Betriebswirtschaftslehre im Universitätsrang. DieserVorreiterrolle fühlen sich ihre Professoren und Doktoranden in Forschung und Lehre verpflichtet. Mit der Schriftenreihe präsentiert die European Business School (EBS) ausgewählteErgebnisse ihrer betriebs- und volkswirtschaftlichen Forschung.
Christian Funke
Selected Essays in Empirical Asset PricingInformation Incorporationat the Single-Firm, Industry,and Cross-Industry Level
With a foreword by Prof. Dr. Lutz Johanning
GABLER EDITION WISSENSCHAFT
Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.
Dissertation European Business School (EBS) Oestrich-Winkel, 2008
D1540
.
.
1. Aulage Dezember 19971st Edition 2008
All rights reserved© Gabler | GWV Fachverlage GmbH, Wiesbaden 2008
Editorial Office: Frauke Schindler / Britta Göhrisch-Radmacher
Gabler is part of the specialist publishing group Springer Science+Business Media.www.gabler.de
No part of this publication may be reproduced, stored in a retrieval systemor transmitted, mechanical, photocopying or otherwise without priorpermission of the copyright holder.
Registered and/or industrial names, trade names, trade descriptions etc. cited in this publica-tion are part of the law for trade-mark protection and may not be used free in any form or byany means even if this is not specifically marked.
Cover design: Regine Zimmer, Dipl.-Designerin, Frankfurt/MainPrinted on acid-free paperPrinted in Germany
ISBN 978-3-8349-1142-1
V
Foreword
Research in empirical asset pricing has – fostered by the availability of new databases –
become an important field of research within the last three decades. This kind of re-
search contributes to the ongoing and exciting debate between the neoclassical and the
behavioral explanation of asset pricing and can help to better explain the evolvement of
asset prices in capital markets.
Research in empirical asset pricing requires multiple competences: a sound un-
derstanding of capital markets, market designs, trading processes, and asset pricing
models, a superior handling of large databases, and efficient programming skills. Chris-
tian Funke lives up to this challenge and his doctoral thesis comprises of three important
essays in empirical asset pricing.
In the first essay, Christian investigates the long term performance of rival com-
panies related to acquisition targets. He documents an underreaction of capital markets
to the information contained in M&A announcements. Following large rival gain events
due to positive information signaling and large rival loss events due to the negative
competitive effects of the transaction, he observes a return drift for up to 12 months
after the announcement.
The second essay documents a strong and prevalent drift in long-term industry
returns after M&A announcements. Specifically, industries that experience positive av-
erage announcement reactions continue to do well in the future, while industries that
experience negative average announcement reactions continue to do poorly. The evi-
dence suggests that capital markets underreact to the industry-wide information pro-
vided by merger announcements.
In the third essay, Christian documents return predictability across stocks, spe-
cifically from customers to their suppliers. He shows that for large positive (negative)
customer price change events supplier stock prices experience significantly positive
(negative) cumulative abnormal returns for up to 20 days after the event. However, the
VI
major part of these returns arises in the first five days and he cannot find such return
predictability for the largest stocks and the most recent time period, indicating that capi-
tal markets are relatively efficient in incorporating extreme customer return shocks into
supplier stock prices.
I wish that researchers and other capital markets professionals will acknowledge
Christians Funke's contribution to the ongoing debate on asset pricing.
Johannisberg, April 2008 Prof. Dr. Lutz Johanning
VII
Preface
“We don't receive wisdom; we must discover it for ourselves after a journey that
no one can take for us or spare us.” Marcel Proust (1871-1922)
Writing a doctoral thesis is never an easy or straightforward endeavor. I would not have
been able to complete this work without the help of many people whom I would like to
acknowledge here.
First, I would like to express my sincere gratitude to my doctoral supervisor Lutz
Johanning for his guidance and encouragement throughout the last three years. I owe
him much for the outstanding opportunities he made possible during my time as Re-
search Assistant at the Dean’s Office of Full-time Programmes and the Endowed Chair
of Asset Management at European Business School (EBS). Especially valuable was the
experience I gained by teaching lectures and case studies on risk management, asset
management, investments, and statistics at EBS and the opportunity to spend a semester
abroad in the PhD program of the Katz Graduate School of Business, University of
Pittsburgh. With regard to my semester abroad, I gratefully acknowledge financial sup-
port provided by the German Academic Exchange Service (DAAD) and the EBS De-
partment of Finance (Stiftung Unternehmensfinanzierung und Kapitalmärkte für den
Finanzstandort Deutschland).
Second, I would like to thank Dirk Schiereck for readily agreeing to provide the
second opinion on my thesis and providing helpful and constructive feedback during my
annual proposal defenses. Also, I would like to appreciate the valuable comments and
insights provided by Yakov Amihud from the Stern School of Business, New York
University, during his research visits to EBS.
Third, I am thankful for the exceptional support and help of my colleague and
friend Timo Gebken. He not only provided me with a vision for innovative empirical
research, a sounding board for research ideas, and a lot of valuable academic advice, but
VIII
also became a good friend during my time at the Chair. In addition, my special appre-
ciation goes to my colleagues and friends Gaston Michel and Sebastian Werner, who
made my time at the Chair a memorable experience, both during and outside office
hours. I would like to thank them and Felix Zeidler for valuable comments and insights
which substantially enhanced this dissertation.
Fourth, I would like to thank all of my colleagues at the Endowed Chair of Asset
Management and the Endowed Chair of Banking and Finance for the fun times we had
on and off work. Also, I would like to thank my fellow doctoral students who formed
the “Underdocs United” soccer team together with me and brought it to some success in
Monday’s Soccerworld Mainz indoor league. The same appreciation goes to all my
friends and colleagues at EBS who joined and supported my newly founded soccer club
FC Altkönig e.V. Finally, I would like to thank Tobias Weigl for being an uncompli-
cated room mate during our PhD days (and nights) in Oestrich-Winkel, and Gerhard
Trautmann for our regular lunch meetings to talk about anything and everything.
Last, but definitely not least, I would like to thank my family and my fiancée
Julie. My parents and my brother always encouraged me and provided me with the free-
dom to pursue my goals. Without their continuous support, love, and patience I could
not have completed this dissertation. Finally, I am especially grateful to Julie, who not
only supported me when I worked long hours and weekends on this thesis, but also
made the tremendous effort to find and correct every single one of my English language
mistakes. Of course, any remaining errors are mine. Her support and love gave me the
necessary strength and motivation to finish this doctoral thesis. I dedicate this disserta-
tion to her and to my family, especially to my grandmother Gisela and to my grandfa-
ther Ralf, who sadly passed away before I could present this thesis to him.
Kronberg im Taunus, January 2008 Christian Funke
IX
Overview
List of Tables and Figures........................................................................................... XIII List of Abbreviations ....................................................................................................XV List of Symbols ..........................................................................................................XVII
1 Introduction........................................................................................................... 11.1 Overview and General Research Objective ................................................... 1 1.2 Essay 1: Research Question and Main Findings ............................................ 3 1.3 Essay 2: Research Question and Main Findings ............................................ 5 1.4 Essay 3: Research Question and Main Findings ............................................ 6
2 Information Signaling and Competitive Effects of M&A: Long-Term Performance of Rival Companies ........................................................................ 92.1 Introduction.................................................................................................. 102.2 Related Literature......................................................................................... 132.3 Data and Methodology................................................................................. 152.4 Results .......................................................................................................... 242.5 Conclusion ................................................................................................... 36Appendix ............................................................................................................... 38
3 Predictability of Industry Returns After M&A Announcements................... 393.1 Introduction.................................................................................................. 403.2 Data and Methodology................................................................................. 433.3 Results .......................................................................................................... 483.4 Conclusion ................................................................................................... 65
4 Predictability of Supplier Returns After Large Customer Price Changes.... 674.1 Introduction.................................................................................................. 684.2 Related Literature......................................................................................... 714.3 Data and Methodology................................................................................. 754.4 Results .......................................................................................................... 834.5 Conclusion ................................................................................................... 96
5 Conclusion............................................................................................................ 99
References ................................................................................................................... 101
XI
Table of Contents
List of Tables and Figures........................................................................................... XIII List of Abbreviations ....................................................................................................XV List of Symbols ..........................................................................................................XVII
1 Introduction........................................................................................................... 11.1 Overview and General Research Objective ................................................... 1 1.2 Essay 1: Research Question and Main Findings ............................................ 3 1.3 Essay 2: Research Question and Main Findings ............................................ 5 1.4 Essay 3: Research Question and Main Findings ............................................ 6
2 Information Signaling and Competitive Effects of M&A: Long-Term Performance of Rival Companies ........................................................................ 92.1 Introduction.................................................................................................. 102.2 Related Literature......................................................................................... 132.3 Data and Methodology................................................................................. 15
2.3.1 Data .................................................................................................. 15 2.3.2 Event Identification.......................................................................... 16 2.3.3 Long-Term Performance.................................................................. 21
2.4 Results .......................................................................................................... 24 2.4.1 Buy-and-Hold Abnormal Returns .................................................... 242.4.2 Calendar-Time Portfolios................................................................. 272.4.3 Size Groups ...................................................................................... 312.4.4 Cross-Sectional Regressions ............................................................ 33
2.5 Conclusion ................................................................................................... 36 Appendix ............................................................................................................... 38
3 Predictability of Industry Returns After M&A Announcements................... 393.1 Introduction.................................................................................................. 403.2 Data and Methodology................................................................................. 43
3.2.1 Daily Announcement Effects ........................................................... 433.2.2 Monthly Returns .............................................................................. 46
XII
3.3 Results .......................................................................................................... 483.3.1 Industry Announcement Effects....................................................... 48
3.3.1.1 Persistence in Intra-Industry Effects ................................. 48 3.3.1.2 Predictability of Monthly Industry Returns ...................... 50
3.3.2 Industry Portfolio Investment Strategies.......................................... 52 3.3.2.1 Raw and Adjusted Returns................................................ 523.3.2.2 Subperiod Analysis ........................................................... 553.3.2.3 Size Groups ....................................................................... 563.3.2.4 Liquidity Effects ............................................................... 583.3.2.5 Strength of Intra-Industry Effects ..................................... 593.3.2.6 Three-Factor Risks ............................................................ 61
3.3.3 The Cross-Section of Individual Stock Returns............................... 63 3.4 Conclusion ................................................................................................... 65
4 Predictability of Supplier Returns After Large Customer Price Changes.... 674.1 Introduction.................................................................................................. 68 4.2 Related Literature......................................................................................... 71 4.3 Data and Methodology................................................................................. 75
4.3.1 Customer-Supplier Relationships .................................................. 75 4.3.2 Large Abnormal Customer Price Changes..................................... 76 4.3.3 Post-Event Abnormal Supplier Returns ......................................... 79
4.4 Results .......................................................................................................... 83 4.4.1 Large Abnormal Customer Price Changes..................................... 834.4.2 Contemporaneous Supplier Reactions ........................................... 854.4.3 Within- and Cross-Industry Effects ............................................... 864.4.4 Size Groups .................................................................................... 884.4.5 Time-Period Subsamples ............................................................... 904.4.6 Robustness Tests ............................................................................ 914.4.7 Cumulative Abnormal Return Regressions.................................... 93
4.5 Conclusion ................................................................................................... 96
5 Conclusion............................................................................................................ 99
References ................................................................................................................... 101
XIII
List of Tables and Figures
Table 2.1: Annual Distribution of M&A Transactions and Average Rival Reactions19
Table 2.2: Annual Distribution of Rivals with Large Short-Term Reactions .............21
Table 2.3: Long-Term Rival Buy-and-Hold-Abnormal Returns.................................25
Table 2.4: Calendar-Time Portfolio Returns...............................................................28
Table 2.5: Calendar-Time Portfolio Three-Factor Alphas ..........................................30
Table 2.6: Long-Term Buy-and-Hold Abnormal Returns for Rival Size Groups.......32
Table 2.7: Cross-Sectional Buy-and-Hold Abnormal Return Regressions.................34
Table 3.1: Annual Distribution of M&A Transactions and Average Industry Announcement Effects ...............................................................................45
Table 3.2: Industry Summary Statistics ......................................................................47
Table 3.3: Industry Cumulative Abnormal Returns to M&A Announcements ..........49
Table 3.4: Monthly Industry Size and Book-to-Market Adjusted Returns .................51
Table 3.5: Monthly Industry Portfolio Returns...........................................................54
Table 3.6: Monthly Industry Portfolio Returns for Time-Period Subsamples ............55
Table 3.7: Monthly Industry Portfolio Returns for Size-Quintiles .............................57
Table 3.8: Robustness of Monthly Industry Portfolio Returns ...................................59
Table 3.9: Industry Portfolio Three-Factor Alphas .....................................................62
Table 3.10: Fama/MacBeth Regressions for Individual Stock Returns ........................65
Table 4.1: Summary Statistics.....................................................................................78
Table 4.2: Annual Descriptive Statistics .....................................................................82
Table 4.3: Abnormal Returns to Suppliers after Customer Return Shocks ................84
Table 4.4: Abnormal Returns to Suppliers Conditional on Initial Reaction ...............86
Table 4.5: Abnormal Returns to Suppliers Within and Across Industries ..................87
Table 4.6: Abnormal Returns to Suppliers for Different Size Groups........................89
Table 4.7: Abnormal Returns to Suppliers for Different Sample Periods...................91
Table 4.8: Robustness Tests ........................................................................................92
Table 4.9: Cross-Sectional Cumulative Abnormal Return Regressions .....................94
XIV
Figure 2.1: Historical Distribution of Rival Cumulative Abnormal Returns .............. 20
Figure 2.2: Event-Time Buy-and-Hold Abnormal Rival Returns ............................... 26
XV
List of Abbreviations
ADR American Depository Receipt
AR Abnormal Return
BHAR Buy-and-Hold Abnormal Return
B/M Book-to-Market
CAPM Capital Asset Pricing Model
CAR Cumulative Abnormal Return
CRSP Center for Research in Securities Prices
FASB Financial Accounting Standards Board
GVKEY Compustat Unique Firm Identifier
M&A Mergers and Acquisitions
N Number of Observations
NYSE New York Stock Exchange
PERMNO CRSP Unique Security Identifier
PPE Property, Plant, and Equipment
REIT Real Estate Investment Trust
SBI Shares of Beneficial Interest
S.D. Standard Deviation
SDC Securities Data Company
SFAS Statements of Financial Accounting Standards
SIC Standard Industrial Classification
XVII
List of Symbols
pa Time-series regression intercept for portfolio p
tiAR , Abormal return of firm i on day t
pb Time-series regression factor loading on market factor for portfolio p
],[, TtiBHAR Buy and hold abnormal return for firm i from point in time t until T
]20,1[,iBHAR Buy and hold abnormal return for firm i and 20-day post-event window
]20,1[,iCAR Cumulative abnormal return for firm i and 20-day post-event window
]2,2[CAR Cumulative abnormal return over 5-day event window
1,tiCAR Average 5-day cumulative abnormal return of equal-weighted industry
portfolio i in the prior month t-1
tjCUSTAR , Abnormal return of customer j on event day t
tp, Time-series regression error-term for portfolio p in month t
ti , Cross-sectional regression error-term for event i at point in time t
ph Time-series regression factor loading on HML for portfolio p
HML Fama and French (1993) value factor (high minus low)
tfR , Risk-free return over time interval t
tiR , Return of firm i over time interval t
tjR , Return of control firm j over time interval t
tmR . Return of equal-weighted CRSP market index over time interval t
tsR , Return of size quintile s over time interval t
tpR , Return of event portfolio p at month t after portfolio formation 2R Regression coefficient of determination (R squared)
ps Time-series regression factor loading on SMB for portfolio p
SMB Fama and French (1993) size factor (small minus big)
tiSUPPAR , Abnormal return of supplier i on event day t
tiSUPPSIZE , Natural logarithm of supplier i market value of equity on event day t
tjiWITHININD ,, Dummy equal to one if supplier i and customer j in the same industry
1
1 Introduction
1.1 Overview and General Research Objective
This doctoral thesis comprises three essays that address selected issues in empirical as-
set pricing. All three essays conduct original empirical research using U.S. capital mar-
ket data. The first two essays both investigate the importance of mergers and acquisi-
tions (M&A) for stock prices. The first essay (Chapter 2) focuses on the single-firm
level and investigates the long-term performance of rival firms directly affected by
M&A transactions. Following a similar intuition, the second essay (Chapter 3) takes an
industry perspective and examines the long-term performance of industry portfolios
depending on the impact of M&A on the firms in the industry. Finally, the third essay
(Chapter 4) focuses on the cross-industry level as it follows explicit economic links be-
tween firms across industries and investigates the information transmission from cus-
tomer stock prices to supplier stock prices.
Although different in focus and approach, all three essays contribute to the main
challenge of developing a better understanding of a central asset pricing issue: how effi-
ciently do capital markets incorporate new information into the stock price discovery
process. Are capital markets efficient or can we empirically observe violations of mar-
ket efficiency in the information incorporation process?
This question is a central issue in the asset pricing literature. In general, accord-
ing to Campbell (2000) (p. 1516), “…asset pricing is concerned with the sources of risk
and the economic forces that determine the rewards for bearing risk”. I focus on specific
research questions related to a particular area in asset pricing, i.e., the cross-sectional
structure of stock returns. This topic has generated a lot of research interest given the
discovery of several apparent ‘anomalies’ in the cross-section of stock returns, which
cannot be explained by the Capital Asset Pricing Model (CAPM) of Sharpe (1964),
Lintner (1965), and Mossin (1966). The most prominent cross-sectional anomalies are
the small-firm effect (Banz (1981)), the value effect (Basu (1983), Rosenberg, Reid and
2
Lanstein (1985), Fama and French (1992)), which is related to the contrarian effect (De
Bondt and Thaler (1985)), and finally, the momentum effect (Jegadeesh and Titman
(1993)). In his extensive review of the asset pricing literature, Campbell (2000) notes
several explanation for these anomalies (pp. 1527-1528):
The first and most conservative interpretation is that they are entirely spurious, the result of ‘data
snooping’ that has found accidental patterns in historical data. (…) A second view is that the anoma-
lies result from the inability of a broad stock index to proxy for the market portfolio return. (…) A
third view is that the anomalies provide genuine evidence against the CAPM but not against a
broader rational model in which there are multiple risk factors. (…) A fourth view is that the anoma-
lies do not reflect any type of risk but are ‘mistakes’ that disappear once market participants become
aware of them. (…) The most radical view is that the anomalies reflect enduring psychological biases
that lead investors to make irrational forecasts.
A lot of research has been undertaken in the last years to answer these issues, but
essentially the question of how to best price the cross-section of stock returns has re-
mained largely unresolved. In the last years the ‘radical view’ of behavioral finance has
generated the most attention. In his review of this field, Hirshleifer (2001) argues (p.
1534) “… that the central task of asset pricing is to examine how expected returns are
related to risk and to investor misevaluation”, a view which is opposed to the one repre-
sented by efficient markets proponents such as Fama (1998) and Campbell (2000). In
general, researchers in the behavioral finance field argue that there are a variety of psy-
chological reasons for the mispricing of assets. Therefore, the argument goes, equilib-
rium prices reflect the weighted average of rational and irrational traders, given the fact
that limits of arbitrage exist which prevent rational arbitrageurs from exploiting mispri-
cing to the full extent (Shleifer and Vishny (1997)). In addition, Hirshleifer (2001) notes
that (p. 1539):
(…a) possible reason for persistent mispricing is that some relevant pieces of public information are
ignored or misused by everyone. This can occur either because the signals are obscurely located or
because our shared model of the world is just not sophisticated enough to make their relevance clear.
This intuition provides the motivation and starting point for my research. In this
doctoral thesis, I focus on the information incorporation process for related firms, i.e.
firms only indirectly affected by the “pieces of public information” in question. This is
3
an important topic, as so far researchers have focused on investigating the information
incorporation process for firms directly affected by the information at hand. However,
major corporate events, such as M&A, also have important consequences for rival firms
and the whole industry. So far, mainly the short-term effects of these events for rival
companies have been investigated, i.e., their long-term consequences and the question
of market efficiency in the information incorporation process have been neglected, an
issue which I investigate in the first two essays of this doctoral thesis. In addition, re-
cently cross-industry links have generated increasing attention, as they appear to repre-
sent examples of relevant pieces of information which are often ignored. Therefore, I
complete this doctoral thesis with a third essay examining the importance of economic
links between customers and suppliers in the information incorporation process.
Overall, the research objective of this doctoral thesis is to improve our under-
standing of the efficiency of capital markets by providing new evidence on the informa-
tion incorporation process at the single-firm, industry, and cross-industry level.
1.2 Essay 1: Research Question and Main Findings
The first essay entitled “Information Signaling and Competitive Effects of M&A: Long-
Term Performance of Rival Companies” investigates the long-term performance of rival
firms related to acquisition targets.
The intra-industry effects of M&A announcements are based on two opposing
effects: positive information-signaling regarding future takeover activity and negative
competitive effects of the transaction. The empirical evidence based on short-term event
studies shows that rival firms gain at the M&A announcement. Hence, the positive in-
formation signaling effects of the deal outweigh the negative competitive effects. How-
ever, long-term studies regarding the transaction success of the merging firms show that
prices do not fully adjust to the information implied by M&A transactions. Given the
evidence on the long-term underperformance of acquiring firms, it is straightforward to
ask whether we can observe a similar underreaction of capital markets to the informa-
tion contained in M&A announcements for the case of rival companies. So far research-
4
ers have not investigated the long-term effects of M&A announcements on rivals.
Therefore, the objective of the first essay is to fill this research gap and provide empiri-
cal evidence on the long-term wealth effects of corporate acquisitions on directly af-
fected rival firms. This question is important to understand how capital markets incor-
porate the information contained in M&A announcements into stock prices.
Research Question 1:
What are the long-term wealth effects of corporate acquisitions on directly affected ri-
val firms? Are the positive information-signaling effects and negative competitive ef-
fects of M&A transactions efficiently incorporated into rival stock prices?
I examine this question using a sample of 2,511 deals announced between Janu-
ary 1, 1985 and December 31, 2005. Based on return correlations to the target and the
short-term reaction around the deal announcement, I identify rival firms clearly affected
by the transaction due to the positive information-signaling effects and the negative
competitive effects of the transaction. Overall, I investigate samples of 6,138 large rival
gain events and 5,408 large rival loss events.
I find that rivals with large gains at the announcement of a deal due to positive
information-signaling experience significantly positive buy-and-hold-abnormal returns
for a year after the announcement, while rivals suffering large losses due to the negative
competitive effects of a M&A transaction experience significantly negative returns over
the same horizon. My results are robust to the choice of methodology to calculate long-
term performance, alternative approaches such as the calendar-time portfolio method
using size and book-to-market matched control firms or a Fama and French (1993)
three-factor asset pricing model show similar findings. This result indicates that capital
markets underreact to the information contained in M&A transactions for the stock price
performance of rivals, a result which fits well into the literature on underreaction of
capital markets to new information (Frazzini (2006), Jackson and Johnson (2006),
Zhang (2006)).
5
1.3 Essay 2: Research Question and Main Findings
The second essay entitled “Predictability of Industry Returns After M&A Announce-
ments” investigates long-term industry returns after M&A announcements.
As mentioned above, overall, researchers find that rival firms gain in short-term
windows around M&A announcements. In a recent study, Funke et al. (2008) show that
these positive industry reactions are not pervasive over time. At the beginning of an
industry-wide consolidation process intra-industry effects are significantly positive.
However, due to increasing competition, intra-industry effects decrease steadily and
become negative at the end of such a consolidation process. Hence, time-varying com-
petitive effects of M&A activity, depending on time-varying industry structure, are re-
sponsible for a significant difference in intra-industry reactions over time. However, as
industry structure and, therefore, the competitive effects of M&A activity change rela-
tively slowly over time, their evidence suggests a certain persistence in average monthly
intra-industry announcement reactions. As short-horizon intra-industry effects directly
impact industry returns, such a persistence may also influence long-term industry per-
formance. To the author’s best knowledge, long-term industry returns after merger an-
nouncements have not yet been studied. Therefore, the research objective of the second
essay of this doctoral thesis is to investigate whether there are systematic differences in
industry returns after M&A announcements depending on the initial intra-industry ef-
fects of the deal.
Research Question 2:
Are there systematic differences in industry returns after M&A announcements depend-
ing on the initial intra-industry effects of the deal?
Using a dataset of 16,483 transactions that occur within one of 20 industries be-
tween 1985 and 2002, I investigate whether the average industry reaction in previous
periods affects the intra-industry effects of transaction announcements in later periods. I
find that this is indeed the case: the cumulative abnormal return (CAR) at the an-
nouncement of an M&A transaction depends on the average reaction to merger an-
nouncements in the same industry in the previous month. Based on this result, I look at
6
monthly industry returns after merger announcements. Paralleling the results for CARs,
I find that returns depend on the average industry announcement reaction in the previ-
ous month.
Therefore, the evidence indicates that the industry-wide information provided by
merger announcements does not seem to be incorporated into stock prices immediately.
Industry M&A investment strategies, which buy positively reacting industries and sell
negatively reacting industries, appear profitable even after controlling for size and book-
to-market effects in returns. In robustness tests, I find that profitability has strengthened
over time, that the effect seems to exist also for the largest stocks, and that an alternative
risk-adjustment using the Fama and French (1993) three-factor asset pricing model can-
not explain the returns of the industry portfolio investment strategies. In addition, Fama-
McBeth regressions show that the average industry CAR of the previous month is sig-
nificantly related to individual stock returns of firms in that industry.
1.4 Essay 3: Research Question and Main Findings
The third essay entitled “Predictability of Supplier Returns After Large Customer Price
Changes” investigates return predictability across stocks, specifically from customers to
their suppliers. Limited investor attention (investor inattention) and the consequences
for information processing in capital markets have recently generated increasing interest
in the asset pricing literature. In a new paper Cohen and Frazzini (2006) examine eco-
nomically linked firms and document widespread cross-asset return predictability in
U.S. capital markets. They conclude that investors do not take into account ex-ante
available and often longstanding customer-supplier relationships due to their limited
attention, so that prices of supplier firms have a predictable lag in updating to new in-
formation about its customer firms.
However, Cohen and Frazzini (2006) focus on return predictability using
monthly data and do not investigate short-run effects on the daily level. A short-term
event study approach provides an ideal ground to test limited investor attention: if we
can observe supplier return predictability after extreme attention-grabbing customer
7
price changes, then we can reasonably assume a violation of market efficiency due to
investor inattention to relevant, publicly available information. However, if we can find
no or limited evidence of supplier return predictability after such large customer price
changes, we can conclude that investor inattention to customer information is restricted
to the smaller price changes studied by Cohen and Frazzini (2006) at the monthly hori-
zon. Additionally, a short-term based investigation on how long it takes for customer
stock return information to be incorporated into supplier stock prices can provide valu-
able insights into the information incorporation process of capital markets.
Research Question 3:
3.1: Do we observe supplier return predictability as documented by Cohen and Frazz-
ini (2006) also on a daily basis in an event study framework using extreme atten-
tion-grabbing customer price changes?
3.2: How long does it take, exactly, for customer stock return information to be incor-
porated into supplier stock prices?
Using daily stock returns from 1981 to 2004, I investigate abnormal supplier
stock returns after large abnormal customer price changes, i.e., a daily customer abnor-
mal stock return three standard deviations or more from the mean. I show that for large
positive (negative) customer price change events supplier stock prices experience sig-
nificantly positive (negative) CARs for up to 20 days after the event. However, the ma-
jor part of these returns arises in the first five days after the event. In addition, I show
that if there is a large positive (negative) contemporaneous supplier reaction to a posi-
tive (negative) large customer price change, markets seem to have already incorporated
all information immediately and no post-event supplier drift can be discerned. However,
if there is no reaction or even a negative (positive) reaction to a positive (negative) large
customer price change, I document a drift in supplier stock prices in the same direction
as the customer event. Sometimes, investors seem to shift their limited attention not
immediately to this customer stock return information and, therefore, incorporate it only
slowly into supplier prices.
9
2 Information Signaling and Competitive Effects of M&A:
Long-Term Performance of Rival Companies
Abstract
In this essay, I investigate the long-term performance of rival companies related to ac-
quisition targets. Using a sample of 2,511 deals from 1985 to 2005, I document an un-
derreaction of capital markets to the information contained in M&A announcements.
Following 6,138 large rival gain events due to positive information signaling and 5,408
large rival loss events due to the negative competitive effects of the deal, I find a return
drift for up to 12 months after the announcement. Hence, my results indicate that capi-
tal markets do not immediately incorporate the effects of M&A announcements into ri-
val stock prices.
I thank my co-authors Timo Gebken and Lutz Johanning, as well as Gaston Michel, Julie Ann Ng, Sebastian Werner, and Felix Zeidler for valuable comments and insights. I am also grateful for the input of workshop participants at EBS and WHU. Any remaining errors are mine.
10
2.1 Introduction
Mergers and acquisitions are important corporate events. With regard to the wealth ef-
fects of M&A, evidence based on short-term event studies shows that target sharehold-
ers gain significantly from acquisitions, while acquiring firm shareholders lose or, at
most, break even around M&A announcements. However, long-term studies show that
prices do not fully adjust to the information implied by M&A transactions. Agrawal,
Jaffe and Mandelker (1992) document significantly negative long-term abnormal re-
turns to acquiring firms. These results are confirmed in later studies by Loughran and
Vijh (1997) and Rau and Vermaelen (1998).
Besides measuring the performance of the merging firms, researchers also exam-
ine the wealth effects of M&A on other firms from the same industry. The intra-industry
effects of M&A announcements are based on two opposing effects: positive information
signaling regarding future takeover activity and negative competitive effects of the
transaction. Overall, researchers find that rival firms gain at the M&A announcement.
Hence, the positive information signaling effects of the deal outweigh the negative
competitive effects (Eckbo (1983), Eckbo (1985), Song and Walkling (2000), Fee and
Thomas (2004) and Shahrur (2005)).
However, so far no studies exist that examine the long-term stock price perform-
ance of related rivals. Given the findings on significant abnormal long-term perform-
ance for acquiring firms, it is straightforward to ask whether we can observe a similar
underreaction of capital markets to the information contained in M&A announcements
for the stock prices of related firms. Therefore, the objective of this essay is to fill this
research gap and provide empirical evidence on the long-term wealth effects of corpo-
rate acquisitions on directly affected rival firms. This question is important to under-
stand how capital markets incorporate the information contained in M&A announce-
ments into stock prices.
I examine this question using a sample of 2,511 deals which have a minimum
transaction value of $100 million and are announced between January 1, 1985 and De-
cember 31, 2005. Based on return correlations measured over a year prior to the an-
nounced transaction, I identify the 30 most closely related rivals of each target firm.
11
Next, I determine subsets of these related rivals by measuring the short-term reaction
around the deal announcement: I am interested in large positive or negative rival reac-
tions due to the positive information-signaling or negative competitive effects, respec-
tively, of the transactions. I select the rival events based on a combination of an absolute
percentage cut-off and a relative standard-deviation based requirement. That is, a short-
term absolute rival CAR of 5% (10%) is deemed to be a large (very large) return event,
if it is at least two standard deviations from zero. Overall, I investigate samples of 6,138
large rival gain events and 5,408 large rival loss events, where rivals, on average, ex-
perience absolute abnormal returns of roughly 10.6% in a five-day window around the
transaction announcement.
After controlling for the size and book-to-market effects using matching firms, I
find that rivals with large gains at the announcement of a deal due to positive informa-
tion-signaling (‘gaining rivals’) experience significantly positive buy-and-hold-
abnormal returns (BHAR) of 2.844% for a year after the announcement, while rivals
suffering large losses due to the negative competitive effects of a M&A transaction
(‘losing rivals’) experience significantly negative long-term returns of 8.146%. The dif-
ference between the two samples of 10.990% is highly statistically significant. When I
investigate very large return events, the BHAR difference between heavily gaining and
heavily losing rivals becomes even larger at 17.509% over 12 months. This finding in-
dicates that capital markets underreact to the information contained in M&A transac-
tions for the stock price performance of rivals, a result which fits well into the literature
on underreaction of capital markets to new information (Frazzini (2006), Jackson and
Johnson (2006), Zhang (2006)).
I document the robustness of my results by using an alternative methodology:
the calendar-time portfolio approach. A long-short portfolio with a long investment in
gaining rivals and a short investment in losing rivals earns an average monthly raw re-
turn of 0.629% (7.55% annualized) and an average monthly size and book-to-market
adjusted return of 0.638% (7.66% annualized) over the 12-month horizon, both highly
statistically significant. Using the Fama and French (1993) three-factor asset pricing
model as an alternative return adjustment method results in an average monthly alpha
12
for the long-short portfolio of 0.726% (8.71% annualized), also highly statistically sig-
nificant. Overall, the results indicate that capital markets do not immediately incorpo-
rate the positive information signaling and negative competitive effects of M&A an-
nouncement into rival stock prices resulting in a return drift of up to a year after the
event.
Next, I document the impact of rival size on the observed long-term return drift.
Only micro-cap gaining rivals smaller than the 20th NYSE market capitalization percen-
tile experience significantly positive long-term returns: their 12-month post-event
BHAR is a statistically significant 8.548%. Small gaining rivals (between the 20th and
50th percentile) and big gaining rivals (above the 50th percentile) experience positive but
insignificant BHARs of 3.051% and 1.049%, respectively. This concurs with the con-
jecture that positive information signaling is more important for the smallest rivals: they
face a higher takeover likelihood and, therefore, experience a larger post-event drift.
However, for losing rivals, there is no discernible impact of rival size: the micro-cap,
small, and big firm subsamples earn comparable negative 12-month BHARs of 7.136%,
10.086%, and 7.345%, respectively.
Finally, I employ cross-sectional BHAR regressions in order to better understand
the drivers of the return drift. However, the regressions do not paint a precise picture of
the long-term BHARs to rivals. Further research into the underlying causes of the large
rival return events, i.e., positive information signaling and negative competitive effects
of M&A, is warranted in order to better understand the underreaction of capital markets
and the ensuing return predictability.
The remainder of this essay is structured as follows. Section 2.2 discusses the
findings of the related literature. Section 2.3 introduces the data and methodology used
in the empirical investigations. Section 2.4 presents the empirical results and Section 2.5
concludes.
13
2.2 Related Literature
Many financial researchers have addressed the question of the wealth effects of M&A.
In general, they find that target shareholders gain significantly from acquisitions, while
acquiring firm shareholders lose or, at most, break even around M&A announcements.
The evidence is usually based on event studies using short-term event windows around
the announcement date and assumes that prices fully adjust to the expected efficiency
gains from acquisitions.1
Following these short-term event studies, some researchers have examined the
assumption of market efficiency by measuring the long-term performance following
acquisitions. Agrawal, Jaffe and Mandelker (1992) show that acquiring firms experience
significantly negative abnormal returns over a five-year post-merger period. Loughran
and Vijh (1997) document that cash bidders experience significantly positive abnormal
returns in the three to five years following a transaction announcement, although ac-
quirers experience negative long-term returns on average. Rau and Vermaelen (1998)
show that the long-term underperformance of acquiring firms is predominantly caused
by the poor post-acquisition performance of low book-to-market “glamour” firms.2
Researchers do not only examine the transaction success of the merging firms,
but also extensively investigate how other firms in the same industry are affected by the
transaction. The intra-industry effects of an acquisition are essentially based on two
factors. Firstly, rival firms may benefit from the positive information-signaling effect of
a transaction regarding future takeover activity in the industry. Secondly, rivals may be
hurt by the competitive effects of the M&A announcements due to increased competi-
tion from the newly merged firms. Overall, researchers have found that the positive in-
formation signaling effects outweigh the negative competitive effects of M&A transac-
1 The early literature is summarized in Jensen and Ruback (1983) and Jarrell, Brickley and Netter (1988). More recent evidence on the determinants of acquiring firm abnormal returns includes Fuller, Netter and Stegemoller (2002), Mitchell, Pulvino and Stafford (2004), Moeller, Schlingemann and Stulz (2004), Masulis, Wang and Xie (2007), and Moeller, Schlingemann and Stulz (2007).
2 Other studies documenting the long-term underperformance of acquiring firms include Asquith (1983), Loderer and Martin (1992) and Agrawal, Jaffe and Mandelker (1992). The literature on long-term performance after M&A announcement is reviewed in Andrade, Mitchell and Stafford (2001).
14
tions. For example, Eckbo (1983) and Eckbo (1985) find positive intra-industry effects
around the announcement of horizontal mergers. In a more extensive study, Song and
Walkling (2000) stress the importance of positive information-signaling effects due to
the increased takeover likelihood and document positive abnormal returns for rivals of
the target in their sample of horizontal and non-horizontal deals. More recently, Fee and
Thomas (2004) and Shahrur (2005) also show that competitors experience positive
stock market effects at the announcement of horizontal deals.3
Nevertheless, so far researchers have only investigated the short-term effects of
M&A announcements on rival companies. Given the evidence on the long-term under-
performance of acquiring firms, it is straightforward to ask whether we can observe a
similar underreaction of capital markets to the information contained in M&A an-
nouncements for the case of rival companies. To the author’s best knowledge this issue
has not been examined so far. Hence, in this essay I address the question of whether
capital markets efficiently incorporate the implications from M&A transactions into
competitors’ stock prices or whether we are able to observe any abnormal long-run re-
turns for rivals of the target company.
In general, besides the M&A literature, this essay fits into the extensive body of
evidence examining long-run stock returns following various kinds of corporate events,
such as earnings announcements (Bernard and Thomas (1989)), IPOs (Loughran and
Ritter (1995)), SEOs (Spiess and Affleck-Graves (1995)), share repurchases (Ikenberry,
Lakonishok and Vermaelen (1995)), dividend initiations and omissions (Michaely,
Thaler and Womack (1995)), stock splits (Ikenberry, Rankine and Stice (1996)),
spinoffs (Desai and Jain (1999)), debt offerings (Spiess and Affleck-Graves (1999)), or
R&D increases (Eberhart, Maxwell and Siddique (2004)). I extend this literature in an
important aspect by examining the long-term consequences of the corporate event for
related companies. In a recent study, Massa, Rehman and Vermaelen (2007) investigate
the information which open-market share repurchase announcements provide about the
3 However, Shahrur (2005) shows that rivals only gain significantly at the announcement of value-increasing transactions. Value-decreasing acquisitions are accompanied by negative wealth effects to the competitors.
15
competitors of the repurchasing firms. In the course of their investigation, they also ana-
lyze the implications for rival firm stock price performance in the post-announcement
period. In this essay, I study the consequences of M&A announcements for long-
horizon rival returns. This topic is of particular interest: acquisitions arguably form the
most important type of corporate events, but the long-term performance of rivals has not
been investigated so far.
2.3 Data and Methodology 2.3.1 Data
Firstly, my data of mergers and acquisitions are from the Thomson One SDC Platinum
Database.4 I consider all deals that were announced between January 1, 1985 and De-
cember 31, 2005, which have a transaction value of at least $100 million and where the
acquirer is seeking to purchase at least 50% of shares. In addition, I require both ac-
quirer and target firm to be publicly listed U.S. companies, excluding foreign acquirers
and private firms. This results in an initial sample of 2,836 deals. I assign each transac-
tion to one of 12 industry sectors5 based on the target’s SIC code.6
Secondly, my stock return data are from CRSP. I only use stocks identified as
common equity (share code 10 and 11), excluding all certificates, ADRs, SBIs, REITs,
etc. In order to minimize the influence of market microstructure effects and to ensure
that the results are not driven by the tiniest micro-cap stocks, I exclude all firms which
have a closing stock price of less than 5$ and/or which have a market value of equity
4 The mergers and acquisitions section (SDC Platinum) of the Thomson One Banker Database was formerly known as the Securities Data Company’s (SDC) U.S. Mergers and Acquisitions Database.
5 As industry sector definitions I use the most recent version of the 12 industry portfolios defined by Fama and French (1997). I thank Ken French for providing the data on his website. The 12 industries are consumer nondurables, consumer durables, manufacturing, energy, chemicals, business equip-ment, telecom, utilities, wholesale and retail, healthcare, finance, and other.
6 I use target SIC codes from CRSP, which allows for time-series variation in industry classification. However, I can report that all our main results remain qualitatively and quantitatively similar if I use the SIC code as recorded by Thomson.
16
below the 5th market capitalization percentile of NYSE stocks in the month before the
M&A announcement.7
Finally, my accounting data are from the annual Compustat files. Following the
usual practice in the asset pricing literature based on Fama and French (1992), I match
accounting variables for fiscal year-ends in year t 1 with CRSP returns from July of
year t to June of year t+1. Detailed definitions of the variables can be found in the ap-
pendix of this chapter.
2.3.2 Event Identification
I am interested in investigating the abnormal long-term return of rivals that experience
heavy positive information signaling effects or heavy competitive effects at the an-
nouncement of a transaction. In the course of the investigation, the rival identification
approach is as follows. I identify a set of potential target rivals based on the industry
classification of each transaction. Then, I narrow down this relatively broad set of firms
by calculating the return correlation of each potential rival with the target.8 For the ini-
tial analysis, I define the 30 firms as the rivals of the target which have the highest cor-
relation measured over a 250-day window ending the month before the transaction.9 In
addition, I impose a minimum return correlation of 0.15 before I classify a firm as a
7 I thank Kenneth French for providing the NYSE breakpoints on his website. I can report that my findings are not sensitive to these exclusions.
8 Such a correlation-based approach to find the most related rivals is also used by Akhigbe and Martin (2000). I base my approach on a broad industry classification using 12 industry sectors (see footnote 5), as I rely on the correlation to clearly identify related rivals and want to avoid reducing the pool of potential rival firms too much by using a narrow industry definition, such as the 48 Fama and French (1997) industries.
9 In the following analysis I focus on results using the 30 most-correlated rivals to guarantee a suffi-cient sample size for the empirical investigations involving large return events. However, I can report that results are qualitatively and quantitatively similar using only the 10 or 20 most correlated rivals.
17
rival.10 The sample is reduced to 2,511 deals, as I exclude all deals where the target did
not have stock returns for the correlation-estimation window.11
To identify those rivals that experience heavy positive information-signaling ef-
fects or heavy negative competitive effects, I rely on the short-term reactions of the re-
spective rival company. Rivals are assumed to experience positive information signaling
effects if they have large (more than 5%) or very large (more than 10%) abnormal gains
around the transaction announcement, in the following simply referred to as ‘gaining
rivals’ and ‘heavily gaining rivals’, respectively. Rivals are assumed to experience
negative competitive effects if they have large (less than -5%) or very large (less than -
10%) abnormal losses around the transaction announcement, in the following simply
referred to as ‘losing rivals’ and ‘heavily losing rivals’, respectively.
In addition, I choose to select the rival events not only by using an absolute per-
centage cut-off, but by also including a relative standard-deviation based requirement.
That is, a short-term absolute rival cumulative abnormal return (CAR) of 5% (10%) is
deemed to be a large (very large) return event if it is at least two standard deviations
from zero. I include this additional requirement, as a constant absolute return may be
important or not depending on the volatility of the rival’s stock returns. By combining
an absolute cut-off with the requirement for a two standard deviation move, I make sure
that the sample is not dominated by high volatility stocks and that, at the same time, a
minimum absolute price change is needed for low volatility stocks to qualify for an
event, ensuring a general ‘relevance’ of the rival return events.12
10 The mean (median) correlation of the 30 rivals is 0.257 (0.222). However, as the minimum correla-tion is only 0.008, I require a minimum correlation of 0.15. This removes about 13% of potential ri-vals. This screen ensures that I do not spuriously include lowly correlated rivals in the analysis.
11 I end the correlation estimation the month before the transaction so as to not run the risk of any bias due to market rumours in the run-up to the acquisition. I can report that the results are robust both to the length of the correlation estimation window (250, 180, or 120 days) and the gap before the trans-action (none, one month, two months).
12 To my knowledge I am the first to use such a combination. So far, the literature has focused on either absolute cut-offs (Brown, Harlow and Tinic (1988), Bremer and Sweeney (1991), Cox and Peterson (1994), Park (1995)) or relative cut-offs (Pritamani and Singal (2001)). However, I can report that my results remain quantitatively and qualitatively similar, if I use absolute cut-offs of 5% and 10% only.
18
For the short-term event study, I base the methodology on Brown and Warner
(1985) and use the modified market model to estimate abnormal returns. I do not use the
market model because, for many rivals, there are months with several merger an-
nouncements, which implies a high probability of other bid announcements occurring in
the estimation period. Any abnormal returns caused by these announcements may bias
the estimation.13 I calculate the CAR for a five-day event window [-2,2] around the an-
nouncement date as:14
2
2,,]2,2[ )(
ttmti RRCAR , (2.1)
where Ri,t is rival i’s return on date t and Rm,t is the return for the equal-weighted CRSP
index on date t.
For each sample year, Table 2.1 shows the number of deals, the percentage of
the total number of deals during that specific year, as well as the average reaction of
equal-weighted portfolios of rivals to the M&A announcements. The table reveals that
most of the transactions take place at the height of the economy-wide merger wave at
the end of the 1990s. On average, positive information-signaling outweighs the negative
competitive effects of M&A, resulting in positive and significant rival reactions of
0.056%.15
13 Fuller, Netter and Stegemoller (2002) do not use the market model for the same reasons. Brown and Warner (1980) show that weighting the market return by a firm’s beta does not significantly improve the estimation.
14 My main findings are not susceptible to the choice of the event window length. They also hold for a shorter time horizon, such as three days, to estimate CARs.
15 This finding is consistent with the findings of Song and Walkling (2000), Fee and Thomas (2004), and Shahrur (2005) who document that the positive information-signaling effects of M&A an-nouncements, on average, outweigh the competitive effects of the deals.
19
Table 2.1: Annual Distribution of M&A Transactions and Average Rival Reactions The sample contains all U.S. M&A announcements between January, 1985 and December, 2005 with a transaction value of at least $100 million, publicly listed target and acquirer firms, and the intent of the acquirer to purchase at least 50% of shares. For each year, the table shows the number of deals, the per-centage of the total number of deals during that specific year, as well as the average rival reaction to M&A announcements during that year. Rivals are the 30 firms from the target’s industry with the highest return correlation to the target measured over a 250-day window ending the month before the transaction. CAR is the cumulative abnormal return to the equal-weighted rival portfolio for a five-day event window [-2,2] around the announcement date. Abnormal returns are calculated using the modified market model with the equal-weighted CRSP index as benchmark. t-statistics for the total sample are provided in brack-ets.
Year Number As % of Average Rival of Deals All Deals CAR in %
1985 93 3.70 0.338 1986 90 3.58 0.304 1987 85 3.39 -0.323 1988 74 2.95 -0.146 1989 58 2.31 0.426 1990 31 1.23 -0.054 1991 41 1.63 -0.059 1992 52 2.07 -0.699 1993 73 2.91 -0.456 1994 106 4.22 0.195 1995 149 5.93 0.086 1996 167 6.65 0.211 1997 259 10.31 0.183 1998 259 10.31 0.178 1999 260 10.35 -0.098 2000 199 7.93 0.231 2001 131 5.22 -0.345 2002 70 2.79 -0.762 2003 101 4.02 0.078 2004 111 4.42 0.412 2005 102 4.06 0.332
Total 2,511 100.00 0.056 (2.02)
Figure 2.1 shows the historical probability distribution of the 30 most-correlated rival’s
CARs. As expected, most reactions are centered around zero, with 71.2% of rivals ex-
periencing CARs between -5% and +5%. What is more interesting, 14.7% of rivals ex-
perience CARs of more than 5%, a clear sign of positive information signaling of the
M&A announcement, and 14.1% of rivals experience CARs of less than -5%, indicative
of the negative competitive effects of the transaction. In the following, I will concen-
trate on these extremes of the CAR distribution to examine the long-term effects of
M&A transactions on rivals who are clearly affected by the announcement.
20
Figure 2.1: Historical Distribution of Rival Cumulative Abnormal Returns The sample contains all U.S. M&A announcements between January, 1985 and December, 2005 with a transaction value of at least $100 million, publicly listed target and acquirer firms, and the intent of the acquirer to purchase at least 50% of shares. The figure shows a histogram of rival short-term reactions to the deal announcement. Rivals are the 30 firms from the target’s industry with the highest return correla-tion to the target measured over a 250-day window ending the month before the transaction. CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Abnormal returns are calculated using the modified market model with the equal-weighted CRSP index as benchmark.
Historical Distribution of Rival Cumulative Abnormal Returns
0,6% 0,9% 1,0% 1,4% 1,8%2,3%
3,1%
4,1%
5,7%
7,6%
9,2%10,1% 9,7%
8,5%
6,9%
5,4%
4,1%3,1%
2,3% 1,9%1,5% 1,1% 0,9% 0,7%
0%
2%
4%
6%
8%
10%
12%
-12% to
-11%
-11%to
-10%
-10% to
-9%
-9%to
-8%
-8%to
-7%
-7% to -6
%
-6%to
-5%
-5% to -4
%
-4%to
-3%
-3%to
-2%
-2% to -1
%
-1%to
0%
0% to 1%
1%to
2%
2%to 3%
3% to 4%
4%to
5%
5% to 6%
6%to
7%
7% to
8%
8% to 9%
9% to 10
%
10%to
11%
11% to
12%
CAR Range
Prob
abili
ty
Historical Distribution of Rival Cumulative Abnormal Returns
0,6% 0,9% 1,0% 1,4% 1,8%2,3%
3,1%
4,1%
5,7%
7,6%
9,2%10,1% 9,7%
8,5%
6,9%
5,4%
4,1%3,1%
2,3% 1,9%1,5% 1,1% 0,9% 0,7%
0%
2%
4%
6%
8%
10%
12%
-12% to
-11%
-11%to
-10%
-10% to
-9%
-9%to
-8%
-8%to
-7%
-7% to -6
%
-6%to
-5%
-5% to -4
%
-4%to
-3%
-3%to
-2%
-2% to -1
%
-1%to
0%
0% to 1%
1%to
2%
2%to 3%
3% to 4%
4%to
5%
5% to 6%
6%to
7%
7% to
8%
8% to 9%
9% to 10
%
10%to
11%
11% to
12%
CAR Range
Prob
abili
ty
Table 2.2 shows the number of rivals experiencing large and very large return events in
each sample year as well as the respective average CAR. Overall, there are 5,408 large
and 2,079 very large loss events, while there are 6,138 large and 2,401 very large gain
events. The table also shows the size of the events: on average, rivals experience abso-
lute returns of roughly 10.6% (16%) for the case of large (very large) return events. The
CAR pattern is fairly consistent over time, with a maximum in the high volatility period
of 2000.
21
Table 2.2: Annual Distribution of Rivals with Large Short-Term Reactions The sample contains all U.S. M&A announcements between January, 1985 and December, 2005 with a transaction value of at least $100 million, publicly listed target and acquirer firms, and the intent of the acquirer to purchase at least 50% of shares. For each year, the table shows the number of rivals with large (very large) short-term gains and losses. Rivals are the 30 firms from the target’s industry with the highest return correlation to the target measured over a 250-day window ending the month before the transaction. Large (very large) gains and losses are defined as (1) a positive or negative CAR of 5% (10%), respec-tively, given (2) at least a 2 standard deviation distance from zero. CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Abnormal returns are calculated using the modified market model with the equal-weighted CRSP index as benchmark. The standard deviation of abnormal returns is estimated over the same 250-day window as the correlations.
Year Number of Rival Events Average Rival CAR in % Very
Large Losses
Large Losses
Large Gains
VeryLarge Gains
Very Large
Losses
Large Losses
Large Gains
VeryLarge Gains
1985 22 122 178 36 -13.501 -8.030 8.262 13.792 1986 25 123 174 43 -14.744 -8.591 8.857 14.898 1987 65 226 228 52 -16.110 -9.694 8.674 14.116 1988 29 141 123 34 -13.100 -8.167 9.203 14.977 1989 5 65 113 20 -12.421 -7.260 8.367 14.464 1990 21 81 83 27 -16.223 -9.282 9.396 15.125 1991 35 110 110 42 -13.965 -9.208 10.194 14.999 1992 37 131 72 27 -13.272 -8.794 10.199 15.580 1993 35 116 95 37 -13.375 -9.022 9.934 14.372 1994 33 168 202 70 -14.600 -8.505 9.408 13.224 1995 62 226 269 86 -14.127 -8.938 9.555 15.135 1996 89 244 326 144 -17.080 -10.784 10.661 15.027 1997 178 507 695 228 -16.607 -10.487 9.630 14.932 1998 320 757 838 361 -14.921 -10.444 10.961 15.867 1999 283 739 695 345 -15.796 -10.556 12.165 17.197 2000 360 599 742 425 -18.604 -14.145 13.673 18.249 2001 232 374 344 171 -17.520 -13.581 12.445 17.548 2002 100 205 197 71 -16.810 -11.976 10.123 15.140 2003 50 160 190 67 -13.616 -9.230 9.551 14.228 2004 48 151 230 65 -13.771 -9.003 8.745 13.361 2005 50 163 234 50 -13.406 -8.877 8.416 13.556
Total 2,079 5,408 6,138 2,401 -16.103 -10.578 10.613 16.013
2.3.3 Long-Term Performance
For the evaluation of long-term performance, appropriate adjustment for risk becomes
critical. First, even small errors in risk adjustment can make an economically large dif-
ference when calculating abnormal returns over longer horizons spanning months or
years, while such errors make little difference for short-term event windows. Second,
the question of the correct model for expected returns is unresolved in the asset pricing
literature, and therefore long-horizon abnormal return estimates suffer from the bad
22
model problem (Kothari and Warner (2007)). Essentially, as emphasized by Fama
(1970, 1998), any test of long-horizon abnormal returns is a joint test of market effi-
ciency and the model of expected returns employed. Bearing this caveat in mind, I care-
fully proceed in selecting the best available empirical methods for the evaluation of ri-
val long-term performance.
Generally, there are two main methods used for measuring long-term stock price
performance: buy-and-hold abnormal returns (BHAR) and the calendar-time portfolio
approach. Despite of an extensive literature, there is no clear winner in a comparison of
the two.16 Therefore, I use both in the empirical investigations. In general, BHARs more
accurately represent an investors’ actual investment experience than the periodic
(monthly) rebalancing entailed in the calendar-time portfolio approach (Lyon, Barber
and Tsai (1999)). Furthermore, Loughran and Ritter (2000) show that the BHAR meth-
odology has substantially higher statistical power to detect abnormal performance than
the calendar-time method.17 The BHAR approach is also known as the characteristic-
based matching approach as it involves matching event firms to non-event firms based
on stock characteristics such as size and book-to-market. I use a matching firm approach
to avoid the skewness problems associated with the use of matching reference portfolios
(Barber and Lyon (1997)). Specifically, for each of the event firms, I first identify all
firms with a market capitalization between 70% and 130% of the market capitalization
of the event firm. Among these firms, I then choose the firm with the book-to-market
ratio closest to that of the event firm. Once a control firm j is identified, a T-month
BHAR for event firm i can be calculated by:
T
ttj
T
ttiTti RRBHAR
1,
1,],[, )1()1( , (2.2)
16 A selected list of this literature includes Barber and Lyon (1997), Kothari and Warner (1997), Fama (1998), Lyon, Barber and Tsai (1999), Mitchell and Stafford (2000) and Kothari and Warner (2007).
17 The use of the BHAR methodology is substantiated by the fact that its statistical problems only be-come severe at very long post-event horizons such as three to five years. Therefore, as I focus on shorter horizons up to a year after the event, I encounter less severe statistical problems than other long-term performance studies.
23
If a matching firm is delisted from CRSP, the next firm with the closest book-to-market
ratio is chosen as additional matching firm. Following Loughran and Vijh (1997), re-
turns for the additional matching firms are spliced in from that point in time going for-
ward. For the 12-month post-event horizon, only 9% of the rival sample needs a second
matching firm and only 1% a third. Finally, for the sample of event firms, the mean
BHAR is calculated as the simple average of the individual firm BHARs. Statistical
significance can be evaluated using a standard cross-sectional t-statistic as the use of a
single control firm alleviates the skewness bias (Barber and Lyon (1997)).
However, the issue of cross-sectional correlation remains and cannot be ne-
glected as the majority of transactions and, therefore, the majority of rival events cluster
in the five years from 1996 to 2000 (see Table 2.1 and Table 2.2). Furthermore, cross-
sectional correlation may be due to two additional sources: over-representation of some
industries with high merger activity in the sample and overlapping long-horizon returns
for multiple-event firms. In principle, these issues can be accommodated in a BHAR
approach using the methodology developed by Jegadeesh and Karceski (2004). How-
ever, I choose to use a calendar-time portfolio approach instead to evaluate the robust-
ness of the results with an alternative methodology.
The calendar-time methodology was initially developed by Jaffe (1974) and
Mandelker (1974) and is especially advocated by Fama (1998) and Mitchell and Staf-
ford (2000). The approach involves calculating calendar-time portfolios for firms ex-
periencing an event. Suppose I want to evaluate long-term performance over T=12
months. Then, in each calendar month over the entire sample period, a portfolio is con-
structed comprising all firms with an event within the previous 12 months. As the num-
ber of firms in a portfolio is not constant over time, the portfolio is rebalanced every
month and the equal-weighted return is calculated for every sample month. The result-
ing time-series of returns is free of any cross-correlation issues and can therefore be
used for standard statistical tests.
I report average monthly raw returns as well as average monthly abnormal re-
turns using the size and book-to-market matched control firm from the BHAR approach.
However, more interestingly, the calendar-time portfolio approach is usually employed
24
with a different risk-adjustment methodology by estimating abnormal returns using a
multi-factor asset pricing model in the spirit of the intertemportal capital asset pricing
model of Merton (1973) and the arbitrage pricing theory of Ross (1976). Therefore, the
time-series of calendar-time portfolio excess returns is regressed on a three-factor model
comprising the three Fama and French (1993) market, size, and book-to-market factors.
Specifically, I regress:
tptptptftmpptftp HMLhSMBsRRbaRR ,,,,, )()()( (2.3)
where Rp,t is the portfolio return comprising t months of event firms, for t=(1,3,6,9,12),
Rf,t is the risk-free rate, Rm,t is the value-weighted CRSP index, bp, sp, and hp are the co-
efficients of the Fama and French (1993) market (Rm-Rf), size (SMB), and book-to-
market (HML) portfolio returns.18 Inferences about the abnormal performance can be
drawn on the basis of the estimated intercept ap and its statistical significance. To ac-
count for heteroskedasticity due to the changing number of portfolio firms and possible
autocorrelation, I use Newey and West (1987) heteroskedasticity- and autocorrelation-
corrected standard errors with three monthly lags to calculate t-statistics.19
2.4 Results2.4.1 Buy-and-Hold Abnormal Returns
Based on the events identified by the short-term event study, I proceed in evaluating the
long-term performance of rival companies. Table 2.3 displays the main findings using
the BHAR methodology with size and book-to-market matched control firms. For the
large return events in Panel A, I find positive long-horizon post-event BHARs for gain-
ing rivals and negative long-horizon post-event BHARs for losing rivals. Rivals who
18 SMB (small minus big) is the difference between the return on the portfolio of “small” and “big” stocks and HML (high minus low) is the difference between the return on the portfolio of “high” and “low” book-to-market stocks. The construction of these portfolio returns is discussed in detail in Fama and French (1993). I thank Kenneth French for providing the factor data on his website.
19 I chose the number of monthly lags using the time-series-dependent rule of thumb 0.5T1/3. However, the results are similar for any other reasonable number of lags. In addition, a simple White (1980) correction for heteroskedasticity shows similar results.
25
have large gains due to positive information-signaling at the announcement of a deal
experience continued positive long-term returns amounting to a significant average
BHAR of 2.844% after 12 months. Rivals who suffer large losses due to the negative
competitive effects of a M&A transaction experience continued negative long-term re-
turns resulting in a highly significant negative average BHAR of 8.146% after 12
months. Testing the two samples for differences shows an average BHAR difference of
10.990% with a t-statistic of over six.
Table 2.3: Long-Term Rival Buy-and-Hold-Abnormal ReturnsThe sample contains all U.S. M&A announcements between January, 1985 and December, 2005 with a transaction value of at least $100 million, publicly listed target and acquirer firms, and the intent of the acquirer to purchase at least 50% of shares. The table shows the long-term buy-and-hold abnormal returns (BHAR) of rivals with large (Panel A) and very large (Panel B) short-term returns around the deal an-nouncement. Rivals are the 30 firms from the target’s industry with the highest return correlation to the target measured over a 250-day window ending the month before the transaction. Large (very large) re-turn events are defined as (1) a positive or negative CAR of 5% (10%), respectively, given (2) at least a 2 standard deviation distance from zero. CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Abnormal returns are calculated using the modi-fied market model with the equal-weighted CRSP index as benchmark. The standard deviation of abnor-mal returns is estimated over the same 250-day window as the correlations. Long-term performance is evaluated using buy-and-hold abnormal returns with size and book-to-market matched control firms as benchmark. Returns are in percent, t-statistics are shown in parentheses.
Holding Period 1 3 6 9 12
Panel A: Large Return Events
BHAR in % 0.383 0.207 0.431 1.101 2.844 (t-statistic) (1.51) (0.47) (0.65) (1.19) (2.27) Gaining
Rivals (1) N 6,111 6,048 5,934 5,818 5,708
BHAR in % -0.164 -0.225 -2.494 -4.532 -8.146 (t-statistic) (-0.60) (-0.43) (-2.96) (-4.31) (-6.29) Losing
Rivals (2) N 5,386 5,349 5,276 5,193 5,096
BHAR in % 0.547 0.432 2.925 5.633 10.990 Difference (1)-(2) (t-statistic) (1.47) (0.64) (2.76) (4.03) (6.09)
Panel B: Very Large Return Events
BHAR in % 0.354 -0.057 0.767 1.259 4.614 (t-statistic) (0.71) (-0.07) (0.59) (0.69) (1.87)
HeavilyGaining
Rivals (1) N 2,388 2,353 2,301 2,249 2,198
BHAR in % -0.369 -0.512 -2.689 -5.486 -12.895 (t-statistic) (-0.69) (-0.48) (-1.78) (-2.55) (-4.80)
HeavilyLosing
Rivals (2) N 2,069 2,052 2,024 1,992 1,946
BHAR in % 0.722 0.455 3.455 6.745 17.509 Difference (1)-(2) (t-statistic) (0.99) (0.34) (1.74) (2.41) (4.81)
26
Turning to the very large return events in Panel B, we can observe that the results are
stronger if we examine a more extreme event definition. Over a holding period of 12
months, heavily gaining rivals experience a significantly positive average post-event
BHAR of 4.614%, while heavily losing rivals experience a significantly negative aver-
age BHAR of 12.895%. The average BHAR difference of 17.509% is highly statisti-
cally significant with a t-statistic of almost five.
Figure 2.2 visualizes the results for large rival return events in an event-time
graph displaying average BHARs. The upper solid line shows the results for gaining
rivals, the lower solid line for losing rivals; 95% confidence intervals are indicated by
the dotted lines.
Figure 2.2: Event-Time Buy-and-Hold Abnormal Rival ReturnsThe sample contains all U.S. M&A announcements between January, 1985 and December, 2005 with a transaction value of at least $100 million, publicly listed target and acquirer firms, and the intent of the acquirer to purchase at least 50% of shares. The figure shows event-time buy-and-hold-abnormal returns of rivals with large short-term gains (upper solid line) or losses (lower solid line) around the deal an-nouncement (month 0). Rivals are the 30 firms from the target’s industry with the highest return correla-tion to the target measured over a 250-day window ending the month before the transaction. Large gains and losses are defined as (1) a positive or negative CAR of 5% given (2) at least a 2 standard deviation distance from zero. CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Abnormal returns are calculated using the modified market model with the equal-weighted CRSP index as benchmark. The standard deviation of abnormal returns is esti-mated over the same 250-day window as the correlations. Long-term performance is evaluated using buy-and-hold abnormal returns with size/book-to-market matched control firms as benchmark. Returns are in percent, 95% confidence intervals are indicated by the dotted lines.
Event-Time Buy-and-Hold Abnormal Rival Returns
-20-16-12
-8-4048
121620
-1 0 1 2 3 4 5 6 7 8 9 10 11 12
Event Month
BH
AR
in %
Event-Time Buy-and-Hold Abnormal Rival Returns
-20-16-12
-8-4048
121620
-1 0 1 2 3 4 5 6 7 8 9 10 11 12
Event Month
BH
AR
in %
27
After zero returns in pre-event month t-1, rival firms experience per definitionem large
gains or large losses in the transaction month t. This event is followed by a continuous
drift of returns in the same direction as the initial reaction to the deal announcement. For
positive events, the drift is only marginal in the first six post-event months as already
indicated in Table 2.2, but afterward picks up pace and becomes highly significant after
12 months. For negative events, the line is consistently downward sloping with statisti-
cally significant negative post-event performance already reached after six months. I
also evaluated longer time-horizons of two and three years after the event, but I was not
able to find any reliable additional drift beyond the first post-event year.20
Overall, from these results we can conclude that rivals of M&A target firms who
show a large reaction to the deal announcement experience a continued return drift in
the same direction as the initial reaction. This indicates that capital markets underreact
to the information contained in M&A transactions for the stock price performance of
rivals, a result which fits well in the literature on underreaction of capital markets to
new information (Frazzini (2006), Jackson and Johnson (2006), Zhang (2006)). Also
consistent with the literature, the return drift after negative events is considerably larger
and more consistently statistically significant, implying that “bad news travels slowly”
and is only incorporated with a considerable time-lag into stock prices, possibly due to
market frictions such as short-sale constraints (Hong, Lim and Stein (2000), Chan
(2003)).
2.4.2 Calendar-Time Portfolios
In order to investigate the robustness of the main findings, I employ calendar-time port-
folio returns as an alternative methodology to the use of BHARs. Table 2.4 shows the
results when I use the calendar-time methodology based on raw returns and size and
book-to-market adjusted returns. For the 12-month holding period, Panel A shows that
the average monthly raw return for the gaining rivals portfolio of 1.423% is substan-
tially larger than the average return of 0.794% for the losing rivals portfolio, resulting in
20 The peak of the post-event BHAR appears after about 15 months; afterwards, I cannot find any fur-ther abnormal performance. Results are available on request from the author.
28
a highly significant average monthly raw return of 0.629% (annualized 7.55%) for the
long-short portfolio.
Table 2.4: Calendar-Time Portfolio ReturnsThe sample contains all U.S. M&A announcements between January, 1985 and December, 2005 with a transaction value of at least $100 million, publicly listed target and acquirer firms, and the intent of the acquirer to purchase at least 50% of shares. The table shows average monthly calendar-time portfolio raw returns (Panel A) and adjusted returns (Panel B) for rivals with large short-term returns around the deal announcement. Adjusted returns are calculated with size and book-to-market matched control firms as benchmark. Rivals are the 30 firms from the target’s industry with the highest return correlation to the target measured over a 250-day window ending the month before the transaction. Gaining and losing rivals are defined as rivals with (1) a positive or negative CAR of 5% given (2) at least a 2 standard devia-tion distance from zero. CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Abnormal returns are calculated using the modified market model with the equal-weighted CRSP index as benchmark. The standard deviation of abnormal returns is estimated over the same 250-day window as the correlations. Returns are in monthly percent, t-statistics are shown in parentheses.
Holding Period
1 3 6 9 12
Panel A: Raw Returns
1.861 1.772 1.522 1.482 1.423 GainingRivals (1) (3.61) (4.16) (3.70) (3.65) (3.54)
0.701 0.810 0.742 0.846 0.794 LosingRivals (2) (1.36) (1.70) (1.57) (1.78) (1.71)
1.194 0.963 0.780 0.636 0.629 Long-Short (1)-(2) (2.83) (3.65) (4.04) (3.52) (4.03)
Panel B: Size-BM Adjusted Returns
1.105 0.598 0.414 0.326 0.257 GainingRivals (1) (2.62) (2.52) (2.54) (2.32) (2.09)
-0.448 -0.103 -0.336 -0.261 -0.381 LosingRivals (2) (-1.27) (-0.47) (-1.86) (-1.52) (-2.55)
1.584 0.701 0.751 0.587 0.638 Long-Short (1)-(2) (2.89) (2.37) (3.63) (3.41) (4.46)
This result carries over into Panel B where I consider the same risk-adjustment as in the
BHAR methodology: returns adjusted by size and book-to-market matched control
firms. With the use of a benchmark, we can now also directly interpret the statistical
significance of the individual portfolios. For the 12-month holding period, average
monthly size and book-to-market adjusted returns are a significantly positive 0.257%
for the portfolio of gaining rivals and a significantly negative 0.381% for the portfolio
of losing rivals. This translates into a highly statistically significant adjusted return of
29
0.638% (annualized 7.68%) for the long-short portfolio, a finding comparable to the raw
return results of Panel A.
Overall, calendar-time portfolio returns appear generally larger at the shorter ho-
rizon: scaled up to the respective holding periods, the adjusted returns for the long-short
portfolio of 1.584% after one month, 2.103% after three months and 4.504% after six
months are larger than the BHAR differences reported in Panel A of Table 2.3 (0.547%,
0.432%, and 2.925%, respectively). However, for the nine and 12-month holding peri-
ods the claendar-time adjusted returns of 5.282% and 7.655%, respectively, are compa-
rable with and smaller than the average BHARs of 5.633% and 10.990%, respectively.
These differences in average returns show the effects of event clustering in the sample.21
Overall, however, the claendar-time portfolio results indicate that my general findings
of underreaction by capital markets to the information contained in M&A announce-
ments for rival firms is robust to the issue of cross-sectional event correlation, as evi-
denced by the high significance of the long-short portfolio returns at all post-event hori-
zons.
In addition, calendar-time portfolios provide for another possibility: the use of
an alternative risk-adjustment methodology based on calendar-time multi-factor regres-
sions. Table 2.5 shows the results using the three-factor model described in Section
2.3.3. In addition to the regression intercepts, Table 2.5 also reports the factor loadings
of the portfolios. Examining the results, it is evident that both the portfolios of rivals
with large gains and large losses represent stocks with high risk with respect to the
Fama-French three-factor model: both portfolios have market betas above one and sig-
nificantly positive loadings on the size-factor (with negligible exposure to the value
factor). Combining the two portfolios, the long-short portfolio of rivals shows balanced
risk exposure: it has a significantly negative exposure to the market factor and no expo-
sure to the size and value factors.
21 These results imply that during the time period of the majority of the events (1996 to 2000) the return drift was less pronounced for the shorter holding periods and more pronounced for the 12-month ho-rizon, a deduction which I confirmed in unreported subsamples.
30
Table 2.5: Calendar-Time Portfolio Three-Factor AlphasThis table shows three-factor alphas, coefficients, t-statistics, and R² values from a monthly time-series regression of , where the left-hand side variable is the monthly excess return for a portfolio of rivals with large return events in the last t months and the right-hand side variables are contemporaneous Fama and French (1993) market, size, and book-to-market factor mimicking portfolio returns. Panel A reports the results for gaining rivals, Panel B presents the findings for losing rivals, and Panel C displays the results for a long-short portfolio. Rivals are the 30 firms from the target’s industry with the highest return correlation to the target measured over a 250-day window ending the month before the transaction. Gaining and losing rivals are defined as rivals with (1) a positive or negative CAR of 5% given (2) at least a 2 standard deviation distance from zero. CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Abnormal returns are calculated using the modified market model with the equal-weighted CRSP index as benchmark. The standard deviation of abnormal returns is estimated over the same 250-day win-dow as the correlations. Alphas are in monthly percent, t-statistics based on Newey and West (1987) autocorrelation- and heteroskedasticity robust standard errors with three lags are shown in parentheses below the coefficient estimates.
Holding Period Pa (%) Pb (Rm-Rf) Ps (SMB) Ph (HML) R² (%)
Panel A: Gaining Rivals 1 0.849 1.107 0.628 -0.220 64.39 (2.59) (17.25) (6.52) (-1.75) 3 0.619 1.125 0.599 -0.032 78.86 (3.01) (24.42) (8.03) (-0.32) 6 0.273 1.213 0.598 0.077 89.48 (1.79) (36.18) (11.80) (1.15) 9 0.248 1.202 0.585 0.059 90.82 (2.04) (38.91) (11.51) (0.95) 12 0.173 1.225 0.538 0.073 92.12 (1.57) (40.66) (8.42) (1.32) Panel B: Losing Rivals 1 -0.456 1.273 0.382 -0.159 63.80 (-1.53) (13.63) (2.25) (-1.07) 3 -0.455 1.327 0.449 -0.056 77.59 (-1.99) (18.34) (3.41) (-0.45) 6 -0.599 1.397 0.485 0.026 83.80 (-2.87) (19.52) (3.70) (0.21) 9 -0.482 1.383 0.474 0.013 81.95 (-2.10) (19.24) (3.51) (0.11) 12 -0.554 1.378 0.494 0.075 82.82 (-2.62) (20.12) (3.73) (0.63) Panel C: Long-Short 1 1.297 -0.157 0.239 -0.075 2.60 (3.49) (-1.53) (1.13) (-0.41) 3 1.073 -0.202 0.150 0.025 5.63 (3.71) (-2.58) (1.18) (0.16) 6 0.873 -0.183 0.113 0.051 8.94 (4.03) (-2.97) (1.08) (0.37) 9 0.729 -0.182 0.111 0.046 9.84 (3.75) (-3.18) (1.13) (0.40) 12 0.726 -0.154 0.044 -0.002 7.62 (4.60) (-3.00) (0.57) (-0.03)
tPtPtPtftmPPtftP HMLhSMBsRRbaRR ,,,,, )()()(
31
Overall, we can conclude that the results are robust to the particular risk-adjustment
methodology chosen. My findings consistently indicate that capital markets do not im-
mediately incorporate the positive information signaling and negative competitive ef-
fects of M&A announcement into rival stock prices, resulting in a return drift of up to a
year after the event.
2.4.3 Size Groups
I show that both gaining and losing rivals experience a return drift into the same direc-
tion as the initial reaction for up to 12 months after the announcement. This raises the
question of whether this return drift is exploitable by the general investor. Therefore, I
investigate whether the results hold across size-based subsamples of stocks or whether
the findings are only relevant for small, illiquid stocks. I follow Fama and French
(2007) in differentiating between three size groups: micro-cap, small, and big firms. The
breakpoints are the 20th and 50th market capitalization percentiles of NYSE stocks in the
month before the M&A announcement.22 These three size groups are roughly in line
with the definition used by many investment managers. To put the breakpoints into per-
spective, in November 2005 (the last pre-event month) the breakpoint between micro-
cap and small firms is $627 million, and the breakpoint between small and big firms is
$1,954 million.23
The long-term returns for the different size based subsamples of rivals are shown
in Table 2.6. For micro-cap (Panel A), small (Panel B) and large rivals (Panel C), the
12-month BHAR after large losses is -7.136%, -10.086% and -7.345%, respectively.
Thus, losing rivals experience a negative post-announcement drift irrespective of their
22 However, as opposed to Fama and French (2007), I exclude the tiniest micro-caps from the analysis by requiring all event firms to have a stock price of at least 5$ and to be larger than the 5th market capitalization percentile of NYSE stocks in the month before the M&A announcement. To put the latter cutoff into perspective, the 5th percentile in November 2005 (the last pre-event month) is $176 million.
23 In addition, Fama and French (2007) report that their group of big firms is sometimes split again into mega-caps and mid-caps by investment managers. I checked for the robustness of my results across the group of big firms by splitting it again at the 80th market capitalization percentile of NYSE stocks. However, I can report that the results for mid-caps and mega-caps are qualitatively and quan-titatively similar.
32
size. As size is a proxy for short-sale constraints, this result implies that market frictions
are not a key driver of the negative post-event drift.
Table 2.6: Long-Term Buy-and-Hold Abnormal Returns for Rival Size GroupsThe sample contains all U.S. M&A announcements between January, 1985 and December, 2005 with a transaction value of at least $100 million, publicly listed target and acquirer firms, and the intent of the acquirer to purchase at least 50% of shares. The table shows the long-term buy-and-hold abnormal returns (BHAR) of microcap (Panel A), small (Panel B) and big (Panel C) rivals with large short-term returns around the deal announcement. The breakpoints for the size groups are the 20th and 50th market capitaliza-tion percentiles of NYSE stocks in the month before the transaction. Rivals are the 30 firms from the target’s industry with the highest return correlation to the target measured over a 250-day window ending the month before the transaction. Gaining and losing rivals are defined as rivals with (1) a positive or negative CAR of 5% given (2) at least a 2 standard deviation distance from zero. CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Ab-normal returns are calculated using the modified market model with the equal-weighted CRSP index as benchmark. The standard deviation of abnormal returns is estimated over the same 250-day window as the correlations. Long-term performance is evaluated using buy-and-hold abnormal returns with size and book-to-market matched control firms as benchmark. Returns are in percent, t-statistics are shown in parentheses.
Holding Period
1 3 6 9 12
Panel A: Microcap RivalsBHAR in % -0.549 -0.818 -3.023 2.164 8.548 (t-statistic) (-0.75) (-0.61) (-1.60) (0.74) (2.04)
GainingRivals (1)
N 997 987 958 929 904 BHAR in % 0.859 0.858 -0.584 -2.565 -7.136 (t-statistic) (1.16) (0.57) (-0.18) (-0.75) (-1.46)
LosingRivals (2)
N 928 914 902 885 861 BHAR in % -1.408 -1.676 -2.438 4.729 15.684 Difference
(1)-(2) (t-statistic) (-1.35) (-0.83) (-0.66) (1.06) (2.45)
Panel B: Small RivalsBHAR in % 0.605 1.052 2.444 2.272 3.051 (t-statistic) (1.22) (1.23) (1.89) (1.22) (1.23)
GainingRivals (1)
N 1,875 1,846 1,813 1,776 1,733 BHAR in % -0.066 0.864 -3.200 -5.861 -10.086 (t-statistic) (-0.12) (0.81) (-2.32) (-2.83) (-4.46)
LosingRivals (2)
N 1,659 1,646 1,624 1,590 1,554 BHAR in % 0.671 0.188 5.644 8.133 13.137 Difference
(1)-(2) (t-statistic) (0.92) (0.14) (2.99) (2.93) (3.87)
Panel C: Big Rivals BHAR in % 0.540 0.036 0.324 0.116 1.049 (t-statistic) (1.75) (0.07) (0.40) (0.11) (0.75)
GainingRivals (1)
N 3,239 3,215 3,163 3,113 3,071 BHAR in % -0.562 -1.223 -2.704 -4.395 -7.345 (t-statistic) (-1.66) (-2.05) (-2.94) (-3.78) (-5.35)
LosingRivals (2)
N 2,799 2,789 2,750 2,718 2,681 BHAR in % 1.103 1.259 3.027 4.511 8.394 Difference
(1)-(2) (t-statistic) (2.41) (1.57) (2.47) (2.88) (4.27)
33
A different picture evolves for gaining rivals. For micro-cap rivals in Panel A, the aver-
age 12-month BHAR is a significantly positive 8.548%. This return drift is substantially
larger than in the overall sample. It concurs with the intuition that positive information
signaling is more important for the smallest rivals: they face a higher takeover likeli-
hood and, therefore, experience a larger post-event drift. On the contrary, small rival
companies (Panel B) experience a positive but insignificant post-event drift of 3.051%.
For the big rival size-group, the 12-month average BHAR of 1.049% is even smaller
and without any statistical significance. Hence, we can conclude that the positive post-
event drift for gaining rivals is only relevant for micro-cap stocks.
2.4.4 Cross-Sectional Regressions
Finally, I employ cross-sectional regressions to investigate possible predictors of the
return drift. More specifically, I regress rival 12-month BHARs on a host of explanatory
variables related to the underlying causes of the rival events: the positive information-
signaling and negative competitive effects of M&A announcements. Any variables pre-
dicting takeover probabilities may be able to predict the long-term positive returns to
rivals following positive information-signaling events, while any variables determining
the negative competitive effects of M&A transactions may be able to predict the long-
term negative returns to rivals following announcements accompanied by negative com-
petitive effects.
To investigate the influence of the focus of the firm I employ a dummy equal to
one if the rival only has one business segment contributing more than 10% of sales ac-
cording to the Compustat segment files. More focused firms face a higher future take-
over probability given that they are more easily integrated into an acquirers existing
business lines without the need to finance, restructure, and eventually spin off unwanted
business segments in a conglomerate takeover. Therefore, I expect a positive sign for
this dummy in the cross-sectional BHAR regressions of gaining rivals, while the seg-
ment dummy should not have any significant influence in the BHAR regressions of los-
ing rivals. Also, I employ a number of independent variables that have been used in the
prior literature to explain the probability of takeovers and the competitive effects of
34
M&A (see, e.g., Hasbrouck (1985), Ambrose and Megginson (1992), Akhigbe and Mar-
tin (2000) or Cremers, Nair and John (2007)). These variables include the rival’s size
(the natural logarithm of the market capitalization), valuation as measured by Tobin’s
Q, asset structure (property, plant, and equipment (PPE)), cash holdings, financial lev-
erage, and profitability. Table 2.7 presents the results of the cross-sectional regressions.
Table 2.7: Cross-Sectional Buy-and-Hold Abnormal Return Regressions This table shows the results of cross-sectional regressions of 12-month rival buy-and-hold abnormal re-turns (BHAR) on a host of explanatory variables. Panel A shows the results for large gain events, Panel B displays the finding for large loss events. Rivals are the 30 firms from the target’s industry with the high-est return correlation to the target measured over a 250-day window ending the month before the transac-tion. Gaining and losing rivals are defined as rivals with (1) a positive or negative CAR of 5% given (2) at least a 2 standard deviation distance from zero. The event CAR is the cumulative abnormal return to the rival firm for a five-day event window [-2,2] around the announcement date. Abnormal returns are calcu-lated using the modified market model with the equal-weighted CRSP index as benchmark. The standard deviation of abnormal returns is estimated over the same 250-day window as the correlations. The seg-ment dummy in the regression is equal to one for firms with only one business segment and ln(market cap) is the natural logarithm of market capitalization. Tobin’s Q is calculated as market value of assets over book value of assets, where the market value of assets is estimated by the book value of assets plus market value of equity minus book value of equity. PPE is the ratio of net property, plant and equipment to total assets, cash is the ratio of cash and short-term investments to total assets, leverage is the ratio of book debt to total assets, and profitability is income before extraordinary items plus deferred taxes over total assets. All Compustat variables are industry-adjusted by subtracting the respective industry mean using Fama and French (1997) 48 industries. The regressions are estimated using a (I) pooled time-series cross-sectional regression approach and a (II) Fama and MacBeth (1973) approach based on each annual cohort of events from 1985 to 2005. For the pooled regressions, the t-statistics in parentheses are calcu-lated using White (1980) heteroskedasticity-robust standard errors. For the Fama-MacBeth-regressions, the t-statistics in parentheses are calculated using Newey and West (1987) autocorrelation-correction with 1 lag. All coefficients are reported x100, the number of observations for the Fama-MacBeth approach is the number of years in the time-series.
Panel A: Gaining Rivals Panel B: Losing Rivals Method (I) Pooled (II) Fama-
MacBeth (I) Pooled (II) Fama-
MacBethSegment Dummy 0.369 5.768 2.336 0.512
(0.10) (1.30) (0.81) (0.17) Ln(Market Cap) -1.118 -0.761 0.767 1.284
(-1.16) (-0.79) (0.84) (1.22) Tobin’s Q 1.211 2.106 -0.533 -1.158
(1.80) (2.52) (-1.73) (-1.39) PPE 1.589 7.427 -4.290 1.104 (0.32) (0.99) (-0.73) (0.14) Cash 12.391 7.072 -3.554 -5.830
(1.24) (0.69) (-0.45) (-0.53) Leverage 0.332 0.423 1.103 0.949
(0.73) (0.46) (2.15) (1.82) Profitability 40.847 -11.662 17.624 -3.452 (2.80) (-0.64) (1.51) (-0.38) R² (%) 0.59 6.99 0.29 7.53 Observations 4,763 21 4,335 21
35
Following Purnanandam and Swaminathan (2004), I estimate the regressions using (I) a
pooled time-series cross-sectional approach and (II) a Fama and MacBeth (1973) re-
gression approach. For the Fama-MacBeth approach, I estimate annual cross-sectional
regressions using each annual sample of gaining and losing rivals from 1985 to 2005
and then report the time-series average of the resulting 21 slope coefficients.24
Panel A of Table 2.7 presents the results for the BHAR regressions of gaining
rivals. First, we observe a statistically significant positive coefficient on Tobin’s Q for
both regression specifications. This finding is counterintuitive as firms with a high
valuation face a lower takeover likelihood (Hasbrouck (1985) and Cremers, Nair and
John (2007)), but apparently do face a higher post-event BHAR in my sample. Second,
the coefficient on profitability is significantly positive in the pooled regression ap-
proach, indicating that more profitable firms face a higher takeover likelihood and,
therefore, a higher post-event BHAR.25 The other regression variables, the segment
dummy, size, cash, PPE and leverage, are not statistically significant in any regression
specification.26
Panel B of Table 2.7 displays the findings for the BHAR regression after large
loss events. First, we observe consistently negative coefficients on Q in both regression
specifications, with statistical significance in the pooled regression. This finding indi-
cates that highly valued rivals are more negatively impacted by the competitive effects
associated with the related M&A announcements, therefore suffering more negative
long-term returns. Second, leverage has a statistically significant positive coefficient in
both regression specifications. This does not concur with the notion that financial lever-
24 This approach accounts for the effects of cross-sectional correlation discussed in the prior section, as cross-sectional information is only used for the coefficient estimation but the tests of statistical sig-nificance solely rely on time-series information. All independent variables are winsorized at the 1% and 99% level to reduce the influence of outliers on the regression results.
25 However, the coefficient is insignificant and even reverses sign in the Fama-MacBeth regression, indicating that cross-sectional correlation due to event clustering may be responsible for spuriously significant results in the pooled regression approach.
26 The negative coefficient of size is consistent with the size group results from Section 2.4.3. The lack of statistical significance indicates that the relationship may not be linear. If I use dummy variables to replicate the size groups used in Section 2.4.3, the dummy for micro-cap gaining rivals is signifi-cantly positive.
36
age limits a firms ability to make investments to respond to competitive challenges
(Stulz (1990)), thereby resulting in more negative competitive effects of a related M&A
announcement (Akhigbe and Martin (2000)). All other variables, the segment dummy,
size, PPE, cash holdings, and profitability are not statistically significant in any regres-
sion specification.
Overall, the regressions are only modestly successful in explaining the long-term
performance of related rivals. Further research into the underlying causes of the large
rival return events, the positive information signaling and negative competitive effects
of M&A announcements, is warranted in order to better understand the underreaction of
capital markets and the ensuing return predictability.
2.5 Conclusion
In this essay, I investigate the long-term performance of rival companies related to the
target of an acquisition. Using a sample of 2,511 deals from 1985 to 2005, I document
an underreaction of capital markets to the information contained in M&A announce-
ments. Following 6,138 large rival gain events due to positive information signaling and
5,408 large rival loss events due to the negative competitive effects of the deal, I find a
return drift into the same direction as the initial reaction for up to 12 months after the
announcement. These findings are robust to the methodology I use to measure long-
term performance: 12-month BHARs using size and book-to-market matched control
firms show a highly significant return difference of 10.990% between the large gain and
large loss sample, while the calendar-time portfolio approach using the same return ad-
justment results in an average monthly long-short return of 0.638% (7.66% annualized).
Using a three-factor asset pricing model for risk adjustment in the calendar-time ap-
proach results in an average monthly three-factor alpha for the long-short portfolio of
0.726% at the 12-month horizon (8.71% annualized).
In subsample tests, I find that the return drift after positive rival return events is
most pronounced for micro-cap rivals: they experience a 12-month post-event BHAR of
8.548%, while small and big rivals experience substantially smaller and insignificant
37
return drift. This finding is intuitive as the smallest rivals face a higher takeover likeli-
hood, resulting in a higher positive return drift after the initial reaction to the M&A an-
nouncement. However, I find that rival size is not important in determining the return
drift after negative return events. I use cross-sectional BHAR regressions in order to
better understand the drivers of the return drift, but the regressions do not paint a precise
picture of the long-term BHARs to rivals. Further research into the underlying causes of
the large rival return events, positive information signaling and negative competitive
effects of M&A, is warranted in order to better understand the underreaction of capital
markets and the ensuing return predictability.
Overall, my results indicate that capital markets do not immediately incorporate
the positive information signaling and negative competitive effects of M&A announce-
ments into rival stock prices, resulting in a return drift of up to a year after the event.
38
Appendix
My accounting data are from the annual Compustat files. Book equity is defined as total
assets (data item 6), minus liabilities (181), plus balance sheet deferred taxes and in-
vestment tax credit (35) if available, minus preferred stock liquidating value (10) if
available, or redemption value (56) if available, or carrying value (130). Asset structure
is calculated as the ratio of net property, plant, and equipment (8) to total assets (6),
cash holdings as the ratio of cash and short-term investments (1) to total assets, financial
leverage as the ratio of long-term debt (9) plus debt in current liabilities (34) to total
assets, and profitability as the ratio of income before extraordinary items (18) plus in-
come statement deferred taxes (50) to total assets.
Following the usual practice in the asset pricing literature based on Fama and
French (1992), I match accounting variables for fiscal year-ends in year t 1 with CRSP
returns from July of year t to June of year t+1. The book-to-market ratio is book equity
divided by CRSP market capitalization (number of shares outstanding times share price)
at the end of December of year t 1. Tobin’s Q is the ratio of market assets to book as-
sets, where the market value of assets is total assets minus book equity plus market eq-
uity. As before, market equity is the CRSP market capitalization at the end of December
of year t 1.
39
3 Predictability of Industry Returns After M&A
Announcements
Abstract
This essay documents a strong and prevalent drift in long-term industry returns after
M&A announcements. Specifically, industries that experience positive average an-
nouncement reactions continue to do well in the future, while industries that experience
negative average announcement reactions continue to do poorly. Industry M&A invest-
ment strategies, which buy positively reacting industries and sell negatively reacting
industries, appear profitable even after controlling for size and book-to-market effects
in returns. Profitability has strengthened over time and seems to exist also for the larg-
est stocks. The evidence suggests that capital markets underreact to the industry-wide
information provided by merger announcements.
I thank my co-authors Timo Gebken and Lutz Johanning, as well as Yakov Amihud, Hermann Lo-carek-Junge (discussant), Gaston Michel, Julie Ng, Andreas Rathgeber (discussant), Dirk Schiereck, Sebastian Werner and Felix Zeidler for valuable comments and insights. I am also grateful for the input of conference participants at the 13th German Finance Association (DGF) conference in Oestrich-Winkel, the WHU Campus For Finance conference in Vallendar, the HVB doctoral seminar in Eltville, and work-shop participants at European Business School. Any remaining errors are mine.
40
3.1 Introduction
The financial literature has extensively examined the effects of M&A announcements
on the stock returns of bidding firms. Short-horizon event studies show that bidders lose
or, at most, break even when announcing a transaction. In addition, recent long-term
event studies measure significantly negative abnormal returns for the three to five years
following merger completion.27 Although the acquiring firm shareholders lose signifi-
cantly at the announcement of a transaction, studies that measure the intra-industry ef-
fects of M&A transactions show that rivals experience significantly positive abnormal
returns when the industry consolidates. Thus, positive information-signaling effects, on
average, outweigh negative competitive effects for rivals.28
In a recent study, Funke et al. (2008) show that these positive industry reactions
are not pervasive over time. At the beginning of an industry-wide consolidation process
intra-industry effects are significantly positive. However, due to increasing competition,
intra-industry effects decrease steadily and become negative at the end of such a con-
solidation process. Hence, time-varying competitive effects of M&A activity, depend-
ing on time-varying industry structure, are responsible for a significant difference in
intra-industry reactions over time. However, as industry structure and therefore the
competitive effects of M&A activity change relatively slowly over time, their evidence
suggests a certain persistence in average monthly intra-industry announcement reac-
tions.
As short horizon intra-industry effects directly impact industry returns, such a
persistence may also influence long-term industry performance. To the author’s best
knowledge, long-term industry returns after merger announcements have not yet been
27 A number of deal and firm characteristics impact bidding firms’ returns. Moeller, Schlingemann and Stulz (2004) show that only large bidders lose significantly around the transaction announcement. Loughran and Vijh (1997) document that cash bidders gain significantly in the months following a transaction announcement, although acquirers experience negative long-term returns on average.
28 Eckbo (1983) and Eckbo (1985) find positive intra-industry effects for the announcement of horizon-tal mergers. Song and Walkling (2000) document positive abnormal returns to rivals of the target for their sample of horizontal and non-horizontal deals. In more recent studies, Fee and Thomas (2004) and Shahrur (2005) also show that competitors experience positive stock market effects at the an-nouncement of horizontal deals.
41
studied. Therefore, the objective of this essay is to investigate whether there are system-
atic differences in industry returns after M&A announcements depending on the initial
intra-industry effects of the deal.
Using a dataset of 16,483 transactions that occur within one of 20 industries be-
tween 1985 and 2002 I investigate whether the average industry reaction in previous
periods affects the intra-industry effects of transaction announcements in later periods. I
find that this is indeed the case: the cumulative abnormal return (CAR) at the an-
nouncement of an M&A transaction depends on the average reaction to merger an-
nouncements in the same industry in the previous month. Thus, my findings concur with
the intuition in Funke et al. (2008) that intra-industry effects are persistent over time.
This raises an interesting question as daily abnormal announcement returns directly im-
pact monthly industry returns for the respective month. Do monthly industry returns
differ conditional on the direction of the average announcement reaction in the previous
month?
Based on this result I look at monthly industry returns after merger announce-
ments. Paralleling the results for CARs I find that returns depend on the average indus-
try announcement reaction in the previous month. Average monthly size and book-to-
market (B/M) adjusted returns are significantly positive 0.266% after months with aver-
age positive industry reactions, and significantly negative 0.454% after months with
average negative industry reactions. The difference of 0.720% is highly statistically
significant and is economically large (8.64% annualized).
Therefore, the evidence indicates a profitable investment opportunity as the in-
dustry-wide information provided by merger announcements does not seem to be incor-
porated into stock prices immediately. This pattern suggests a certain underreaction by
capital markets as past information in the form of the intra-industry effects of merger
announcements in the previous month seem to have an ability to predict the cross-
section of industry returns in the current month. Hence, I examine industry portfolio
investment strategies in order to investigate the profitability of this cross-sectional pat-
tern in industry returns.
42
Each month, I form portfolios of industries depending on the sign of the average
announcement return, and examine the profits to a zero cost investment strategy taking a
long position in all industries with positive average announcement returns and taking a
short position in all industries with negative announcement returns. The results indicate
that the effect is most pronounced for a one month holding period, where I find a statis-
tically significant return of 1.045% for raw returns (12.54% annualized) and 0.748% for
size-B/M adjusted returns (8.98% annualized). Thereafter, returns decrease steadily with
longer holding periods, though they remain statistically significant. This pattern indi-
cates that the industry return drift is most pronounced at shorter horizons.
The results hold for both the 1980s and the 1990s, while being stronger in the
latter half of the sample period. Across different size quintiles, the effect is more rele-
vant for smaller companies, although it remains statistically significant for all size-
based subsamples. Additional robustness checks confirm the pervasiveness of the effect
across different methodologies. A possible explanation for these findings may be that
the zero cost long-short industry portfolio is not free of factor-related risks, i.e., that the
significant returns represent compensation for risk inherent in those portfolios. How-
ever, when I examine the exposure of the industry portfolio returns to Fama-French
three-factor portfolios in a time-series regression, I do not find support for this hypothe-
sis. Instead, at the one month horizon I find a statistically significant intercept for the
zero cost industry portfolio not explained by exposure to factor-related risks. In a final
test I examine the relation between the cross-section of individual stock returns and av-
erage industry announcement returns. Fama-McBeth regressions show that the average
industry CAR of the previous month is significantly related to individual stock returns
of firms in that industry.
My results indicate that the industry-wide information provided by merger an-
nouncements does not seem to be incorporated into stock prices immediately, a finding
that is apparently at odds with market efficiency in the sense of Fama (1970, 1991).
However, considering the transaction costs associated with monthly portfolio rebalanc-
ing on an industry basis to capture the one month effect, the profitability of the industry
portfolio investment strategies may be called into question. Nevertheless, the effect is
43
strong and pervasive enough to be of interest for academic researchers as well as market
participants.
The remainder of this essay is organized as follows. Section 3.2 introduces the
data and describes the empirical methodology. Section 3.3 discusses the results and Sec-
tion 3.4 concludes.
3.2 Data and Methodology 3.2.1 Daily Announcement Effects
The data of mergers and acquisitions comes from the Thomson One SDC Platinum Da-
tabase.29 I consider all deals that were announced between January 1, 1985 and Decem-
ber 31, 2002 with an inflation-adjusted transaction value of at least $25 million.30 I as-
sign each transaction to one of 20 industry groups, defined as in Moskowitz and Grin-
blatt (1999), based on the bidder’s SIC code recorded by Thomson at the time of the
announcement.
Daily stock return data is from the daily CRSP files using all stocks identified as
common equity (share code 10 and 11), which excludes all certificates, ADRs, SBIs,
REITs, etc. As in previous studies intra-industry reactions are measured as the stock
market response of an equal-weighted portfolio of all firms within the industry of the
bidder.31 Following Brown and Warner (1985), I use the modified market model to es-
timate abnormal returns. I do not use the market model because, for many industries,
there are months with several merger announcements, which means a high probability
of other bid announcements occurring in the estimation period. Any abnormal returns
29 The mergers and acquisitions section (SDC Platinum) of the Thomson One Banker Database was formerly known as the Securities Data Company’s (SDC) U.S. Mergers and Acquisitions Database.
30 The $25 million cutoff is imposed for 1985 and adjusted for inflation according to the U.S. consumer price index in the following years. I can report, however, that all main results remain qualitatively and quantitatively similar without inflation adjustment.
31 SIC Codes from the CRSP file, which allow for time-series variation in industry classification, are used to assign stocks to industries according to the Moskowitz and Grinblatt (1999) two-digit classi-fication.
44
caused by these announcements will bias the estimation.32 I calculate the cumulative
abnormal return (CAR) for a five-day event window [-2,2] around the announcement
date.33 Specifically, the five-day CAR is:
2
2,,]2,2[, )(
ttmtii RRCAR , (3.1)
where Ri,t is industry i’s daily return on date t and Rm,t is the return for the equal-
weighted CRSP index on date t.
For each sample year, Table 3.1 shows the number of deals, the percentage of
the total number of deals during that specific year, as well as the average industry reac-
tion to M&A announcements. Panel A reports the results for the 22,853 deals with a
minimum transaction value of $25 million, Panel B and C display the results for the
16,483 and 11,341 deals with minimum transaction values of $50 million and $100 mil-
lion, respectively. The pattern is similar across all three panels: the two economy-wide
merger waves of the 1980s and 1990s are clearly visible with deal numbers peaking in
1988 and 1998, respectively. Annual average industry announcements effects are con-
sistent with earlier evidence such as Fee and Thomas (2004) and Shahrur (2005): a
slightly positive average around 0.05%. In addition, a certain time pattern can be ob-
served. In general, announcement effects seem to be positive, but towards the peak of
the merger waves they become negative.
32 Fuller, Netter and Stegemoller (2002) do not use the market model for the same reasons. Brown and Warner (1980) show that weighting the market return by a firm’s beta does not significantly improve the estimation.
33 The main findings are not susceptible to the choice of the event window length. They also hold for a shorter horizon, such as three days, to estimate CARs.
Tab
le 3
.1:
Ann
ual D
istr
ibut
ion
of M
&A
Tra
nsac
tions
and
Ave
rage
Indu
stry
Ann
ounc
emen
t Eff
ects
The
sam
ple
cont
ains
all
U.S
. M&
A a
nnou
ncem
ents
bet
wee
n Ja
nuar
y, 1
985
and
Dec
embe
r, 20
02 w
ith a
n in
flatio
n-ad
just
ed tr
ansa
ctio
n va
lue
of a
t lea
st $
25
mill
ion
(Pan
el A
), $5
0 m
illio
n (P
anel
B),
and
$100
mill
ion
(Pan
el C
). Fo
r eac
h ye
ar, t
he ta
ble
show
s th
e nu
mbe
r of d
eals
, the
per
cent
age
of th
e to
tal n
umbe
r of
deal
s dur
ing
that
spec
ific
year
, as w
ell a
s the
ave
rage
indu
stry
reac
tion
to M
&A
ann
ounc
emen
ts d
urin
g th
at y
ear.
The
20 In
dust
ries a
re d
efin
ed a
s in
Mos
kow
itz
and
Grin
blat
t (19
99).
CA
R is
the
cum
ulat
ive
abno
rmal
retu
rn to
the
equa
l-wei
ghte
d in
dust
ry p
ortfo
lio fo
r a fi
ve-d
ay e
vent
win
dow
[-2,
2] a
roun
d th
e an
noun
ce-
men
t dat
e. A
vera
ge In
dust
ry C
AR
den
otes
the
aver
age
CA
R to
the
indu
stry
por
tfolio
at t
he a
nnou
ncem
ent o
f dea
ls th
at ta
ke p
lace
with
in th
at in
dust
ry.
Pane
l A: T
rans
actio
n va
lue
> $
25m
n
Pane
l B: T
rans
actio
n va
lue
> $
50m
n Pa
nel C
:Tra
nsac
tion
valu
e >
$10
0mn
Yea
r N
umbe
r of
Dea
lsA
s % o
f To
tal
Num
ber
Ave
rage
In
dust
ryC
AR
(%)
Num
ber o
f D
eals
As %
of
Tota
lN
umbe
r
Ave
rage
In
dust
ryC
AR
(%)
Num
ber o
f D
eals
As %
of
Tota
lN
umbe
r
Ave
rage
In
dust
ryC
AR
(%)
1985
75
4 3.
340.
074
600
3.64
0.
084
461
4.06
0.09
5 19
86
970
4.30
0.06
877
64.
71
0.07
258
75.
170.
084
1987
93
0 4.
12-0
.034
736
4.47
-0
.034
541
4.77
-0.0
71
1988
1,
197
5.30
0.00
194
75.
75
-0.0
0269
26.
100.
001
1989
1,
040
4.61
0.00
878
34.
75
-0.0
1355
34.
87-0
.023
19
90
657
2.91
-0.0
4646
02.
79
-0.0
4227
62.
43-0
.083
19
91
579
2.56
0.01
940
42.
45
-0.0
0225
82.
27-0
.006
19
92
707
3.13
0.12
749
63.
01
0.12
331
12.
740.
089
1993
92
2 4.
080.
054
621
3.77
0.
044
385
3.39
0.05
0 19
94
1,13
8 5.
040.
121
787
4.77
0.
094
512
4.51
0.10
3 19
95
1,36
9 6.
060.
065
932
5.65
0.
067
610
5.37
0.08
0 19
96
1,70
4 7.
550.
057
1,17
57.
13
0.09
574
56.
560.
127
1997
2,
140
9.48
0.16
71,
541
9.35
0.
185
1,07
49.
460.
182
1998
2,
349
10.4
0-0
.003
1,66
910
.13
-0.0
061,
123
9.89
-0.0
10
1999
2,
011
8.90
-0.0
281,
500
9.10
-0
.044
1,07
69.
48-0
.032
20
00
1,82
4 8.
08-0
.008
1,40
08.
49
0.03
21,
005
8.85
0.02
4 20
01
1,19
6 5.
300.
064
868
5.27
0.
096
603
5.31
0.08
5 20
02
1,09
6 4.
850.
165
788
4.78
0.
162
539
4.75
0.22
0
Tota
lsa
mpl
e 22
,583
10
0.00
0.04
816
,483
100.
00
0.05
211
,351
100.
000.
053
45
46
3.2.2 Monthly Returns
For the tests of the long-run effects of M&A announcements on industry returns, I use
monthly stock return data from the monthly CRSP files. As for daily returns, industry
portfolio returns are calculated as the equal-weighted average of returns of individual
firms. In addition to raw returns, I calculate size (market capitalization) and book-to-
market equity (B/M) adjusted returns, since much research has documented the ability
of these variables to capture the cross-section of expected returns.34 For book equity, I
use data item 60 (book value of common equity) from Compustat as in Fama and
French (1992).35 I construct characteristic-matched benchmark portfolios by independ-
ently sorting stocks into size and B/M quintiles using NYSE breakpoints. I then inter-
sect those sorts to create 25 portfolios of similar size and B/M. Stocks within each in-
dustry are matched with those well-diversified portfolios based on their size and B/M
characteristic, and the equal-weighted average in excess of these size-B/M benchmarks
represents the monthly industry abnormal return.36
Table 3.2 provides a description of the industry portfolios and summary statis-
tics. The average number of stocks per industry is 290, and the fewest number of stocks
at any time in any industry except Railroads is more than 20. Therefore, virtually all
portfolios are well diversified in that they have negligible firm specific risk. Table 3.2
also reports the average monthly raw returns and size-B/M adjusted returns of the 20
industries.37 F-tests of whether these mean returns differ across industries are not re-
34 See, e.g., Banz (1981), Fama and French (1992, 1993, 1996), and Daniel and Titman (1997). 35 To avoid look ahead bias caused by back filled Compustat data, I include firms only once two com-
plete years of history are available. In addition, I exclude firms with negative book value of equity. 36 Following Fama and French (1992) portfolio compositions are changed each year in July based on a
firm’s size quintile in June of year t and its B/M quintile in December of year t-1. Stock i is then matched with one of the 25 portfolios based on its size and B/M characteristic, and the return of the matched portfolio is subtracted from stock i’s return at time t. I employ this characteristic-matched portfolio adjustment method as opposed to a Fama-French three-factor regression in order to avoid estimation issues regarding factor loadings. Furthermore, Daniel and Titman (1997) find that charac-teristics better capture cross-sectional variation in mean returns than factor loadings.
37 The overall negative average of size-B/M adjusted industry returns implies that stocks from larger industries with more stocks have outperformed stocks from smaller industries with less stocks over the sample period. Table 3.2 confirms this deduction, the four industries with positive adjusted re-turns are among the largest industries.
47
jected, which suggests that there is little cross-sectional variation in the industry sample
means. Furthermore, t-statistics for the average size-B/M adjusted returns provide little
evidence that unconditional abnormal industry returns exist per se (t-statistics not re-
ported for brevity).
Table 3.2: Industry Summary Statistics Sample statistics of the 20 industry portfolios are reported below, including the two-digit SIC codes used to form the industries as in Moskowitz and Grinblatt (1999). The industries are formed monthly from January, 1985 to December, 2002 using CRSP SIC codes, which allow for time variation in industrial classification. The average number of stocks assigned to each industry portfolio every month is reported, along with the minimum number of stocks appearing in each portfolio at any point in time (reported in parentheses). Also reported for each industry over the sample period are the average raw return, the aver-age adjusted return in excess of the size-B/M matched benchmarks, and the monthly average CAR to own-industry M&A announcements with an inflation-adjusted transaction value of at least $25m, $50m, or $100m. CAR is the cumulative abnormal return to the equal-weighted industry portfolio for a five-day event window [-2,2] around the announcement date. Industry CAR denotes the sample average of the monthly average CAR to each industry portfolio at the announcement of deals that take place within that industry. In addition, the cross-sectional averages of these statistics across industries are reported at the bottom of the table.
Industry SIC Codes
Average Number of Stocks(Minimum)
RawReturns(%)
AdjustedReturns(%)
IndustryCAR(%)$25m
IndustryCAR(%)$50m
IndustryCAR(%)$100m
1. Mining 10-14 273.5 (149) 0.942 -0.130 0.152 0.145 0.0632. Food 20 109.6 (91) 1.091 -0.086 0.031 0.019 0.0043. Apparel 22-23 83.3 (43) 0.572 -0.511 -0.043 -0.063 -0.0604. Paper 26 47.2 (33) 1.108 -0.109 -0.002 -0.045 -0.0085. Chemical 28 320.6 (199) 1.467 0.387 0.059 0.062 0.0036. Petroleum 29 26.1 (21) 1.277 -0.030 -0.036 -0.021 -0.0017. Construction 32 37.9 (23) 1.015 -0.211 -0.050 -0.038 -0.0378. Prim. Metals 33 73.0 (54) 0.766 -0.627 -0.136 -0.119 -0.0899. Fab. Metals 34 105.4 (59) 1.114 -0.185 0.009 0.001 0.05210. Machinery 35 368.5 (261) 1.165 -0.072 0.080 0.066 0.03711. Electrical Eq. 36 449.8 (375) 1.386 0.238 0.040 0.033 0.04112. Transport Eq. 37 94.2 (75) 0.874 -0.194 -0.021 -0.004 -0.00713. Manufacturing 38-39 376.1 (307) 1.274 0.161 0.181 0.175 0.14414. Railroads 40 12.0 (8) 1.226 -0.050 0.043 0.021 0.04615. Transport 41-47 102.4 (76) 1.014 -0.120 0.052 0.046 0.02416. Utilities 49 205.0 (132) 1.028 -0.176 -0.071 -0.060 -0.01117. Dept. Stores 53 39.3 (29) 0.715 -0.168 0.139 0.154 0.17918. Retail 50-52,
54-59 576.3 (429) 0.917 -0.163 -0.012 -0.030 -0.042
19. Financial 60-69 1165.7 (863) 1.318 -0.015 0.054 0.061 0.06820. Other other 1329.1 (957) 1.118 0.079 0.053 0.061 0.064 Average 289.8 (209.2) 1.069 -0.099 0.026 0.023 0.024
F-statistic (identical means) 0.296 1.124 1.382 1.380 1.064 (p-value) (0.999) (0.318) (0.124) (0.125) (0.382)
48
Finally, Table 3.2 reports average industry CARs for the three different M&A transac-
tion samples with minimum transaction values of $25m, $50m, and $100m, respec-
tively. Similar to adjusted returns there is some cross-sectional variation in returns
across industries visible, but F-tests cannot reject the hypothesis that mean CARs differ.
Hence, there is little evidence that unconditional differences between industry portfolio
returns exist for any of the measures employed.
3.3 Results
In this section I discuss my main empirical findings. Section 3.1 presents evidence on
the persistence in intra-industry effects and the ensuing predictability of monthly indus-
try returns. Section 3.2 discusses the findings related to industry portfolio investment
strategies, and Section 3.3 presents results on the predictability of individual stock re-
turns based on prior intra-industry reactions.
3.3.1 Industry Announcement Effects
3.3.1.1 Persistence in Intra-Industry Effects
Consistent with earlier evidence such as Fee and Thomas (2004) and Shahrur (2005), in
general, industry announcement effects are positive for my samples of M&A transac-
tions. However, Funke et al. (2008) show that industry reactions are only positive at the
beginning of an industry specific consolidation process, but decrease steadily as it con-
tinues and become negative at the end. This pattern is due to increasingly negative com-
petitive effects towards the end of a consolidation process. Hence, time-varying com-
petitive effects of M&A activity, depending on time-varying industry structure, are re-
sponsible for a significant difference in intra-industry reactions over time. However, as
industry structure and therefore the competitive effects of M&A activity change rela-
tively slowly over time, their evidence suggests a certain persistence in intra-industry
announcement reactions, at least at the beginning and end of such industry-specific con-
solidation processes.
49
I investigate this issue by examining the intra-industry effects of M&A an-
nouncements conditional on the average industry announcement reaction in the previous
month. Table 3.3 reports the median and mean industry CAR to M&A announcements
for three subsamples: (1) industries with a positive average CAR in the previous month,
(2) industries without M&A activity in the previous month, and (3) industries with a
negative average CAR in the previous month.
Table 3.3: Industry Cumulative Abnormal Returns to M&A Announcements The sample contains all U.S. M&A announcements between January, 1985 and December, 2002 with an inflation-adjusted transaction value of at least $25 million (Panel A), $50 million (Panel B), and $100 million (Panel C). CAR i, t denotes the cumulative abnormal return to the equal-weighted industry portfo-lio for a five-day event window [-2,2] around the announcement date. The 20 Industries are defined as in Moskowitz and Grinblatt (1999). Average Industry CARi,t-1 denotes the average CAR to the industry portfolio at the announcement of deals that take place within industry i in the previous month t-1. All returns are in percentages, median values are in brackets and absolute t-statistics for mean tests and z-values for median tests in parentheses. The difference tests are based on t-tests for equality of means and a Wilcoxon-test for equality of medians. The final row for each panel lists the number of observations.
All Average Industry CARi,t-1 > 0
(1)
No M&A Activ-ity Industryi,t-1
(2)
Average Industry CARi,t-1 < 0
(3)
Difference ((1)-(3))
Panel A: Transaction value > $25mn
CAR i, t 0.048 0.173 0.030 -0.095 0.268 (%) (5.47) (14.79) (0.65) (7.08) (15.13)
[0.035] [0.140] [-0.0124] [-0.0919] [0.232] (6.99) (18.85) (0.34) (9.48) (19.69)
N 22,583 11,541 1,015 10,027
Panel B: Transaction value > $50mn
CAR i, t 0.052 0.176 0.055 -0.097 0.274 (%) (5.05) (13.34) (1.26) (5.90) (13.10)
[0.036] [0.137] [-0.0019] [-0.0801] [0.217] (6.35) (15.73) (0.65) (7.28) (15.88)
N 16,483 8,351 1,093 7,039
Panel C: Transaction value > $100mn
CAR i, t 0.053 0.160 0.049 -0.070 0.231 (%) (4.31) (9.61) (1.16) (3.52) (8.93)
[0.033] [0.127] [-0.0113] [-0.0764] [0.203] (5.25) (11.75) (0.82) (4.73) (11.42)
N 11,351 5,517 1,089 4,745
50
The table is split into Panels A, B, and C, which report results for the minimum transac-
tion values of $25m, $50m, and $100m, respectively. I report complete results in the
tables, but limit my discussion to the $50m cut-off sample (Panel B) for expositional
simplicity and due to the similarity of the results.For comparison, the average industry
CAR for all transactions, a significantly positive 0.052%, is reported again. Across the
subsamples (1) through (3), however, the average intra-industry announcement reaction
differs completely: significantly positive average industry CARs of 0.177% can only be
observed if the average industry announcement reaction in the previous month was
positive. Otherwise, if there was no M&A activity in the previous month, the CARs are
not distinct from zero, and if M&A activity in the previous month caused a negative
intra-industry reaction on average, the CARs are significantly negative 0.097%. The
difference between the average industry CAR for subsamples (1) and (3) is a highly
significant 0.274%. The results do not depend on outliers, as the median values reported
in brackets display even more significance.
3.3.1.2 Predictability of Monthly Industry Returns
So far, the evidence clearly suggests that industry CARs depend on the average reaction
to merger announcements in the same industry in the previous month. This finding
raises an interesting question given that average daily industry CARs directly impact
monthly industry returns for that month. Do monthly industry returns differ depending
on the direction of the average announcement reaction in the previous month?
I investigate this question by examining monthly size-B/M adjusted industry re-
turns conditional on the average industry announcement reaction in the previous month.
Table 3.4 reports the mean and median size-B/M adjusted industry returns for the whole
sample as well as the three subsamples from Table 3.3, which are conditional on the
average industry CAR in the previous month. Again, the table is split into Panels A, B,
and C depending on the transaction value, but the discussion is limited to the results for
$50m in Panel B.
51
Table 3.4: Monthly Industry Size and Book-to-Market Adjusted Returns The sample contains monthly size-B/M adjusted returns of the 20 Moskowitz and Grinblatt (1999) indus-try portfolios for the period of February, 1985 to December, 2002. Size-B/M Adjusted Industry Return i, tdenotes the equal-weighted portfolio of individually size-B/M adjusted stock returns for all stocks in industry i in month t. Average Industry CARi,t-1 denotes the average CAR to the industry portfolio at the announcement of deals that take place within industry i in month t-1 with transaction values of at least $25mn (Panel A), $50mn (Panel B) and $100mn (Panel C). All returns are in percentages, median values are in brackets and absolute t-statistics for mean tests and z-values for median tests in parentheses. The difference tests are based on t-tests for equality of means and a Wilcoxon-test for equality of medians. The final row for each panel lists the number of observations.
All Average Indus-try CARi,t-1 > 0
(1)
No M&A Activ-ity Industryi,t-1
(2)
Average Indus-try CARi,t-1 < 0
(3)
Difference ((1)-(3))
Panel A: Transaction value > $25mn
-0.098 0.247 -0.061 -0.477 0.725 (2.07) (3.24) (0.61) (6.40) (6.79)
[-0.158] [0.082] [-0.199] [-0.424] [0.506]
Size-B/MAdjustedIndustryReturn i, t (%) (4.33) (1.93) (2.00) (7.22) (6.70)
N 4,300 1,584 1,158 1,558
Panel B: Transaction value > $50mn
-0.098 0.266 -0.116 -0.454 0.720 (2.07) (3.37) (1.35) (5.62) (6.38)
[-0.158] [0.101] [-0.218] [-0.382] [0.483]
Size-B/MAdjustedIndustryReturn i, t (%) (4.33) (2.31) (3.25) (6.35) (6.30)
N 4,300 1,442 1,451 1,407
Panel C: Transaction value > $100mn
-0.098 0.252 -0.142 -0.390 0.643 (2.07) (3.00) (1.89) (4.43) (5.28)
[-0.158] [0.124] [-0.254] [-0.344] [0.467]
Size-B/MAdjustedIndustryReturn i, t (%) (4.33) (2.45) (4.06) (5.27) (5.61)
N 4,300 1,248 1,827 1,225
Overall, size-B/M adjusted returns are unconditionally negative over the sample period,
minus -0.098% with a t-statistic of 2.1.38 However, conditioning on the previous
38 The reported percentage value of -0.098% differs from the value reported in Table 3.2, -0.099%, due to different averaging and a slightly different sample period: the former is the pooled average of all 4,300 monthly industry observations from February, 1985 to December, 2002, while the latter is the cross-sectional average across industries of the time-series average industry size-B/M adjusted re-turns from January, 1985 to December, 2002.
52
month’s average industry announcement reaction changes the picture completely. Mir-
roring Table 3.3, size-B/M adjusted returns are a significantly positive 0.266% follow-
ing months with positive average industry CARs, and a significantly negative 0.454%
following months with negative average industry CARs. The difference in means of
0.720% is highly statistically significant, and the medians confirm the result with
somewhat less magnitude but similar statistical significance. Finally, the difference in
means also suggests economic importance, as a monthly return of 0.720% points im-
plies an annualized return of 8.64%.
So far, my evidence indicates a profitable investment opportunity as the indus-
try-wide information provided by merger announcements does not seem to be incorpo-
rated into stock prices immediately. This pattern suggests a certain underreaction by
capital markets as past information in the form of the intra-industry effects of merger
announcements in the previous month seem to have an ability to predict the cross-
section of industry returns in the current month. Therefore, I turn to examine industry
portfolio investment strategies in the next section in order to investigate the profitability
of this cross-sectional pattern in industry returns.
3.3.2 Industry Portfolio Investment Strategies
3.3.2.1 Raw and Adjusted Returns
In this section I focus on the question of whether the apparent cross-sectional predict-
ability of industry returns by the average industry announcement effects can be ex-
ploited by sufficiently sophisticated investors. I examine this issue by investigating the
profitability of rolling industry portfolio investment strategies over different holding
periods.
Each month, I assign industries to one of three portfolios depending on the aver-
age CAR in industry i in portfolio formation month t (negative reaction, no M&A, posi-
tive reaction). In addition, I form a self-financing portfolio by taking a long position in
all industries with positive average industry CARi,t and taking a short position in all
industries with negative average industry CARi,t. Then, I calculate returns for longer
53
horizons using a standard rolling-portfolio method as in Jegadeesh and Titman (1993)
and Fama (1998) in order to avoid test statistics that are based on overlapping returns.
For example, suppose I want to look at the performance of industries with posi-
tive average industry CAR over three months. For each calendar month, I calculate
equal-weighted industry returns from individual stock returns for all industries with
positive average industry announcement reaction in the previous month. I then average
the returns for the calendar month across industries to get the return on a portfolio of
positively reacting industries. For the same calendar month, I also calculate the returns
on portfolios of all positively reacting industries from two and three months ago and
average the three resulting portfolio returns. This average tracks the calendar month
performance of an investment strategy that holds a series of portfolios formed in the last
month as well as in the previous two months. I repeat this process every month to get a
time series of returns. I follow the same steps for different horizons (one, three, six, and
12 months after the event), and for other portfolios of industries (positively reacting
industries, industries without M&A, negatively reacting industries, and the self-
financing long-short strategy).
Panel A of Table 3. reports average raw returns for these portfolios and holding
periods. Panel B shows average size-B/M adjusted returns. In addition, time-series t-
statistics are given for the self-financing long-short strategy. The results for the one-
month holding period are as expected given the evidence in Table 3.3. The long-short
strategy earns a significantly positive 1.045% (12.54% annualized) when I employ raw
returns. Using size-B/M adjusted returns leads to a significantly positive 0.748%
(8.98% annualized). For longer horizons the effect is less pronounced. Returns diminish
to 0.619% and 0.458% for the three-month horizon. For the 12-month horizon, they
decrease further to 0.300% and 0.242%, respectively. However, for all holding periods
the results remain statistically significant. Nevertheless, the decreasing pattern of aver-
age returns to the long-short strategy suggests that the apparent industry return predict-
ability is most pronounced at shorter horizons.
54
Table 3.5: Monthly Industry Portfolio Returns This table shows average monthly returns for rolling industry portfolio investment strategies over differ-ent holding periods. Each month from January, 1985 to December, 2002, industries are assigned to one of three portfolios depending on the average CAR in industry i in portfolio formation month t. Additionally, a self-financing portfolio is formed by going long all industries with positive average industry CARi,t and going short all industries with negative average industry CARi,t. Returns for longer horizons are calcu-lated using a standard rolling-portfolio method as in Jegadeesh and Titman (1993) and Fama (1998) to create a time-series of non-overlapping returns. Panel A shows average monthly raw returns and Panel B average monthly size-B/M adjusted returns for one, three, six, and 12-month holding periods. All indus-tries are weighted equally in the portfolio averages, and all months are weighted equally in the time-series average. All returns are in percentages, t-statistics are given in parentheses for the self-financing long-short strategy.
Months after portfolioformation
Average Indus-try CARi t > 0
(1)
No M&A Activ-ity Industryi,t
(2)
Average Indus-try CARi t < 0
(3)
Long-Short Strategy(1)-(3)
t-statistic
Panel A: Average monthly raw returns (%)
1 1.545 1.018 0.500 1.045 (4.92)
3 1.327 1.060 0.707 0.619 (3.54)
6 1.211 1.040 0.795 0.416 (2.93)
12 1.145 0.999 0.845 0.300 (2.78)
Panel B: Average monthly size-B/M adjusted returns (%)
1 0.285 -0.153 -0.462 0.747 (4.42)
3 0.128 -0.094 -0.329 0.458 (3.42)
6 0.063 -0.114 -0.260 0.324 (3.05)
12 0.015 -0.140 -0.228 0.242 (3.03)
Examining both legs of the long-short strategy, we can conclude that both sides seem to
contribute equally to the profitability of the long-short strategy for raw returns. Com-
pared to the portfolio without M&A activity, industries with positive average CARs in
the previous month earn half a percent more at the one month horizon. Portfolios with
negative average CARs earn half a percent less. For the case of size-B/M adjusted re-
turns we can meaningfully look at the statistical significance of the different portfolio
returns. Adjusted returns for industries with positive average CARs in the previous
month are only significantly positive at the one month horizon. However, adjusted re-
turns for industries with negative average CARs in the previous month are significantly
negative up to the 12-month holding period (t-statistics not reported for brevity). This
result suggests that the short leg of the industry portfolio investment strategy is more
55
important, which indicates that some part of the apparent underreaction by capital mar-
kets to the industry-wide information provided by merger announcements may be due to
market frictions, such as short-selling constraints.
3.3.2.2 Subperiod Analysis
To check whether the results hold across different time periods, I split the sample into
two subperiods. Table 3.6 shows size-B/M adjusted returns for the different industry
portfolios and holding periods for 1985-1993 (Panel A) and for 1994-2002 (Panel B).
Table 3.6: Monthly Industry Portfolio Returns for Time-Period Subsamples This table shows average monthly size-B/M adjusted returns for rolling industry portfolio investment strategies over different holding periods for two time-period subsamples. Panel A reports results for Janu-ary, 1985 to December, 1993. Panel B shows the performance of the portfolios for January, 1994 to De-cember 2004. Each month, industries are assigned to one of three portfolios depending on the average CAR in industry i in portfolio formation month t. In addition, a self-financing portfolio is formed by go-ing long all industries with positive average industry CARi,t and going short all industries with negative average industry CARi,t. Returns for longer horizons are calculated using a standard rolling-portfolio method as in as Jegadeesh and Titman (1993) and Fama (1998). Panel A shows average monthly raw returns and Panel B average monthly size-B/M adjusted returns for one, three, six, and 12-month holding periods. All industries are weighted equally in the portfolio averages, and all months are weighted equally in the time-series average. All returns are in percentages, t-statistics are given for the self-financing long-short strategy.
Months after portfolioformation
Average Indus-try CARi t > 0
(1)
No M&A Activ-ity Industryi,t
(2)
Average Indus-try CARi t < 0
(3)
Long-Short Strategy(1)-(3)
t-statistic
Panel A: Average monthly size-B/M adjusted returns (%) for subperiod 1985-1993
1 0.256 0.135 -0.273 0.529 (3.46)
3 0.156 0.147 -0.200 0.356 (3.96)
6 0.067 0.135 -0.097 0.164 (2.38)
12 0.039 0.099 -0.073 0.112 (2.12)
Panel B: Average monthly size-B/M adjusted returns (%) for subperiod 1994-2002
1 0.329 -0.449 -0.646 0.975 (3.22)
3 0.107 -0.337 -0.469 0.576 (2.25)
6 0.054 -0.355 -0.431 0.485 (2.35)
12 -0.005 -0.366 -0.379 0.374 (2.33)
56
In terms of time pattern, it is interesting to observe that the profitability of the industry
portfolio investment strategies seems to have strengthened over time. For the first pe-
riod I find a statistically significant size-B/M adjusted return of 0.529% (6.35% annual-
ized) for the long-short strategy, which increases to 0.975% (11.70% annualized) for the
latter period. This may be due to the increasing importance of M&A for corporate
strategies in the 1990s.
3.3.2.3 Size Groups
I employ equal-weighting for monthly returns to be in concurrence with my measure of
short-term intra-industry effects of M&A announcements, the average industry CAR.
This raises the question of whether my findings are only relevant for small, illiquid
stocks, or whether the effect can also be found in large, liquid stocks which are more
accessible for the general investor. I investigate this issue by dividing the sample into
size quintiles.
I divide stocks at the end of each June based on month-end market capitalization
using within-industry NYSE break points. Then, I calculate equal-weighted industry
portfolios from the stocks in the respective quintile. I use within-industry breakpoints to
mitigate the problems associated with splitting industries with a few number of stocks,
such as Petroleum or Railroads, into quintiles.
Table 3.7 shows how the industry portfolio investment strategies perform across
the different size quintiles. There are several patterns to note when comparing the raw
returns to the results reported in Table 3.5. Firstly, the effect indeed seems to be
stronger for smaller stocks. The long-short strategy earns a highly significant return of
1.342% (16.10% annualized) in the smallest quintile at the one-month horizon (Panel
A). It decreases almost monotonically to 0.464% (5.57% annualized) for the largest
quintile (Panel E), but remains statistically significant at the 5% level of confidence.
Secondly, the results confirm that the effect is most pronounced at the shorter horizons,
average raw returns for the long-short strategy at the three, six, and 12-month horizons
are decreasing monotonically, and lose their statistical significance for the largest quin-
tiles.
57
Table 3.7: Monthly Industry Portfolio Returns for Size-Quintiles This table shows average monthly raw returns for rolling industry portfolio investment strategies over different holding periods for five size quintiles. Stocks are assigned each year to size quintiles based on within-industry NYSE break points for market capitalization as of December of year t-1. Each month from January, 1985 to December, 2002, industries are assigned to one of three portfolios depending on the average CAR in industry i in portfolio formation month t. In addition, a self-financing portfolio is formed by going long all industries with positive average industry CARi,t and going short all industries with negative average industry CARi,t. Returns for longer horizons are calculated using a standard rolling-portfolio method as in as Jegadeesh and Titman (1993) and Fama (1998). Panels A through E show aver-age monthly raw returns for one, three, six, and 12 month holding periods in ascending order for smallest quintiles to largest. All industries are weighted equally in the portfolio averages, and all months are weighted equally in the time-series average. All returns are in percentages, t-statistics are given for the self-financing long-short strategy.
Months after portfolioformation
Average Indus-try CARi t > 0
(1)
No M&A Activ-ity Industryi,t
(2)
Average Indus-try CARi t < 0
(3)
Long-Short Strategy(1)-(3)
t-statistic
Panel A: Average monthly raw returns (%) for quintile 1 (smallest)
1 1.729 1.039 0.387 1.342 (5.74)
3 1.459 1.039 0.669 0.789 (4.07)
6 1.299 1.051 0.799 0.500 (3.24)
12 1.250 1.015 0.890 0.360 (3.28)
Panel B: Average monthly raw returns (%) for quintile 2
1 1.263 1.112 0.365 0.898 (3.77)
3 1.121 1.066 0.593 0.528 (3.12)
6 1.011 0.960 0.662 0.349 (2.74)
12 0.917 0.880 0.673 0.244 (2.49)
Panel C: Average monthly raw returns (%) for quintile 3
1 1.481 0.994 0.525 0.956 (3.95)
3 1.201 0.973 0.783 0.418 (2.21)
6 1.100 0.942 0.841 0.259 (1.78)
12 1.040 0.891 0.820 0.220 (1.99)
Panel D: Average monthly raw returns (%) for quintile 4
1 1.400 1.154 0.854 0.547 (2.61)
3 1.239 1.284 0.948 0.292 (1.89)
6 1.176 1.271 0.964 0.212 (1.75)
12 1.142 1.176 0.975 0.166 (1.78)
58
Table 3.7 (Continued): Monthly Industry Portfolio Returns for Size-Quintiles
Months after portfolioformation
Average Indus-try CARi t > 0
(1)
No M&A Activ-ity Industryi,t
(2)
Average Indus- try CARi t < 0
(3)
Long-Short Strategy(1)-(3)
t-statistic
Panel E: Average monthly raw returns (%) for quintile 5 (largest)
1 1.329 1.202 0.865 0.464 (2.21)
3 1.249 1.235 0.964 0.285 (1.78)
6 1.192 1.201 0.982 0.209 (1.57)
12 1.131 1.130 0.954 0.178 (1.66)
3.3.2.4 Liquidity Effects
The apparent profitability of the rolling industry portfolio investment strategy, espe-
cially in the smaller size quintiles, raises the question whether it can be explained by
liquidity effects. One way to investigate how liquidity affects the industry portfolio in-
vestment strategies is to exclude those stocks that have high transaction costs. I follow
Chan (2003) and eliminate all stocks with prices of $5 or less from the sample before I
calculate industry returns. Dropping low-priced stocks considerably reduces the number
of stocks underlying the industry portfolios. On average, 30 percent of observations
(stocks in all months) are priced at $5 or below. The remaining sample should consist of
more liquid stocks, since price is related to ease of buying or selling.
Panel A of Table 3.8 reports size-B/M adjusted returns for the rolling industry
portfolio investment strategies based on the reduced set of stocks. The results are simi-
lar to those of Table 3.5. However, the long-short strategy is slightly less profitable with
a return of 0.614% (7.37% annualized) at the one-month horizon. The statistical signifi-
cance, though, is still considerably high, which indicates that the effects of liquidity can
only account for a small portion of the profitability of the industry portfolio investment
strategies.
59
Table 3.8: Robustness of Monthly Industry Portfolio Returns This table shows average monthly returns for rolling industry portfolio investment strategies over differ-ent holding periods. Each month from January, 1985 to December, 2002, industries are assigned to one of three portfolios depending on the average CAR in industry i in portfolio formation month t. In addition, a self-financing portfolio is formed by going long industries with positive average industry CARi,t and going short industries with negative average industry CARi,t. Returns for longer horizons are calculated using a standard rolling-portfolio method as in as Jegadeesh and Titman (1993) and Fama (1998), and average monthly size-B/M adjusted returns for one, three, six, and 12-month holding periods are reported. All industries are weighted equally in the portfolio averages, and all months are weighted equally in the time-series average. Panel A excludes stocks priced below 5$ before forming industries. Panel B reports returns if only two industries are included each month in the positive and negative average CAR i, t portfo-lios, respectively. These two industries are selected by ranking by formation month average industry CAR i, t and selecting the two portfolios each at the top and bottom of the ranking. The middle portfolio still includes all industries without M&A activity. All returns are in percentages, t-statistics are given for the self-financing long-short strategy.
Months after portfolioformation
Average Indus-try CARi t > 0
(1)
No M&A Activ-ity Industryi,t
(2)
Average Indus-try CARi t < 0
(3)
Long-Short Strategy(1)-(3)
t-statistic
Panel A: Average monthly size-B/M adjusted returns (%) excluding stocks priced < 5$
1 0.223 -0.152 -0.391 0.614 (4.09)
3 0.094 -0.124 -0.280 0.374 (3.27)
6 0.035 -0.143 -0.204 0.240 (2.70)
12 0.011 -0.170 -0.190 0.201 (2.91)
Panel B: Average monthly size-B/M adjusted returns (%) including only two industries on each side
1 0.346 -0.153 -0.535 0.881 (3.18)
3 0.105 -0.094 -0.462 0.567 (2.79)
6 0.094 -0.114 -0.334 0.428 (2.63)
12 -0.011 -0.140 -0.321 0.311 (2.69)
3.3.2.5 Strength of Intra-Industry Effects
Another issue regarding the industry investment strategy is the number of industries
invested in. For all results reported so far, the strategies take a long position in all indus-
tries with positive average industry CAR in the previous month, and a short position in
all industries with negative average industry CAR in the previous month. This corre-
sponds, according to the median, to six industries on each side of the long-short portfo-
lio, with a minimum investment in two industries and a maximum investment in 12 in-
60
dustries.39 Judging from the standard deviation of the number of industries per portfolio,
which is slightly larger than two, there is substantial variation in what constitutes the
long and short portfolio. In addition, there is substantial variation in average industry
announcement reactions, i.e., in the factor which determines which portfolio an industry
enters. Average industry CARs range from values only marginally different from zero
to values reliably greater or smaller than zero.
Hence, the question arises whether it is possible to increase the profitability of
the rolling industry portfolio investment strategies by only considering a fixed number
of industries for each leg of the trading strategy, and by only considering those indus-
tries with announcement reactions reliably different from zero. I investigate this issue
by only investing in two industries for the long and short portfolio, respectively,40 and
by selecting these two industries from the top and bottom, respectively, of a formation
month ranking on average industry CARs. Panel B of Table 3.8 reports size-B/M ad-
justed returns for these alternative rolling industry investment portfolios.41 Indeed, the
profitability of the long-short strategy improves for all holding periods. At the one-
month horizon, returns increase from 0.748% in Table 3.5 to 0.881% (10.57% annual-
ized), and at the 12-month horizon they increase from 0.242% to 0.311% (3.73% annu-
alized). The majority of the improvement in profitability seems to come from the short
portfolio, while returns to the long portfolio only change considerably at the one-month
horizon. However, the t-statistics for the long-short strategy are lower across all hori-
zons compared to Table 3.5, which indicates that only investing in two portfolios on
each side of the investment strategy increases the time-series variation in the adjusted
returns considerably.
39 On average, the long portfolio consists of 6.71 industries, the portfolio without M&A contains 6.75 industries, and the short portfolio holds 6.54 industries. The standard deviations of the number of in-dustries are 2.20, 2.37, and 2.11, respectively.
40 I select two industries, as it conveniently constitutes not only the top and bottom deciles, but also the minimum number of industries in each portfolio. Results are quantitatively similar when I select a fixed number of three industries, allowing exceptions for months where there are only two industries available.
41 The industry portfolio without M&A activity is unchanged and therefore its returns are identical to Tale 3.5.
61
3.3.2.6 Three-Factor Risks
In the analysis of size-B/M adjusted returns I already account for the two most impor-
tant factors driving cross-sectional differences in stock returns: size and book-to-market
characteristics. I still find statistically significant returns to the long-short industry strat-
egy unexplained by those characteristics. However, to investigate an alternative choice
of methodology relying on risk factor loadings instead of characteristic-matched returns,
Table 3.9 reports month-by-month alphas and three-factor loadings for the industry
portfolios. These loadings are from a time-series regression of portfolio excess returns
on contemporaneous Fama-French factors in calendar time. Specifically, I regress:
tptptptftmpptftp HMLhSMBsRRbaRR ,,,,, )()()( , (3.2)
where Rp,t is portfolio p’s return t months after formation, for t=(1,3,6,12), and bp, sp,
and hp are the coefficients on the Fama and French (1993) market, size, and book-to-
market portfolio returns from the same calendar months.42 Note that the portfolio re-
turns are unadjusted and non-rolling. I conduct these time-series regressions for the
long-short industry portfolio (Panel A), as well as the individual portfolios with a long
position in all industries with positive average industry CAR (Panel B) and a short posi-
tion in all industries with negative average industry CAR (Panel C). t-statistics are cal-
culated from White (1980) heteroskedasticity-robust standard errors.43
Month-by-month alphas confirm the evidence presented before. The long-short
strategy has a significantly positive intercept of 0.719% one month after portfolio for-
mation. This is due to a significantly positive intercept of 0.537% for the portfolio tak-
ing a long position in all industries with positive average industry CAR, and a signifi-
cantly negative 0.611% for the portfolio taking a short position in all industries with
negative average industry CAR in the previous month.
42 The construction of these portfolio returns is discussed in detail in Fama and French (1993). I thank Kenneth French for providing monthly time-series data for the factors on his website.
43 t-statistics are qualitatively unchanged if I use Newey and West (1987) autocorrelation- and het-eroskedasticity-corrected standard errors at various lags instead of the White (1980) correction.
62
tppptftmpptftp HMLhSMBsRRbaRR ,,,,, )()()(
Table 3.9: Industry Portfolio Three-Factor Alphas This table shows month-by-month alphas, coefficients, t-statistics, and R² values from a monthly time-series regression where the left-hand side variable is monthly excess returns for a portfolio formed t month ago and the right-hand side variables are contemporaneous Fama and French (1993) market, size, and book-to-market factor mimicking portfolio returns. Industry portfolios are formed each month from January, 1985 until December 2002 by assigning industries to one of three portfolios depending on the average CAR in industry i in portfolio formation month t. In addition, a self-financing portfolio is formed by going long all industries with positive average industry CARi,t and going short all industries with negative average industry CARi,t. Panel A reports the regression results for these long-short industry portfolios, while Panel B presents results for the positive average industry CAR portfolios and Panel C displays the findings for negative average industry CAR portfolios. t-Statistics are based on White (1980) heteroskedasticity-robust standard errors.
Post-formation month Pa (%) Pb (Rm-Rf) Ps (SMB) Ph (HML) R² (%) N
Panel A: Long-Short Industry Portfolio
1 0.719 -0.108 0.182 -0.094 7.24 215 (3.06) (-2.07) (1.24) (-0.81)
3 0.017 -0.063 0.328 0.000 13.86 213 (0.07) (-1.20) (2.44) (-0.00)
6 -0.074 -0.087 0.269 -0.058 13.66 210 (-0.39) (-2.02) (2.98) (-0.53)
12 -0.001 0.020 0.117 -0.176 9.72 204 (-0.01) (0.38) (1.25) (-1.60)
Panel B: Positive Industry CAR Portfolio
1 0.537 0.927 0.830 0.309 84.29 215 (3.32) (23.84) (9.57) (4.46)
3 0.194 0.964 0.926 0.342 84.58 213 (1.19) (23.63) (9.98) (3.90)
6 0.030 0.976 0.853 0.383 85.99 210 (0.21) (22.51) (11.39) (4.76)
12 0.192 1.008 0.776 0.273 76.70 204 (0.92) (20.12) (6.96) (2.80)
Panel C: Negative Industry CAR Portfolio
1 -0.611 1.036 0.656 0.411 78.97 215 (-3.45) (22.24) (5.89) (4.06)
3 -0.250 1.028 0.607 0.350 79.95 213 (-1.48) (24.17) (6.09) (3.39)
6 -0.320 1.064 0.593 0.449 79.87 210 (-1.74) (30.51) (6.29) (5.10)
12 -0.226 0.990 0.668 0.457 83.15 204 (-1.47) (26.95) (11.32) (6.31)
63
In contrast to the next-month regressions, all month-by-month alphas for three, six and
12 months after portfolio formation are not significantly different from zero at conven-
tional levels of confidence.44 This confirms the earlier conclusion that the industry re-
turn predictability by intra-industry announcement effects is most pronounced at the
shorter horizon.
3.3.3 The Cross-Section of Individual Stock Returns
The main objective of this essay relates to the study of long-term industry returns after
M&A announcements, depending on the initial intra-industry effects of the announce-
ment. I find that the short-term announcement reaction has an ability to predict the
cross-section of future industry returns in the sense that a self-financing long-short
strategy buying industries with positive average industry CARs and selling industries
with negative average industry CARs yields significantly positive profits not accounted
for by known risk factors. But as industry returns are calculated directly from individual
stock returns, an alternative question arises: does the average announcement reaction to
M&A have any predictive power for the cross-section of expected individual stock re-
turns? I investigate this issue as an additional robustness check on the results with an
alternative methodology.
Specifically, I use Fama and MacBeth (1973) cross-sectional regressions to de-
termine the impact of average industry announcement CARs controlling for factors
known to predict stock returns. Each month from 1985 to 2002, I regress the cross-
section of individual stock returns over one, three, six and 12-month horizons on a con-
stant and various past firm characteristics. I then average the coefficients across months,
and calculate standard errors from the time-series of coefficients. In regressions where
individual stock returns over several months are employed as dependent variable, I use
Newey and West (1987) autocorrelation-robust standard errors with a lag equal to the
number of months over which the returns overlap. The set of independent variables in-
44 The longer-horizon evidence is most consistent for the short portfolio, where all intercepts are nega-tive and the six-month intercept is significant at the 10 percent level of confidence.
64
cludes: market ,45 size (log of market capitalization of month t-1), book-to-market (log
of book-to-market ratio calculated using past data as in Fama and French (1992)), sev-
eral past individual stock return variables (last month’s return, ret-1:-1, last years return
from t-12 to t-2, ret-12:-2, as well as the return from t-36 to t-13, ret-36:-13), as well as last
month’s industry announcement reaction (average industry CAR of month t-1).
The results on the control variables reported in Table 3.10 confirm previous
findings in the literature. The use of market , size and book-to-market as regressors
allows to capture their ability, if any, to explain the cross-section of expected stock re-
turns. The one-month past return controls for liquidity and microstructure effects docu-
mented by Jegadeesh (1990) which induce a reversal in short-term stock returns. The
nearest year of returns captures the Jegadeesh and Titman (1993) momentum effect, and
the long-run return the De Bondt and Thaler (1985) three- to five-year reversal effect.
However, for my purposes the coefficient on the average industry CAR in month t-1 is
of main interest: at the one-month horizon is positive and highly statistically significant,
with a t-statistic of over seven. For the regressions with longer-horizon returns the pre-
dictive ability of the average industry CAR is somewhat less reliable as t-statistics de-
crease substantially, but the influence remains positive and the coefficients are still sig-
nificant, at least at the 5 percent level. This ability of the past month’s average industry
CAR to predict the cross-section of individual stock returns confirms the earlier evi-
dence on differences in long-term industry returns after observing the initial intra-
industry effects and provides strong evidence affirming the robustness of the results.
45 Market s are estimated by regressing the prior 36 months of excess returns for each stock on a con-stant and the past 36 months of excess returns of the CRSP value-weighted index. Stocks are then ranked based on their coefficient estimates from this regression (pre-ranking betas) and assigned to one of 100 groups based on this ranking. Stocks within a particular beta group are assigned the equal-weighted average beta for that group. This is essentially the procedure employed by Moskowitz and Grinblatt (1999) in their Table VI.
65
Table 3.10: Fama/MacBeth Regressions for Individual Stock Returns Fama and MacBeth (1973) cross-sectional regressions are run every month on the universe of securities from February 1985 to December 2002. Specifically, the cross-section of individual stock returns over one, three, six, and 12-month horizons are regressed on a constant and a host of firm characteristics: mar-ket (estimated using the prior 36 months of returns), size (log of market capitalization at month t-1), B/M (log of book-to-market ratio calculated using past data as in Fama and French (1992)), several past return variables, as well as the average industry CAR of month t-1. Time series t-statistics are in brackets, below each average coefficient. Newey and West (1987) heteroskedasticity and autocorrelation-robust standard errors are used in all calculations for the three, six, nine and 12-month regressions.
Hori-zon Intercept ˆ ln(Size) ln(B/M) ret-1:-1 ret-12:-2 ret-36:-13
Average IndustryCARt-1
AverageR² (%)
1 0.0228 0.0008 -0.0009 0.0000 -0.0661 0.0310 -0.1588 0.5036 5.34(3.05) (0.48) (1.46) (0.02) (12.02) (1.25) (4.75) (7.07)
3 0.0684 0.0017 -0.0026 -0.0006 -0.0061 0.0485 -0.4816 0.4917 5.56(3.27) (0.39) (1.64) (0.19) (0.88) (0.90) (4.20) (2.54)
6 0.1406 0.0041 -0.0054 -0.0010 0.0096 -0.0491 -0.9129 0.7572 5.79(4.06) (0.53) (2.00) (0.17) (0.88) (0.43) (4.11) (2.15)
12 0.2154 0.0091 -0.0081 -0.0020 0.0164 -0.2538 -1.3609 0.9855 5.90(4.59) (0.80) (2.15) (0.21) (1.16) (1.54) (3.98) (2.39)
3.4 Conclusion
Most studies that measure the intra-industry effects of M&A transactions show that ri-
vals experience significantly positive abnormal returns when the industry consolidates.
In a recent study, however, Funke et al. (2008) show that these positive industry reac-
tions are not pervasive over time. At the beginning of an industry-wide consolidation
process intra-industry effects are significantly positive. However, due to increasing
competition, they decrease steadily and become negative at the end of such a consolida-
tion process. Hence, time-varying competitive effects of M&A activity, depending on
time-varying industry structure, are responsible for a significant difference in intra-
industry reactions over time. However, as industry structure and therefore the competi-
tive effects of M&A activity change relatively slowly over time, this suggests a certain
persistence in intra-industry announcement reactions.
66
I confirm this implication by documenting that industries which experience posi-
tive average announcement reactions continue to do well at future merger announce-
ments, while industries that experience negative average announcement reactions con-
tinue to do poorly. These differences in short-term announcement reactions directly im-
pact monthly industry returns, resulting in a difference in monthly returns of 0.720%
(8.64%) between industry groups depending on the direction of the previous month’s
average industry reaction. Industry portfolio investment strategies, which buy positively
reacting industries and sell negatively reacting industries, appear profitable even after
controlling for size and book-to-market effects in returns. Profitability has strengthened
over time and seems to exist also for the largest stocks. The findings cannot be ex-
plained by liquidity effects or exposure to three-factor risks. Additional robustness
checks with Fama-McBeth regressions confirm the ability of the past month’s average
industry announcement reaction to predict the cross-section of stock returns.
In this essay I document a strong and prevalent drift in long-term industry re-
turns after M&A announcements. The evidence suggests that capital markets underreact
to the industry-wide information provided by merger announcements.
67
4 Predictability of Supplier Returns After Large Customer
Price Changes
Abstract
This essay documents return predictability across stocks, specifically from customers to
their suppliers. I extend the investigation of Cohen and Frazzini (2006), who examine
monthly return predictability for economically linked firms, to analysis at the daily level
and show that for large positive (negative) customer price change events supplier stock
prices experience significantly positive (negative) cumulative abnormal returns for up
to 20 days after the event. However, the major part of these returns arises in the first
five days and such return predictability cannot be observed for the largest stocks and
the most recent time period, indicating that capital markets are relatively efficient in
incorporating extreme customer return shocks into supplier stock prices. Hence, limited
investor attention to information on economically linked firms seems to be less severe
for the attention-grabbing events investigated in this essay.
I thank my co-authors Timo Gebken and Lutz Johanning, as well as Yakov Amihud, Julie Ng, Gas-ton Michel, Dirk Schiereck, Sebastian Werner, and Felix Zeidler for valuable comments and insights. Any remaining errors are mine.
68
4.1 Introduction
Limited investor attention (investor inattention) and its consequences for information
processing in capital markets have recently generated a lot of interest in the asset pricing
literature.46 In a new paper Cohen and Frazzini (2006) examine economically linked
firms and document widespread cross-asset return predictability in U.S. capital markets.
They conclude that investors do not take into account ex-ante available and often long-
standing customer-supplier relationships due to their limited attention, so that prices of
supplier firms have a predictable lag in updating to new information about its customer
firms.
I examine this apparently wide-spread return predictability across assets further.
Specifically, my interest lies in the short-run return dynamics at the daily horizon not
examined by Cohen and Frazzini (2006) who base their study of supplier-customer rela-
tionships on monthly returns. I want to know more precisely for how long return pre-
dictability from customer to supplier firms can be observed, and to which extent this
return predictability is also visible in the extremes of the return distribution.
In the light of this, I focus on extreme daily customer price changes, i.e., events
which should grab investor’s attention due to related underlying news or by becoming
news themselves. In this situation, I want to know whether we can still observe supplier
return predictability due to limited investor attention to the consequences of this cus-
tomer stock return information for supplier prices. Alternatively, I hypothesize that
given such attention-grabbing customer events, limited attention is less of an issue for
the related supplier firms and cross-asset return predictability could be relatively short-
lived.
My results show that this is the case. I use publicly available customer informa-
tion from firms’ financial statements to build my database on customer-supplier rela-
46 Recent empirical works on limited attention, among others, include Hong, Lim and Stein (2000), Hou and Moskowitz (2005), Hou, Peng and Xiong (2006), Menzly and Ozbas (2006), Cohen and Frazzini (2006), Hou (2007) and Hong, Torous and Valkanov (2007). In addition, Gervais, Kaniel and Mingelgrin (2001), Barber and Odean (2006), and Frazzini and Lamont (2006) examine shifts in in-vestor attention and their price effects. See the literature review in Section 4.2.
69
tionships.47 Using an event study framework and daily stock returns from 1981 to 2004,
I investigate abnormal supplier stock returns after large abnormal customer price
changes, i.e., a daily customer abnormal stock return three standard deviations or more
from the mean. I show that after such large customer price changes there is indeed some
supplier return predictability: after large positive events supplier 20-day cumulative
abnormal returns (CAR) are statistically significant positive 0.569% and after large
negative events supplier 20-day CAR amount to statistically significant negative
0.372%.
However, this supplier return predictability mainly stems from the first week af-
ter the event: for positive events about half of the 20-day CAR arises over the [t+1,t+5]
window, and for negative events return predictability is completely attributable to these
five trading days. This finding is interesting given the somewhat contrary evidence in
Cohen and Frazzini (2006) who document a drift between economically linked firms at
the monthly horizon. However, Cohen and Frazzini (2006) examine differences between
quintile portfolios, while I look farther into the tails of the distribution examining ex-
treme return events. I surmise that my different finding arises as capital markets incor-
porate large abnormal customer price changes more quickly into supplier stock prices.
Limited investor attention may be grabbed by such large price changes and shifted rela-
tively quickly to the importance of this customer information for supplier stock prices.
To shed more light on this finding I investigate the best proxy for limited inves-
tor attention in my research setting: the initial supplier reaction to the large abnormal
customer price change. I show that if there is a large positive (negative) contemporane-
ous supplier reaction to a positive (negative) large customer price change, markets seem
to have already incorporated all information immediately and no post-event supplier
drift can be discerned. However, if there is no reaction or even a negative (positive)
reaction to a positive (negative) large customer price change, I document a drift in sup-
plier stock prices in the same direction as the customer event. Sometimes, investors
47 Currently, FASB regulation SFAS No. 131 governs segment reporting in the U.S. and requires nam-ing customers which account for 10 percent or more of a firm’s sales.
70
seem to shift their limited attention not immediately to this customer stock return infor-
mation but incorporate it only slowly into supplier prices.
I investigate the robustness of my results in several respects. First, I show that
my findings are not due to industry effects, given the findings of Hou and Moskowitz
(2005) and Hou (2007) who demonstrate that lead-lag effects among stocks are primar-
ily an industry phenomenon. I show that supplier return predictability also exists across
industries and when excluding the smallest supplier decile. When I examine the impor-
tance of size, I find that supplier return predictability disappears in the largest supplier
size tertile, consistent with the notion that limited investor attention is a more important
issue for smaller supplier firms. This finding is intuitive given the greater news cover-
age larger firms receive, and taking into account prior evidence on the importance of
firm size for return predictability. Hence, the effects of limited attention for asset prices
are more important for smaller firms. In addition, I examine the stability of the return
predictability over time and show that, at least for large negative customer price
changes, supplier return predictability has completely disappeared in the more recent
part of my sample period.
Overall, my finding of only very short-run supplier return predictability for the
tails of the customer return distribution extends the existing literature on limited inves-
tor attention. We know from empirical evidence that there exists a widespread return
predictability across assets in capital markets (Menzly and Ozbas (2006), Cohen and
Frazzini (2006), Hong, Torous and Valkanov (2007)) at odds with market efficiency in
the sense of Fama (1970, 1991), most likely due to limited investor attention and capac-
ity constraints in information processing. However, investor attention is only ‘partially
limited’ when examining the large, attention-grabbing customer price changes investi-
gated in this essay: return predictability mainly occurs in the first week after the event,
does not apply to the largest supplier stocks, and has disappeared, at least for negative
events, in the more recent past.
The remainder of this essay is organized as follows. Section 4.2 gives an over-
view of the related literature on the effects of limited investor attention on asset prices.
71
Section 4.3 introduces the customer-supplier data and describes the research methodol-
ogy. Section 4.4 discusses the empirical results and Section 4.5 concludes.
4.2 Related Literature
Limited investor attention is a phenomenon which has recently generated a lot of re-
search interest in the asset pricing literature. The psychology literature acknowledged
that attention is a scarce cognitive resource as early as Kahneman (1973), who proposed
a psychological model of divided attention which is based on the idea of mental efforts.
In his model, cognitive resources are allocated accommodating differing mental re-
quirements by different tasks, and attention to one specific task requires the substitution
of mental resources from another task.48 In the asset pricing literature, Merton (1987)
provides the first rigorous theoretical approach to limited investor attention with a
model where investors restrict their information gathering and trading to a small subset
of all available stocks. As a result, given asymmetry in the amount of information about
any stock, those with less information available are less recognized by investors, have a
smaller investor base, and trade at a greater discount due to the inability of their inves-
tors to appropriately share the risks of these stocks (neglected stocks).
More recently, Hong and Stein (1999) propose a model with two boundedly ra-
tional investor types in which information diffuses slowly across the investment public
and in which return predictability ensues from investors failing to rationally extract all
available information from market prices. Hirshleifer and Teoh (2003) model firms’
choices between alternative means of presenting financial information in accounting
reports and the resulting effects on market prices when investors have limited attention
and processing power. Peng (2005) provides a model analyzing the effects of capacity
constraints in the learning processes of investors. Finally, Peng and Xiong (2006) de-
velop a model investigating investors’ attention allocation, their ensuing learning behav-
ior given limited attention, and the resulting effects on asset price dynamics.
48 For summaries of the psychology literature see also Nisbett and Ross (1980) and Pashler (1998).
72
Building on these theoretical models, a growing body of literature is examining
empirical evidence on limited investor attention. Huberman and Regev (2001) provide a
particularly obvious case in point: a firm’s stock price soaring in the wake of a New
York Times article on a potential development of new cancer-curing drugs. However,
this potential breakthrough in cancer research had been reported in the journal Nature
and in various popular newspapers (including The Times) more than five months earlier.
In a broader empirical investigation, Hou and Moskowitz (2005) study the severity of
market frictions affecting a stock using the delay with which its price responds to new
information. They show that frictions associated with investor recognition appear to be
related to this delay effect. Hou (2007) finds that slow diffusion of common information
is the main reason for such lead-lag effects and that it is mainly an intra-industry phe-
nomenon: big firm returns lead small firm returns in the same industry.
Hong, Lim and Stein (2000) investigate momentum in stock returns to test the
gradual-information-diffusion model of Hong and Stein (1999) and find that firm-
specific information, especially negative information, diffuses only slowly across inves-
tors. Hou, Peng and Xiong (2006) examine the implications of investor attention for
both price momentum and earnings momentum (post-earnings announcement drift).
They show that variation in trading volume, which they use as a proxy for attention,
causes significant variation in abnormal returns to both price momentum and earnings
momentum.49
Besides these works on the effects of limited attention, there is a strand of litera-
ture examining shifts in investor attention due to events such as earnings announce-
ments or firm-specific news. Barber and Odean (2006) use different attention proxies,
i.e., stocks appearing in the news, stocks with high abnormal trading volume, and stocks
with extreme one-day returns, and show that individual investors are net-buyers of such
attention-grabbing stocks. They also document that institutional investors are less prone
to attention-driven purchases. Frazzini and Lamont (2006) examine scheduled earnings
49 The original finding that price momentum varies with trading volume is due to Lee and Swaminathan (2000).
73
announcement dates and show that, on average, stock prices rise around these dates
concurrent with an increase in volume. They suggest that this apparent ‘earnings an-
nouncement premium’ is driven by small investor buying and the resulting price pres-
sure when the announcement brings these stocks to their attention. Gervais, Kaniel and
Mingelgrin (2001) use a similar argument to explain their finding of a return premium
to stocks with an abnormal increase in trading volume: they propose that shocks to the
trading activity of a stock affect its visibility, and therefore the demand and price for
that stock.
Recent papers on limited attention related to this essay are Menzly and Ozbas
(2006) and Hong, Torous and Valkanov (2007). Menzly and Ozbas (2006) document a
cross-momentum effect among industries which are related to each other along the sup-
ply chain. They show that a profitable industry momentum strategy can be constructed
based on upstream and downstream industry returns.50 Hong, Torous and Valkanov
(2007) show that the returns of industry portfolios are able to predict the movements of
stock markets, for the U.S. and international equity markets. They attribute their finding
of cross-asset return predictability to slow information diffusion across markets and
limited investor attention to information from markets the investors do not specialize in.
Finally, the paper by Cohen and Frazzini (2006) is most closely related to my
work. They take a new approach to investigating limited investor attention and slow
information diffusion by focusing on a more intuitive link between stocks: firm-specific
customer-supplier relationships. They create a long-short customer momentum strategy
which yields monthly alphas of over 1.5%. Therefore, they conclude that stock prices
do not promptly incorporate news about economically related firms, leading to return
predictability across assets. Naturally, the customer-supplier link provides an excellent
framework for testing limited investor attention. Given U.S. disclosure requirements on
major customers, these links are publicly available to all investors, and they do have
real effects on cash flows (Cohen and Frazzini (2006)).
50 They document that their results are different from individual stock momentum (Jegadeesh and Tit-man (1993)) and industry momentum (Moskowitz and Grinblatt (1999)).
74
However, Cohen and Frazzini (2006) focus on return predictability using
monthly data in an asset pricing framework and do not investigate short-run effects on
the daily level. This leaves some important research questions unanswered which I ad-
dress in this essay: do we observe such cross-asset return predictability only at a
monthly horizon taking all customer information into account? Or do we also observe it
on a daily basis in an event-study framework using only selected customer-supplier
events? How long does it take, exactly, for customer stock return information to be in-
corporated into supplier stock prices?51 The event study approach also provides the op-
portunity to study the more extreme cases of return predictability: i.e., really large cus-
tomer price changes in the tail of the return distribution and the ensuing supplier stock
price reaction.
Such a setting provides an ideal ground to test limited investor attention: if we
can observe supplier return predictability after extreme attention-grabbing customer
price changes, then we can reasonably assume a violation of market efficiency due to
investor inattention to relevant, publicly available information. However, if we can find
no or limited evidence of supplier return predictability after such large customer price
changes, we can conclude that investor inattention to customer information is restricted
to the smaller price changes studied by Cohen and Frazzini (2006) at the monthly hori-
zon.52 This seems plausible as the news driving such extreme customer performance
will catch the attention of some investors, while the large price change itself will catch
the attention of others. Even in the absence of any information, extreme price changes
can become news themselves and generate investor attention (Barber and Odean
(2006)). This leaves the question to be answered empirically: will this attention to the
customer result in an efficient market reaction for suppliers, or can we observe supplier
51 Another question Cohen and Frazzini (2006) do not address is long-term performance beyond the monthly horizon. This may be an avenue for further research, however, keeping in mind the accom-panying asset pricing model specification problems.
52 Cohen and Frazzini (2006) study return differences between quintile portfolios. Assuming a normal distribution, the extreme quintile portfolios have a distance of 0.84 standard deviations from the cen-ter of the distribution.
75
return predictability due to particularly obvious limited investor attention? I will inves-
tigate this issue in the following sections.
4.3 Data and Methodology 4.3.1 Customer-Supplier Relationships
I want to investigate return predictability across assets to examine the extent of investor
limited attention. My empirical analysis is based on long-standing important customer-
supplier relationships between firms. The data on such relationships is taken from the
Compustat segment files and covers the period from 1980 to 2004. U.S. regulation re-
quires publicly listed firms to report information on operating segments in their share-
holder reports.53 Most interesting for this study, firms are required to disclose informa-
tion on any major customer accounting for more than 10% of total sales. This allows me
to identify suppliers and their customers in the merged CRSP/Compustat database. For
the analysis I only use common equity (share codes 10 and 11), which excludes all cer-
tificates, ADRs, SBIs, REITs, etc.
However, the customer identification process is rather tedious: customer identi-
ties are listed in the segment files without unique identifiers such as CRSP’s PERMNO
or Compustat’s GVKEY. The database only contains company name abbreviations be-
fore 1998 and full names afterwards. In addition, the full names do not always corre-
spond exactly to the names listed in the CRSP/Compustat files, and both full names and
abbreviations vary across reporting firms and throughout the years. Therefore, I use a
visual inspection procedure to link the customers to the corresponding PERMNO identi-
fier. First, for the pre-1998 data, I visually compare the abbreviations used in the
Compustat segment files to all company names listed on CRSP. In cases in which visual
inspection determines an almost certain, distinct match, I link the abbreviation with the
corresponding PERMNO identifier. Second, for the post-1998 data, I automatically link
53 Since December 15, 1997, FASB regulation SFAS No. 131 governs segment reporting. From 1977 until 1997, regulation SFAS No. 14 required segment disclosure.
76
full names from the Compustat segment files which exactly match company names
listed on CRSP with the corresponding PERMNO identifier. Next, I visually compare
the remaining unmatched full names to all company names listed on CRSP and link the
corresponding entries whenever I find a unique match. Finally, all customers without a
unique match are excluded from the sample. However, some discretion is involved in
matching name abbreviations and full names with firm identities, so I am deliberately
conservative in assigning firm identifiers. I choose to exclude firms which are only
probable matches because I recon that the statistical noise generated by mistakenly iden-
tifying unrelated firms as customers is greater than the noise associated with failing to
identify actual customers.
Finally, following Cohen and Frazzini (2006) I use the standard time lag im-
posed to match accounting variables with subsequent return data and, therefore, impose
a six month lag between fiscal yearend customer-supplier information and matched
stock returns. My final sample has 11,641 unique customer-supplier relationships with
33,711 distinct annual customer-supplier relationships. I draw confidence in the match-
ing precision from the fact that I am reasonably close to the 11,484 unique customer-
supplier relationships reported by Cohen and Frazzini (2006) over the same time period.
4.3.2 Large Abnormal Customer Price Changes
My daily stock price and return data is from CRSP. A daily customer stock return that
represents a large abnormal price change is called an event. A change is deemed to be
‘large’ if the abnormal return after subtracting the value-weighted CRSP index is more
than three standard deviations above or below the mean. Mean and standard deviation
are calculated over rolling windows comprising the preceding 250 trading days. I follow
Pritamani and Singal (2001) in using such a relative standard deviation based approach
as opposed to an absolute percentage cutoff as I believe that this is more appropriate for
my research objective.54 In order to ensure that I do not include multiple events in the
54 Since a constant percentage price change, say 10%, may be important or not depending on the vola-tility of a customer stock, it appears preferable to use a relative standard deviation based cutoff. The distance of three standard deviations follows Pritamani and Singal (2001). Table 4.3, however, also re-
77
same time period for the same firm, I require customer firms not to have a large abnor-
mal price change over the preceding 20 trading days. Events are restricted to lie on or
between January 2, 1981 and December 2, 2004 to reserve the last 20 trading days in
2004 for examining post-event performance.
I match all such large abnormal price change by a customer with the respective
supplier stock return data. As I am interested in examining only genuinely delayed sup-
plier stock price reactions due to limited investor attention, and not the effects of infre-
quent trading or other market microstructure issues, I impose certain requirements on
these suppliers. First, I exclude all events where the supplier had a closing price of less
than $5 on the event date. Low price stocks usually have high transaction costs, making
it more difficult to incorporate all relevant information.55 Second, I require suppliers to
have positive volume on each of the 60 days preceding the event date. This ensures that
infrequently traded stocks are excluded from the sample. Third, I require the supplier to
have positive volume on the event day itself. This last screen ensures that I only include
supplier stocks which actually traded on the day of the large abnormal price change by
the customer, i.e., stocks which could have reacted to this change contemporaneously.56
The final sample consists of 26,021 events.
I divide this sample into positive and negative events depending on the sign of
the event day customer abnormal return. Of the 26,021 events, the return is significantly
positive for 15,369 events and significantly negative for 10,652 events.57 Table 4.1 pro-
vides summary statistics on the sample. On average, there are 640 positive and 444
negative events per year, with as little as 118 and 22 events and as many as 1,323 and
ports the results for more extreme price changes using four standard deviations as distance. Further-more, I can report that my results remain qualitatively similar using an absolute cut-off of similar magnitude such as 8%, 10% or 12%.
55 I chose the threshold of $5 to be consistent with Cohen and Frazzini (2006). However, I can confirm that my results are qualitatively similar using a screen of $10 as in Pritamani and Singal (2001).
56 Of course, we do not observe intraday dynamics in these price changes and the corresponding sup-plier reactions. Hence, there will be some supplier stocks that did not trade anymore after witnessing a large abnormal price change by their customer towards the end of the trading day. To this extent the market reaction may be deemed efficient even if it only occurs on the next trading day.
57 This preponderance of large positive price changes is also reported by Pritamani and Singal (2001) in the same relative proportions of 3 to 2.
78
905 events in a given year, respectively. Interestingly, for both subsamples only roughly
23% of suppliers are in the same industry as their customer based on the 48 industries of
Fama and French (1997). This finding mitigates any concerns that I may only be pick-
ing up intra-industry effects, an issue that will be examined in more detail in Section
4.4.3.
Table 4.1: Summary Statistics This table shows summary statistics for the annually pooled time series of events as well as the com-pletely pooled sample. The sample contains all large customer abnormal return shock events, defined as customer abnormal returns (AR) three standard deviations above (positive events) or below (negative events) the mean, which can be matched with an economically-linked supplier between January 1981 and December 2004. The customer abnormal return is the customer raw return minus the CRSP value-weighted return, its mean and standard deviation are estimated over rolling 250-day windows. Panel A shows statistics for positive events and Panel B for negative events.
Panel A: Positive Events Panel B: Negative Events
Mean Me-dian
Min Max S.D. Mean Me-dian
Min Max S.D.
Number of events per year 640 564 118 1323 348 444 442 22 905 238
Number of suppli-ers per year 328 294 72 652 160 264 247 20 470 125
Number of cus-tomers per year 464 417 92 1090 250 348 339 18 784 200
% of supplier-customer in same industry
23.5 23.7 15.0 28.8 3.7 23.2 22.9 16.8 32.4 4.6
Customer abnor-mal return (%) 8.4 6.5 1.6 216.0 6.5 -8.8 -6.8 -88.7 -2.2 6.3
Events per customer 5.8 3.0 1.0 37.0 6.5 4.6 2.0 1.0 29.0 5.1
Events per supplier 6.2 3.0 1.0 83.0 7.9 4.8 3.0 1.0 45.0 5.5
Number of suppli-ers per customer 4.6 1.0 1.0 193.0 13.8 4.7 1.0 1.0 185.0 13.5
Number of cus-tomers per supplier 1.8 1.0 1.0 17.0 1.3 1.7 1.0 1.0 14.0 1.2
% of sales to customer 19.1 14.0 0.0 600.0 17.6 19.2 14.0 0.0 330.7 17.0
Examining the statistics on all observations pooled, on average, event day customer
abnormal returns are a significantly positive 8.4% for positive events and significantly
negative 8.8% for negative events, with considerable variation resulting in a standard
deviation of 6.5% and 6.3%, respectively. The minimum value for positive events of
79
1.6% and the maximum value of -2.2% for negative events indicate that the standard
deviation based cut-off procedure clearly allows ‘large’ price change events in low-
volatility customer firms to be relatively ‘small’ in absolute terms.
For positive events (negative events), on average, there are 5.75 (4.56) events
per customer and 6.23 (4.76) events per supplier, with as many as 37 (29) events per
customer and 83 (45) per supplier. Turning to supplier-customer relationships, there are,
on average 4.56 (4.74) unique suppliers per customer and 1.79 (1.71) unique customers
per supplier over the whole time span, with as many as 193 unique suppliers per cus-
tomer and 17 unique customers per supplier. This indicates that some very large cus-
tomer firms, such as General Motors, IBM, Wal-Mart, Ford, or AT&T, have a huge
number of different suppliers, an issue which I take up in the robustness tests in Section
4.4.6. However, the median customer firm is only linked to one supplier and the median
supplier firm is only linked to one customer. On average, 19% of supplier sales go to
each customer, with as little as zero (missing values) and as much as 100%.
4.3.3 Post-Event Abnormal Supplier Returns
To measure abnormal returns, cumulative abnormal returns (CAR) are computed for an
equal-weighted portfolio of supplier firms for up to 20 days following a customer event.
I choose a 20-day period to be able to capture the short-run stock return dynamics fol-
lowing large abnormal customer price changes. I am interested in examining the speed
of supplier price adjustments given limited attention to such customer stock return in-
formation at the monthly horizon as discovered by Cohen and Frazzini (2006). There-
fore, my choice of a 20-day period covers one trading month. To be able to examine the
stock return dynamics more precisely, I report results for various periods up to 20 days,
but focus my discussion on the 20-day period.
80
Daily abnormal returns itAR are calculated as the daily supplier stock return less
the average return to a size-matched control portfolio.58 The control portfolios are
equal-weighted size decile portfolios. Deciles are determined at the beginning of each
year using NYSE size breakpoints. The control portfolio return and breakpoint data is
from Ken French’s homepage.59 Abnormal returns are then simply calculated as the
difference:
tstiti RRAR ,,, (4.1)
where tiR , is the daily return for each supplier i in the sample and tsR , is the respective
size quintile s return on day t . Post-event cumulative abnormal returns ]20,1[,iCAR for
the 20-day window are then simply calculated as the sum: 20
1,]20,1[,
ttii ARCAR (4.2)
The statistical significance of these CARs is then gauged from the cross-sectional mean
and standard deviation over positive and negative events separately. Alternatively, to
assure the robustness of the results I also calculate buy-and-hold abnormal returns
(BHAR) as follows: 20
1,
20
1,]20,1[, )1()1(
tts
ttii RRBHAR (4.3)
Table 4.2 reports some annual descriptive statistics. For conciseness, the discussion is
limited to the positive events displayed in Panel A; the results for negative events
shown in Panel B are largely similar. The table shows that there is considerable varia-
tion in the number of events per year, with as little as 118 events in a low-volatility pe-
riod such as 1981 and a many as 1,323 events in a high volatility period such as 2000.
Furthermore, event-day average customer abnormal returns also show substantial varia-
tion ranging from as low as 5.08% in 1989 to as much as 17.64% in 2001. Interestingly,
58 Opposed to customer abnormal returns, the precise model for measuring supplier abnormal returns does matter regarding my research objective. I follow Pritamani and Singal (2001) in choosing a size-matched control portfolio approach to capture any size effect in returns.
59 I thank Ken French for supplying this data. My results are not susceptible to alternative control port-folios: I reach qualitatively similar conclusions using size quintiles or calculating control portfolio re-turns myself.
81
contemporaneous average supplier abnormal returns are positive in each single year and
reach their maximum simultaneously with average customer abnormal returns in 2001
at 3.52%.
Furthermore, an important feature of the data is immediately visible from the ta-
ble: by construction (due to the reporting requirements) customer firms are, on average,
substantially larger than supplier firms. This raises the question whether I may be pick-
ing up lead-lag effects from large to small stocks as documented by Lo and MacKinlay
(1990) and recently shown to be a within-industry phenomenon by Hou and Moskowitz
(2005) and Hou (2007). Alternatively, I could be picking up some short-term version of
the industry momentum effect found by Moskowitz and Grinblatt (1999). Therefore,
this issue will be addressed in the analysis of industry effects in Section 4.4.3. However,
it should not play an important role in explaining my results as 77% of the customer-
supplier relationships are across industries as reported in Table 4.1.
Tab
le 4
.2:
Ann
ual D
escr
iptiv
e St
atis
tics
The
sam
ple
cont
ains
all
larg
e cu
stom
er a
bnor
mal
ret
urn
shoc
k ev
ents
, def
ined
as
cust
omer
abn
orm
al r
etur
ns (
AR
) th
ree
stan
dard
dev
iatio
ns a
bove
(po
sitiv
e ev
ents
) or b
elow
(neg
ativ
e ev
ents
) the
mea
n, w
hich
can
be
mat
ched
with
an
econ
omic
ally
-link
ed s
uppl
ier b
etw
een
Janu
ary
1981
and
Dec
embe
r 200
4. T
he c
us-
tom
er a
bnor
mal
retu
rn is
the
cust
omer
raw
retu
rn m
inus
the
CR
SP v
alue
-wei
ghte
d re
turn
, its
mea
n an
d st
anda
rd d
evia
tion
are
estim
ated
ove
r rol
ling
250-
day
win
dow
s. Fo
r eac
h ye
ar, t
he ta
ble
show
s th
e nu
mbe
r of e
vent
s, ev
ent d
ay a
vera
ge a
bnor
mal
retu
rn, s
tock
pric
e, a
nd s
ize
(mar
ketc
apita
lizat
ion)
for c
usto
mer
s an
d as
wel
l as t
heir
resp
ectiv
e su
pplie
rs, f
or b
oth
posi
tive
(Pan
elA
) and
neg
ativ
e (P
anel
B) c
usto
mer
retu
rn sh
ocks
. The
supp
lier a
bnor
mal
retu
rn is
the
supp
lier
raw
retu
rn m
inus
the
resp
ectiv
e eq
ual-w
eigh
ted
size
-dec
ile p
ortfo
lio.
Pa
nel A
: Pos
itive
Eve
nts
Pane
l B: N
egat
ive
Even
ts
C
usto
mer
Ave
rage
Supp
lier A
vera
ge
Cus
tom
er A
vera
ge
Su
pplie
r Ave
rage
Yea
r N
umbe
r of
Eve
nts
AR
(%)
Pric
e ($
)Si
ze
($m
)A
R(%
)Pr
ice
($)
Size
($
m)
Num
ber
of E
vent
sA
R(%
)Pr
ice
($)
Size
($
m)
AR
(%)
Pric
e($
)Si
ze
($m
)
1981
11
8 7.
13
30.0
62,
884
0.21
23.4
138
322
-5.8
642
.38
3,33
90.
1717
.90
223
1982
33
8 6.
26
35.3
29,
292
0.34
20.8
338
714
7-6
.56
31.3
92,
568
0.03
21.0
435
119
83
223
8.09
43
.32
4,22
00.
5421
.73
249
184
-7.0
350
.44
10,2
83-0
.36
19.8
225
919
84
236
7.85
36
.02
2,34
70.
7217
.75
266
301
-6.0
044
.04
10,4
56-0
.50
19.3
736
1 19
85
434
6.31
49
.92
9,20
60.
3621
.32
474
291
-5.6
960
.35
12,0
17-0
.31
20.9
465
1 19
86
758
5.83
56
.20
12,1
090.
2824
.24
766
446
-5.4
356
.62
19,1
06-0
.01
23.2
673
519
87
971
6.07
62
.63
20,1
600.
1022
.59
703
511
-6.6
446
.52
12,4
04-0
.80
18.7
884
119
88
178
8.62
43
.51
9,33
20.
4719
.37
615
161
-7.4
354
.25
16,9
76-0
.63
17.2
345
719
89
643
5.08
52
.79
19,8
640.
1722
.80
1,38
129
2-5
.66
55.6
518
,717
-0.3
620
.55
1,01
119
90
518
6.29
41
.65
15,1
770.
3018
.82
701
486
-6.2
242
.58
18,4
66-0
.08
20.9
91,
664
1991
40
6 8.
67
38.5
410
,340
0.75
22.0
086
230
8-7
.44
48.7
019
,471
-0.4
420
.53
798
1992
58
0 7.
43
42.4
818
,020
0.04
21.9
31,
236
327
-9.4
447
.80
19,8
47-0
.31
19.8
661
3 19
93
837
7.17
45
.57
24,6
320.
4221
.10
1,12
446
9-7
.03
43.9
419
,509
-0.4
920
.98
1,01
9 19
94
823
7.65
43
.10
21,9
480.
2519
.82
911
490
-7.4
941
.06
19,3
43-0
.46
20.9
71,
002
1995
82
7 7.
20
49.2
725
,592
0.75
23.0
71,
118
715
-6.9
750
.76
32,3
34-0
.95
23.1
91,
131
1996
1,
182
8.46
55
.19
25,9
620.
3720
.20
747
739
-9.3
544
.75
30,8
75-0
.66
21.3
01,
357
1997
1,
181
8.21
55
.69
44,3
550.
2822
.84
1,18
782
9-8
.10
56.7
936
,990
-0.3
723
.61
1,28
919
98
1,12
5 8.
79
54.2
642
,963
0.22
21.3
11,
350
780
-8.8
052
.32
43,0
91-0
.51
21.3
41,
775
1999
89
1 10
.31
60.0
984
,226
0.79
25.6
12,
465
567
-10.
7558
.91
66,5
25-0
.53
23.8
03,
347
2000
1,
323
12.1
2 53
.33
102,
380
1.02
28.4
72,
768
905
-15.
1044
.63
88,6
11-1
.29
26.9
94,
790
2001
43
9 17
.64
26.0
244
,129
3.52
17.9
32,
237
380
-14.
4832
.96
44,3
89-2
.44
20.9
42,
089
2002
54
7 11
.70
28.4
060
,289
1.26
18.0
12,
299
689
-10.
1132
.98
66,2
60-0
.57
19.4
41,
829
2003
39
9 10
.00
31.2
723
,071
0.60
20.3
02,
372
437
-10.
1037
.03
69,7
93-0
.17
20.5
63,
437
2004
39
2 7.
42
43.9
484
,996
0.53
21.3
22,
993
176
-9.4
833
.57
33,7
77-0
.80
19.1
51,
333
All
15,3
69
8.44
48
.76
37,1
720.
5522
.15
1,35
510
,652
-8.7
946
.95
37,4
79-0
.61
21.7
01,
694
82
83
4.4 Results4.4.1 Large Abnormal Customer Price Changes
Panel A of Table 4.3 reports post-event supplier abnormal returns for large abnormal
customer price changes using a three standard deviations cut-off definition as discussed
in the prior section. There is a significant contemporaneous supplier reaction: average
abnormal returns are 0.552% for positive events and -0.613% for negative events, both
highly statistically significant. This contemporaneous reaction to the large abnormal
customer price changes is followed by drift in the directions predicted: on average, sup-
pliers experience statistically significant CARs of 0.569% and -0.372%, respectively,
over the 20-day post-event window. However, this drift appears to be rather short-lived,
as among the shorter windows only the [t+1,t+5] window shows CARs statistically sig-
nificant at the one percent level.60
This finding is interesting given the evidence in Cohen and Frazzini (2006) who
document a drift between economically linked firms at the monthly horizon. Using a
different methodology to discover short-run supplier return dynamics, I document
somewhat contrarily that capital markets seem to react rather quickly to the information
contained in large abnormal customer price changes, as they do not show statistically or
economically significant drift for windows more than a week after the event. However,
Cohen and Frazzini (2006) examine differences between quintile portfolios, while I
look farther into the tails of the distribution, namely three standard deviations from the
mean.61 Such extreme customer price changes can be considered as attention-grabbing
events (Barber and Odean (2006)), after which the (limited) investor attention is appar-
ently relatively quickly shifted onto the importance of such customer stock return in-
formation for supplier stock prices. I conjecture that this is the reason why the major
part of the supplier return predictability I uncover occurs within the first five days fol-
lowing the large customer price change.
60 I also checked for any continuing drift beyond the 20-day post-event window, but did not find any significant effects.
61 For a normal distribution of returns, the extreme quintile portfolios have a distance of 0.84 standard deviations from the center of the distribution.
84
Table 4.3: Abnormal Returns to Suppliers after Customer Return Shocks The sample contains all large customer abnormal return shock events, defined as customer abnormal returns (AR) three standard deviations (Panel A) and four standard deviations (Panel B) above (positive events) or below (negative events) the mean, which can be matched with an economically-linked supplier between January 1981 and December 2004. The customer abnormal return is the customer raw return minus the CRSP value-weighted return, its mean and standard deviation are estimated over rolling 250-day windows. The table shows the number of events, average event day customer abnormal return, as well as supplier cumulative abnormal returns (CAR) for different post-event windows. The supplier ab-normal return is calculated as the supplier raw return minus the respective equal-weighted size-decile portfolio. t-statistics are reported in parentheses.
Panel A: 3 S.D. Distance Panel B: 4 S.D. Distance
Events: Positive Negative Positive NegativeSample Size: 15,369 10,652 5,901 4,944Customer AR (%): 8.436 -8.792 12.016 -12.289 (160.16) (-143.27) (101.58) (-108.65)Trading Day Supplier CAR (%)
t+0 0.552 -0.613 0.614 -0.733 (14.61) (-14.28) (9.18) (-10.88)
t+1 0.231 -0.192 0.218 -0.265 (7.02) (-4.59) (3.86) (-3.97)
t[+1,+5] 0.236 -0.432 0.452 -0.505 (3.56) (-4.92) (4.14) (-3.61)
t[+6,+10] 0.065 0.026 0.124 0.005 (0.97) (0.31) (1.07) (0.04)
t[+11,+15] 0.175 0.022 0.084 0.207 (2.63) (0.27) (0.76) (1.68)
t[+16,+20] 0.093 0.012 0.157 -0.048 (1.43) (0.15) (1.49) (-0.37)
t[+1,+20] 0.569 -0.372 0.817 -0.340 (4.49) (-2.40) (3.92) (-1.41)
Panel B of Table 4.3 reports the results for even larger abnormal customer price changes
four standard deviations from the mean. The number of events is reduced substantially
to 5,901 positive and 4,944 negative events. The findings, however, are largely similar:
return predictability seems to be limited to the post-event window [t+1,t+5] with a sig-
nificantly positive drift of 0.452% and a significantly negative drift of 0.505%, respec-
tively. In addition, we can observe for both panels that a large part of the post-event
reaction occurs immediately on the day after the event. This supports my conclusion
that capital markets incorporate large abnormal customer price changes fairly quickly
into supplier stock prices as (limited) investor attention is grabbed by the large customer
85
price changes, and relatively quickly shifted onto the importance of this customer in-
formation for supplier stock prices.
4.4.2 Contemporaneous Supplier Reactions
This first finding, however, warrants some detailed scrutiny. Intuitively, the contempo-
raneous supplier reaction on the event day itself plays a crucial role in determining the
amount of post-event drift to expect from supplier firms and can be regarded as the best
proxy for investor attention in this setting: if there is a large positive (negative) contem-
poraneous supplier reaction to a large positive (negative) customer price change, it is
very likely that markets have already incorporated all information immediately and no
post-event supplier drift is to be expected. However, if there is no reaction or even a
negative (positive) reaction to a positive (negative) large customer price change, I ex-
pect a drift in supplier stock prices in the same direction as the customer event as inves-
tors slowly shift their limited attention to this customer stock return information and
incorporate it into supplier prices.
Table 4.4 shows exactly this picture: Panel A shows CARs after positive events
for supplier tertiles sorted on their event day abnormal return. The supplier tertile with
the highest abnormal return, i.e., a positive 4.727%, shows no significant positive post-
event day drift. On the contrary, the tertile with the lowest abnormal return, i.e., a nega-
tive 3.195%, shows considerable drift in the same direction as the customer price
change: a highly statistically significant 1.144% for the 20-day post-event window. The
middle tertile with almost no reaction on the event day itself also shows positive drift
with a statistically significant positive CAR of 0.409%.
The results in Panel B for negative events confirm the findings from Panel A:
the supplier tertile with the lowest abnormal return, i.e., a negative 4.745%, shows no
significant negative post-event drift, while the tertile with the highest abnormal return,
i.e., a positive 3.369%, shows a statistically significant negative CAR of 0.685% for the
20-day post-event window. Overall, the picture is consistent with the hypothesis stated
above: supplier firms who show the largest (smallest) contemporaneous reaction to the
86
large positive (negative) customer price change subsequently show no discernible return
predictability, and vice versa.
Table 4.4: Abnormal Returns to Suppliers Conditional on Initial ReactionThe sample contains all large customer abnormal return shock events, defined as customer abnormal returns three standard deviations above (positive events) or below (negative events) the mean, which can be matched with an economically-linked supplier between January 1981 and December 2004. The cus-tomer abnormal return is the customer raw return minus the CRSP value-weighted return, its mean and standard deviation are estimated over rolling 250-day windows. The table shows the number of events, average event day customer abnormal return, as well as supplier cumulative abnormal returns (CAR) for different post-event windows and subsamples based on event day supplier abnormal return (AR). The supplier abnormal return is calculated as the supplier raw return minus the respective equal-weighted size-decile portfolio. Panel A shows the results for positive events, and Panel B for negative events. t-statistics are reported in parentheses.
Panel A: Positive Events Panel B: Negative Events
Event Day Supplier AR:
LowestTertile Tertile 2 Highest
TertileLowestTertile Tertile 2 Highest
Tertile
Sample Size: 5,123 5,123 5,123 3,551 3,550 3,551Customer AR (%): 8.425 7.595 9.288 -9.852 -7.864 -8.659 (93.73) (96.58) (90.95) (-84.74) (-83.11) (-83.09)Trading Day Supplier CAR (%)
t+0 -3.195 0.124 4.727 -4.745 -0.462 3.369 (-75.71) (16.09) (67.98) (-75.09) (-46.83) (59.51)
t+1 0.370 0.103 0.219 -0.104 -0.276 -0.197 (5.92) (2.39) (3.48) (-1.21) (-4.89) (-2.72)
t[+1,+5] 0.666 0.117 -0.076 -0.099 -0.439 -0.758 (5.43) (1.25) (-0.61) (-0.58) (-3.55) (-4.82)
t[+6,+10] 0.146 0.028 0.020 0.258 -0.160 -0.020 (1.16) (0.30) (0.16) (1.51) (-1.37) (-0.13)
t[+11,+15] 0.256 0.146 0.123 -0.048 0.064 0.049 (2.07) (1.56) (0.98) (-0.30) (0.58) (0.35)
t[+16,+20] 0.076 0.117 0.086 -0.205 0.199 0.044 (0.63) (1.22) (0.72) (-1.26) (1.68) (0.30)
t[+1,+20] 1.144 0.409 0.154 -0.094 -0.336 -0.685 (4.81) (2.30) (0.65) (-0.31) (-1.50) (-2.51)
4.4.3 Within- and Cross-Industry Effects
An issue already raised in the discussion in Section 4.3.3 is the discrepancy in size: cus-
tomer firms are substantially larger than supplier firms. Lo and MacKinlay (1990) re-
port lead-lag effects from large to small stocks. Recently, Hou (2007) shows that the
lead-lag effect is an intra-industry phenomenon: stock returns on big firms lead returns
87
on small firms within the same industry. He argues that slow diffusion of common in-
formation is a leading cause of this lead-lag effect. Table 4.5 investigates whether my
results are simply picking up such intra-industry lead lag effects or if we can also ob-
serve return predictability for supplier-customer relationships across industries. There-
fore, I split my sample into across industry (customer and supplier are in different in-
dustries) and within industry (customer and supplier are in the same industry) subsam-
ples. Industries are defined using the 48 industries from Fama and French (1997), and
given the evidence in Table 4.1 the distribution of events is as expected: 78% happen
across industries and 22% within industries.
Table 4.5: Abnormal Returns to Suppliers Within and Across Industries The sample contains all large customer abnormal return shock events, defined as abnormal returns (AR) three standard deviations above (positive events) or below (negative events) the mean, which can be matched with an economically-linked supplier between January 1981 and December 2004. The Customer abnormal return is the customer raw return minus the CRSP value-weighted return, its mean and standard deviation are estimated over rolling 250-day windows. The table shows the number of events, average event day customer abnormal returns, as well as supplier cumulative abnormal returns (CAR) for different post-event windows across industries (customer and supplier in different industries) and within industries (customer and supplier in the same industry). The supplier abnormal return is calculated as the supplier raw return minus the respective equal-weighted size-decile portfolio. Panel A shows the results for all events, and Panel B excluding suppliers in the smallest size decile. t-statistics are reported in parentheses.
Panel A: Including Smallest Size Decile Panel B: Excluding Smallest Size Decile
Across Industries Within Industries Across Industries Within Industries
Events: Pos. Neg. Pos. Neg. Pos. Neg. Pos. Neg.Sample Size: 11,676 8,122 3,561 2,441 9,252 6,487 3,049 2,064Customer AR (%): 8.289 -8.694 8.855 -9.070 8.349 -8.773 8.716 -9.071 (138.46) (-126.67) (80.53) (-66.09) (120.17) (-112.39) (73.90) (-58.87)Trading Day Supplier CAR (%) 0 0.434 -0.484 0.954 -1.027 0.471 -0.534 0.977 -1.130 (10.21) (-10.01) (11.39) (-10.89) (10.13) (-10.08) (10.96) (-11.15)
+1 0.173 -0.144 0.422 -0.357 0.162 -0.120 0.391 -0.322 (4.60) (-3.01) (6.08) (-3.99) (3.93) (-2.27) (5.35) (-3.38)
[+1,+5] 0.110 -0.464 0.642 -0.342 0.181 -0.434 0.588 -0.264 (1.46) (-4.55) (4.54) (-1.95) (2.20) (-3.90) (3.95) (-1.43)
[+6,+10] 0.070 0.060 0.041 -0.095 0.041 0.037 0.113 -0.010 (0.91) (0.60) (0.29) (-0.55) (0.49) (0.34) (0.73) (-0.06)
[+11,+15] 0.101 0.065 0.442 -0.076 0.066 0.073 0.404 -0.084 (1.32) (0.70) (3.14) (-0.46) (0.79) (0.73) (2.67) (-0.48)
[+16,+20] 0.101 -0.029 0.097 0.142 0.131 -0.018 0.112 0.170 (1.35) (-0.30) (0.71) (0.84) (1.59) (-0.17) (0.76) (0.95)
[+1,+20] 0.381 -0.369 1.223 -0.370 0.419 -0.342 1.217 -0.188 (2.64) (-2.05) (4.55) (-1.18) (2.63) (-1.74) (4.25) (-0.57)
88
Panel A reports the results for all suppliers. For negative events, the 20-day post-event
CAR is comparable for both subsamples: across industries it amounts to a statistically
significant -0.369%, while within industries it is an insignificant -0.370%.62 For positive
events, post-event return predictability is larger within industries: the 20-day CAR is a
highly significant 1.223% within industries. However, across industries it still amounts
to a significant 0.381%. Following intuition, within industries the information link be-
tween customers and suppliers is closer as the contemporaneous reaction is about twice
as large as across industries. This observation may also help to explain the higher drift
after positive events within industries: firstly, it is more likely that I have correctly
matched customers and, therefore, face less statistical noise, and secondly, customer
stock return information may simply be more important within industries than across
industries. Notwithstanding, we can conclude that the supplier return predictability
arises also independent of the within-industry lead-lag effect reported by Hou (2007).
However, Hou (2007) argues that this lead lag effect results from a lead of big
firms over small firms. Therefore, Panel B shows the results excluding the smallest sup-
plier size decile.63 The results do not change the conclusion from Panel A. Return pre-
dictability after positive events is stronger within-industries, but also statistically sig-
nificant across industries. Overall, I conclude from this exercise that the within-industry
lead-lag effect documented by Hou (2007) is not responsible for my findings of short-
run return predictability across economically linked stocks, as we do observe it across
industries and excluding the smallest supplier stocks.
4.4.4 Size Groups
Independent of the aforementioned discussion the influence of supplier size warrants a
more detailed investigation. There is widespread evidence in the asset pricing literature
that the information incorporation process in capital markets depends on firm size. Fur-
62 The difference in significance can be attributed to the substantially higher number of cross-industry events.
63 Obviously, how many small firms to exclude is arbitrary. My results are qualitatively similar if I exclude the two smallest deciles.
89
thermore, it is reasonable to assume that investor inattention may vary with firm size as
larger firms are more likely to be covered by analysts or to be mentioned in the news.
Therefore, Table 4.6 reports post-event CARs for different supplier size tertiles. For
positive events in Panel A, return predictability can only be observed for the first two
tertiles: the smallest and middle tertile show significantly positive 20-day post-event
CARs of 0.446% and 0.945%, while the largest tertile only experiences an insignificant
CAR of 0.337%. The results are similar for negative events in Panel B, only the first
two tertiles experience significantly negative CARs of 0.389% and 0.578%, while the
largest tertile shows no post-event return predictability.
Table 4.6: Abnormal Returns to Suppliers for Different Size Groups The sample contains all large customer abnormal return shock events, defined as customer abnormal returns (AR) three standard deviations above (positive events in Panel A) or below (negative events in Panel B) the mean, which can be matched with an economically-linked supplier between January 1981 and December 2004. The customer abnormal return is the customer raw return minus the CRSP value-weighted return, its mean and standard deviation are estimated over rolling 250-day windows. The table shows the number of events, average event day customer abnormal retrun, as well as supplier cumulative abnormal returns (CAR) for different post-event windows and subsamples based on beginning-of-year supplier market capitalization. The supplier abnormal return is calculated as the supplier raw return minus the respective equal-weighted size-decile portfolio. t-statistics are reported in parentheses.
Panel A: Positive Events Panel B: Negative Events
Supplier Size: Smallest Tertile Tertile 2 Largest
TertileSmallest
Tertile Tertile 2 Largest Tertile
Sample Size: 8,538 4,334 2,497 5,847 3,049 1,756Customer AR (%): 8.385 8.627 8.279 -8.662 -9.157 -8.590 (122.95) (81.33) (63.85) (-112.09) (-71.38) (-57.17)Trading Day Supplier CAR (%)
t+0 0.416 0.689 0.782 -0.432 -0.766 -0.950 (7.97) (9.39) (10.09) (-7.23) (-10.07) (-9.19)
t+1 0.275 0.216 0.104 -0.192 -0.276 -0.047 (5.91) (3.55) (1.56) (-3.31) (-3.57) (-0.49)
t[+1,+5] 0.203 0.286 0.262 -0.538 -0.288 -0.329 (2.17) (2.34) (1.89) (-4.36) (-1.74) (-1.83)
t[+6,+10] 0.075 0.107 -0.041 0.123 0.043 -0.326 (0.80) (0.86) (-0.29) (1.01) (0.28) (-1.73)
t[+11,+15] 0.158 0.318 -0.013 0.012 -0.067 0.206 (1.65) (2.67) (-0.09) (0.11) (-0.47) (1.15)
t[+16,+20] 0.011 0.234 0.130 0.014 -0.266 0.492 (0.11) (1.99) (0.95) (0.12) (-1.75) (2.71)
t[+1,+20] 0.446 0.945 0.337 -0.389 -0.578 0.044 (2.53) (3.87) (1.33) (-1.77) (-2.02) (0.14)
90
Given these findings I conjecture that larger supplier stocks incorporate the information
contained in large abnormal customer price changes immediately into their stock prices
and, therefore, show no discernible post-event drift, due to the higher attention investors
give to large stocks. This conjecture is supported by the observation that the largest size
quintiles also show the largest contemporaneous stock price reaction on the event day
itself. Hence, I conclude that limited attention is more severe for small- and mid-cap
stocks, a finding which is not surprising.
4.4.5 Time-Period Subsamples
Another question frequently asked by studies on return predictability concerns the sta-
bility of the relationship over time. Can we observe increasingly efficient information
processing or does limited attention to customer stock return information persevere until
today? In Table 4.7, I address this issue by dividing the sample into three eight-year
time periods.
It is immediately visible that return predictability after negative events has dis-
appeared over time: the 20-day CAR for negative events is highly significantly negative
for the first period from 1981 to 1988 at a large 1.869%, but declines to values indistin-
guishable from zero for the more recent time periods from 1989 to 1996 and 1997 to
2004. For positive events, no such change in the relationship can be observed: the 20-
day CAR is statistically significant positive for all three periods at 0.510%, 0.675%, and
0.501%, respectively.
In addition, an increase in the contemporaneous reaction to large customer price
changes is visible over time, which is consistent with a more efficient incorporation of
customer return information into supplier stock prices in current capital market condi-
tions. Hence, I conjecture that the short-term return predictability arising from limited
attention to large customer price changes has largely disappeared, at least for negative
events. This may be due to a decrease in the preponderance of investor inattention,
maybe arising from more rapidly shifting investor attention given modern-day commu-
nication systems or from exploitation by arbitrageurs. Apparently, at least large negative
91
customer price changes do not cause return predictability for supplier stock prices any
more.
Table 4.7: Abnormal Returns to Suppliers for Different Sample Periods The sample contains all large customer abnormal return shock events, defined as customer abnormal returns (AR) three standard deviations above (positive events) or below (negative events) the mean, which can be matched with an economically-linked supplier for three eight-year time periods: Jan 1981 to Dec 1988, Jan 1989 to Dec 1996, Jan 1997 to to Dec 2004. The customer abnormal return is the customer raw return minus the CRSP value-weighted return, its mean and standard deviation are estimated over rolling 250-day windows. The table shows the number of events, average event day customer abnormal return, as well as supplier cumulative abnormal returns (CAR) for different post-event windows. The supplier abnormal return is calculated as the supplier raw return minus the respective equal-weighted size-decile portfolio. t-statistics are reported in parentheses.
Time Period: Jan 1981 – Dec 1988 Jan 1989 – Dec 1996 Jan 1997 – Dec 2004
Events: Positive Negative Positive Negative Positive NegativeSample Size: 3,256 2,063 5,816 3,826 6,297 4,763Customer AR (%): 6.510 -6.233 7.325 -7.555 10.458 -10.894 (78.84) (-71.16) (101.57) (-81.75) (109.42) (-108.37)Trading Day Supplier CAR (%) t+0 0.299 -0.392 0.379 -0.525 0.843 -0.779 (5.25) (-4.35) (7.70) (-8.78) (11.32) (-10.63)
t+1 0.215 -0.684 0.225 -0.160 0.244 -0.005 (3.95) (-7.40) (4.99) (-2.77) (3.90) (-0.07)
t[+1,+5] 0.191 -1.580 0.321 -0.253 0.180 -0.079 (1.76) (-8.69) (3.65) (-2.13) (1.41) (-0.52)
t[+6,+10] 0.145 0.141 0.114 0.078 -0.022 -0.065 (1.41) (0.80) (1.24) (0.68) (-0.17) (-0.44)
t[+11,+15] 0.209 -0.340 0.286 0.204 0.056 0.032 (2.04) (-2.34) (3.23) (1.80) (0.43) (0.23)
t[+16,+20] -0.035 -0.091 -0.046 -0.073 0.287 0.126 (-0.35) (-0.62) (-0.50) (-0.61) (2.31) (0.87)
t[+1,+20] 0.510 -1.869 0.675 -0.043 0.501 0.013 (2.71) (-6.93) (3.96) (-0.19) (2.02) (0.05)
4.4.6 Robustness Tests
I conduct further subsample tests to ensure my results are robust. First, I restrict my
sample to those suppliers for which the customer information should matter most: i.e., I
only consider suppliers for whom at least 20% of their sales goes to the respective cus-
tomer. For these, the link to the customer should matter most, implying both a larger
contemporaneous stock price reaction, as well as a larger post-event return predictability
92
given the increased importance of the customer stock return information for the supplier
firm. Panel A of Table 4.8 shows that this is indeed the case. CARs for the 20-day event
window amount to a marginally significant positive 0.513% after positive events64 and a
highly significant negative 0.966% after negative events. So at least for negative events,
supplier return predictability is increased when considering suppliers with the closest
links to the respective customers.
Table 4.8: Robustness Tests The sample contains all large customer abnormal return shock events, defined as customer abnormal returns (AR) three standard deviations above (positive events) or below (negative events) the mean, which can be matched with an economically-linked supplier. The customer abnormal return is the cus-tomer raw return minus the CRSP value-weighted return, its mean and standard deviation are estimated over rolling 250-day windows. The table shows the number of events, average event day customer ab-normal return, as well as supplier cumulative abnormal returns (CAR) for different post-event windows. The supplier abnormal return is calculated as the supplier raw return minus the respective equal-weighted size-decile portfolio. t-statistics are reported in parentheses. Panel A restricts the sample to those suppliers which sell at least 20% to the respective customer. Panel B excludes all events for the five customer firms with the highest number of suppliers. Panel C shows supplier buy-and-hold abnormal returns (BHAR) instead of CAR for the post-event windows.
Panel A: Panel B: Panel C:Sales to Customer>20% Excl. Top 5 Customers Supplier BHAR
Events: Positive Negative Positive Negative Positive NegativeSample Size: 3,236 2,297 11,488 8,246 15,369 10,652Customer AR (%): 8.132 -8.675 9.336 -9.449 8.436 -8.792(t-statistic) (79.87) (-71.41) (138.64) (-125.99) (160.16) (-143.27)Supplier CAR (%) (t-statistic)0 0.790 -0.750 0.688 -0.689 0.552 -0.613
(9.57) (-7.49) (14.85) (-13.93) (14.61) (-14.28)
+1 0.244 -0.430 0.254 -0.254 0.231 -0.192(3.37) (-4.53) (6.35) (-5.40) (7.02) (-4.59)
[+1,+5] 0.188 -0.550 0.224 -0.456 0.199 -0.469(1.24) (-2.82) (2.81) (-4.54) (2.97) (-5.43)
[+6,+10] 0.161 -0.121 -0.007 -0.023 0.050 -0.021(1.00) (-0.63) (-0.09) (-0.24) (0.74) (-0.24)
[+11,+15] 0.243 -0.172 0.140 -0.006 0.152 -0.020(1.66) (-0.96) (1.74) (-0.06) (2.27) (-0.25)
[+16,+20] -0.079 -0.124 0.116 0.014 0.062 -0.001(-0.55) (-0.68) (1.48) (0.14) (0.94) (-0.01)
[+1,+20] 0.513 -0.966 0.472 -0.471 0.392 -0.731(1.74) (-2.87) (3.10) (-2.64) (2.95) (-4.58)
64 The decrease in statistical significance is due to the substantially decreased sample size, which is reduced by 80%.
93
Second, I investigate an issue already raised in Section 4.3.2: the influence of very large
customer firms, such as General Motors, IBM, Wal-Mart, Ford, or AT&T, who have a
huge number of different suppliers. Therefore, I exclude all events for the five customer
firms with the highest number of suppliers as they are responsible for 25% of events, a
clearly influential share for such a small number of customers. However, Panel B of
Table 4.8 offers reassurance that the findings are robust as the results are comparable to
the full sample: after positive events I find a significantly positive CAR of 0.472%, and
after negative events I find a significantly negative CAR of 0.471%.
Third, I take an investors perspective and examine post-event BHARs instead of
CARs. However, Panel C shows that this methodical choice does not affect my findings
much, as there is a significantly positive BHAR of 0.392% after positive events and a
significantly negative BHAR of 0.731% after negative events for the 20-day window.
Hence, given the results of theses robustness tests, I conclude that my findings are quite
reliable.
4.4.7 Cumulative Abnormal Return Regressions
Finally, the question remains how cross-sectional differences in customers and suppliers
interact to determine the supplier post-event CARs. To address this issue Table 4.9 re-
ports event-time regressions of the 20-day supplier CARs on a constant and supplier and
customer characteristics. The base regression equation takes the form:
tititjititjtti SUPPSIZEWITHININDSUPPARCUSTARCAR ,,4,,3,2,10]20,1[, (4.4)
where ]20,1[, ttiCAR is the supplier cumulative abnormal return over the post-event win-
dow [t+1,t+20], tjCUSTAR , is the event-day abnormal return for the respective cus-
tomer, tiSUPPAR , is the event day supplier abnormal return, tiSUPPSIZE , is the natural
logarithm of its market value of equity, and tjiWITHININD ,, a dummy taking the value
of one if supplier and customer are in the same industry and zero otherwise, all for sup-
plier i and its customer j on event day t. All independent variables are winsorized at the
94
1% and 99% percentiles and t-statistics are reported using heteroskedasticity-corrected
standard errors based on White (1980).65
Table 4.9: Cross-Sectional Cumulative Abnormal Return Regressions The sample contains all large customer abnormal return shock events, defined as customer abnormal returns three standard deviations above (positive events) or below (negative events) the mean, which can be matched with an economically-linked supplier between January 1981 and December 2004. The table shows event-time regressions of supplier CAR t[t+1,t+20], the sum of supplier raw returns minus the respec-tive equal-weighted size-quintile portfolio over the 20 post-event day window [t+1,t+20], on a constant and customer and supplier characteristics. CUSTAR is the event day customer raw return minus the CRSP value-weighted return, SUPPAR is the event day supplier raw return minus the respective equal-weighted size-quintile portfolio, SUPPSIZE is the natural logarithm of the event day supplier market capitalization, and WITHININD is a dummy taking the value of one for within-industry customer return shock events (customer and supplier are in the same industry), and 0 for across industry events (customer and supplier are in different industries). Panel A shows the results for positive events and Panel B for negative events. Specification I is the base regression, specification II adds year dummies, specification III adds industry dummies, and specification IV includes both year and industry dummies. Independent variables are winsorized at the 99 and 1% level, heteroskedasticity-corrected White t-statistics are re-ported in parentheses below the coefficient estimates, which are displayed x 100.
Panel A: Positive Events Panel B: Negative Events
Dependent Variable: Supplier CAR[t+1,t+20]
I II III IV I II III IV
CONSTANT 1.63 -0.12 0.45 -1.25 -2.50 -3.05 -4.43 -4.91(1.04) (-0.07) (0.14) (-0.38) (-1.30) (-1.24) (-0.62) (-0.67)
CUSTARt+0 3.94 2.19 2.92 0.88 2.02 5.64 4.33 7.17(1.42) (0.74) (1.03) (0.29) (0.61) (1.57) (1.25) (1.92)
SUPPARt+0 -13.63 -13.81 -14.75 -14.89 -12.27 -13.51 -12.56 -13.83(-3.12) (-3.15) (-3.36) (-3.39) (-2.52) (-2.78) (-2.57) (-2.85)
SUPPSIZE -0.07 -0.04 -0.09 -0.07 0.11 0.02 0.13 0.01(-0.86) (-0.49) (-1.07) (-0.80) (1.19) (0.21) (1.20) (0.06)
WITHININD 0.72 0.92 1.33 1.33 3.05 2.30 4.54 3.88(0.66) (0.82) (0.87) (0.86) (1.55) (1.15) (1.80) (1.52)
Year dummies No Yes No Yes No Yes No Yes
Industry dummies No No Yes Yes No No Yes Yes
Adjusted R² (%) 0.10 0.34 0.16 0.39 0.08 0.92 0.20 1.01
F-statistic 5.06 2.98 1.49 1.81 3.29 4.68 1.43 2.47(p-value) (0.00) (0.00) (0.01) (0.00) (0.01) (0.00) (0.02) (0.00)
Degrees of freedom 15,246 15,223 15,199 15,176 10,573 10,550 10,526 10,503
65 I also experimented with a variable measuring the percentage of supplier sales going to the respective customer (see Table 4.1) to measure the importance of the customer to the supplier as in Panel A of Table 4.8. However, due to missing data for a significant number of suppliers I did not include it in the final regression.
95
Given my prior findings on the changing nature of the return predictability over time
and the importance of industries, in addition to the baseline regression specification I, I
also report a specification II adding year dummies, a specification III adding (supplier)
industry dummies, and a specification IV adding both year and industry dummies. Panel
A reports the regression results for positive events and Panel B for negative events. Ad-
justed R² are relatively low across all specifications as usual in such CAR regressions,
the maximum value reached is only about 1%. All F-statistics, however, indicate overall
significance for the regression models.66
The inclusion of most of the explanatory variables is obvious given my prior
findings. First, for positive events in Panel A, the event-day customer abnormal return
has the expected positive influence on the post-event supplier CAR, albeit without sta-
tistical significance in any of the four regression specifications. Nevertheless, the coef-
ficient is consistently positive, i.e., the higher the large price change the higher the post-
event supplier drift, a result already visible in Table 4.3 comparing Panel A (three stan-
dard deviations distance) and Panel B (four standard deviations distance). A similar
conclusion can be deduced from the consistently positive coefficients in Panel B of
Table 4.9 for negative events, i.e., the more negative the large price change the more
(negative) post-event supplier drift can be observed. However, this finding is only sta-
tistically significant for specification IV.
Second, the coefficient on the event day supplier abnormal return is also consis-
tent with intuition and prior findings: it has a statistically significant negative effect in
all specifications and for both positive and negative events. This is in line with the evi-
dence reported in Table 4.4, i.e., for positive events the larger the contemporaneous
positive supplier reaction the smaller the subsequent positive post-event CAR, and for
negative events the larger the contemporaneous negative supplier reaction the smaller
the subsequent negative post-event CAR.
66 T-statistics are qualitatively unchanged if I use Newey and West (1987) autocorrelation- and het-eroskedasticity-corrected standard errors at various lags instead of the White (1980) correction.
96
Third, I include the within-industry dummy in the regression. Given the evi-
dence in Table 4.5 the results are not surprising: Panel A for positive events shows a
significantly positive coefficient across all specifications while Panel B for negative
events shows no significance at all. Hence, for positive events the post-event supplier
CAR is significantly greater if supplier and customer are in the same industry, a finding
which confirms the subsample results in Table 4.5.
Fourth, I investigate the effect of supplier size. For positive events in Panel A
the coefficients are negative but not statistically significant with t-statistics around 1.
For negative events in Panel B the findings are similar as all coefficients are positive but
do not show statistical significance in any of the four specifications. Hence, according
to the regression results larger firms do show lower return predictability as reported in
Table 4.6, but this finding is not statistically reliable.
Summing up, the CAR regressions demonstrate that my findings are robust to
the consideration of all variables simultaneously and do not change after the addition of
year or industry dummies.
4.5 Conclusion
This essay suggests that the return predictability across assets demonstrated by Cohen
and Frazzini (2006) for the monthly horizon is considerably less pronounced when ex-
amined in an event study framework using daily stock returns and looking at large, at-
tention-grabbing customer price changes. I show that for large positive (negative) cus-
tomer price changes supplier stocks experience significantly positive (negative) CAR
for up to 20 days after the event. However, the major part of these returns arises in the
first five days, indicating that capital markets are relatively efficient in incorporating
extreme customer return shocks into supplier stock prices. Furthermore, I show that no
such return predictability exists for the largest firms and, in the case of negative events,
for the more recent part of the time period investigated. My findings hold both within as
well as across industries, and vary as expected with the initial supplier stock price reac-
tion, the natural proxy for limited investor attention in this setting.
97
My finding of only very short-run supplier return predictability for the tails of
the customer return distribution extends the existing literature on limited investor atten-
tion. We know from the evidence in Menzly and Ozbas (2006), Cohen and Frazzini
(2006), and Hong, Torous and Valkanov (2007), among others, that there exists a wide-
spread return predictability across assets in capital markets, most likely due to limited
investor attention and capacity constraints in information processing. However, investor
attention is only ‘partially limited’ when examining large, attention-grabbing customer
price changes, as the ensuing return predictability is short-lived and confined to the first
week after the event. Hence, the return predictability from customers to suppliers un-
covered by Cohen and Frazzini (2006) does not hold completely for the tails of the cus-
tomer return distribution. This result is reassuring as a longer-lived return predictability
for these extreme events would have indicated a widespread and blatant form of market
inefficiency, given that these customer-supplier links are publicly available to all inves-
tors.
Overall, my evidence refutes only the most obvious form of limited investor at-
tention, i.e., supplier return predictability after large, attention-grabbing customer price
changes. Hence, limited investor attention may still play an important role in under-
standing asset price dynamics for less extreme cases. Future research should examine
the impact of different types of information on investor attention, e.g., by examining
supplier return predictability following large customer price changes classified by news
reports. This may allow us to gain a better understanding of information processing by
investors and the impact of limited attention on asset prices. Furthermore, future re-
search should investigate empirical evidence on limited investor attention carefully and
very closely to determine the precise extent to which any effects of limited investor at-
tention on asset prices can be observed. More detailed scrutiny is certainly warranted.
99
5 Conclusion
Overall, this doctoral thesis follows the research objective to improve our understanding
of the efficiency of capital markets by providing new evidence on the information in-
corporation process at the single-firm, industry, and cross-industry level.
To achieve its research objective, this doctoral thesis comprises three essays that
conduct original empirical research using U.S. capital market data. The first two essays
both investigate the importance of M&A for stock prices. The first essay (Chapter 2)
focuses on the single-firm level and investigates the long-term performance of rival
firms directly affected by M&A transactions. Following a similar intuition, the second
essay (Chapter 3) takes an industry perspective and examines the long-term perform-
ance of industry portfolios depending on the impact of M&A on the firms in the indus-
try. Finally, the third essay (Chapter 4) focuses on the cross-industry level, as it exam-
ines explicit economic links between firms and investigates the information transmis-
sion from customer stock prices to supplier stock prices.
Although different in focus and approach, all three essays contribute to the main
challenge of developing a better understanding of a central asset pricing issue: how effi-
ciently do capital markets incorporate new information into the stock price discovery
process. Are capital markets efficient or can we empirically observe violations of mar-
ket efficiency in the information incorporation process?
In general, I find new evidence supporting the view of many researchers in be-
havioral finance that capital markets are not perfectly efficient. In the three different
essays, I document return predictability at the single-firm, industry, and cross-industry
level, which refutes the notion that the information incorporation process for related
firms is completely efficient.
My findings fit well into the results of recent research on underreaction of capi-
tal markets to new information (Frazzini (2006), Jackson and Johnson (2006), Zhang
(2006)) and the effects of limited investor attention (Hou and Moskowitz (2005), Cohen
100
and Frazzini (2006), Hong, Torous and Valkanov (2007), Hou (2007)). Overall, this
doctoral thesis provides some promising avenues for future research on the information
incorporation process in capital markets.
101
References
Agrawal, Anup, Jeffrey F. Jaffe, and Gershon N. Mandelker, 1992, The Post-Merger Performance of Acquiring Firms: A Re-Examination of an Anomaly, Journal of Finance 47 (4), 1605-1621.
Akhigbe, Aigbe, and Anna D. Martin, 2000, Information-Signaling and Competitive Effects of Foreign Acquisitions in the US, Journal of Banking and Finance 24 (8), 1307-1321.
Ambrose, Brent W., and William L. Megginson, 1992, The Role of Asset Structure, Ownership Structure, and Takeover Defenses in Determining Acquisition Like-lihood, Journal of Financial and Quantitative Analysis 27 (4), 575-589.
Andrade, Gregor, Mark Mitchell, and Erik Stafford, 2001, New Evidence and Perspec-tives on Mergers, Journal of Economic Perspectives 15 (2), 103-120.
Asquith, Paul, 1983, Merger Bids, Uncertainty, and Stockholder Returns, Journal of Financial Economics 11 (1-4), 51-83.
Banz, Rolf W., 1981, The Relationship between Return and Market Value of Common Stocks, Journal of Financial Economics 9 (1), 3-18.
Barber, Brad M., and John D. Lyon, 1997, Detecting Long-Run Abnormal Stock Re-turns: The Empirical Power and Specification of Test Statistics, Journal of Fi-nancial Economics 43 (3), 341-372.
Barber, Brad M., and Terrance Odean, 2006, All That Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional Investors, Re-view of Financial Studies, forthcoming.
Basu, Sanjoy, 1983, The Relationship between Earnings' Yield, Market Value and Re-turn for Nyse Common Stocks: Further Evidence, Journal of Financial Econom-ics 12 (1), 129-156.
Bernard, Victor L., and Jacob K. Thomas, 1989, Post-Earnings-Announcement Drift: Delayed Price Response or Risk Premium?, Journal of Accounting Research 27, 1-36.
Bremer, Marc, and Richard J. Sweeney, 1991, The Reversal of Large Stock-Price De-creases, Journal of Finance 46 (2), 747-754.
102
Brown, Keith C., W. V. Harlow, and Seha M. Tinic, 1988, Risk Aversion, Uncertain Information, and Market Efficiency, Journal of Financial Economics 22 (2), 355-385.
Brown, Stephen J., and Jerold B. Warner, 1980, Measuring Security Price Performance, Journal of Financial Economics 8 (3), 205-258.
Brown, Stephen J., and Jerold B. Warner, 1985, Using Daily Stock Returns: The Case of Event Studies, Journal of Financial Economics 14 (1), 3-31.
Campbell, John Y., 2000, Asset Pricing at the Millennium, Journal of Finance 55 (4), 1515-67.
Chan, Wesley S., 2003, Stock Price Reaction to News and No-News: Drift and Reversal after Headlines, Journal of Financial Economics 70 (2), 223-60.
Cohen, Laura, and Andrea Frazzini, 2006, Economic Links and Predictable Returns, Journal of Finance, forthcoming.
Cox, Don R., and David R. Peterson, 1994, Stock Returns Following Large One-Day Declines: Evidence on Short-Term Reversals and Longer-Term Performance, Journal of Finance 49 (1), 255-267.
Cremers, K. J. Martin, Vinay B. Nair, and Kose John, 2007, Takeovers and the Cross-Section of Stock Returns, Working Paper, Yale, Wharton, and Stern.
Daniel, Kent, and Sheridan Titman, 1997, Evidence on the Characteristics of Cross Sec-tional Variation in Stock Returns, Journal of Finance 52 (1), 1-33.
De Bondt, Werner F. M., and Richard Thaler, 1985, Does the Stock Market Overreact?, Journal of Finance 40 (3), 793-805.
Desai, Hemang, and Prem C. Jain, 1999, Firm Performance and Focus: Long-Run Stock Market Performance Following Spinoffs, Journal of Financial Economics 54 (1), 75-101.
Eberhart, Allan C., William F. Maxwell, and Akhtar R. Siddique, 2004, An Examina-tion of Long-Term Abnormal Stock Returns and Operating Performance Follow-ing R&D Increases, Journal of Finance 59 (2), 623-650.
Eckbo, B. Espen, 1983, Horizontal Mergers, Collusion, and Stockholder Wealth, Jour-nal of Financial Economics 11 (1-4), 241-273.
Eckbo, B. Espen, 1985, Mergers and the Market Concentration Doctrine: Evidence from the Capital Market, Journal of Business 58 (3), 325-349.
Fama, Eugene F., 1970, Efficient Capital Markets: A Review of Theory and Empirical Work, Journal of Finance 25 (2), 383-417.
103
Fama, Eugene F., 1991, Efficient Capital Markets: II, Journal of Finance 46 (5), 1575-617.
Fama, Eugene F., 1998, Market Efficiency, Long-Term Returns, and Behavioral Fi-nance, Journal of Financial Economics 49 (3), 283-306.
Fama, Eugene F., and Kenneth R. French, 1992, The Cross-Section of Expected Stock Returns, Journal of Finance 47 (2), 427-465.
Fama, Eugene F., and Kenneth R. French, 1993, Common Risk Factors in the Returns on Stock and Bonds, Journal of Financial Economics 33 (1), 3-56.
Fama, Eugene F., and Kenneth R. French, 1996, Multifactor Explanations of Asset Pric-ing Anomalies, Journal of Finance 51 (1), 55-84.
Fama, Eugene F., and Kenneth R. French, 1997, Industry Costs of Equity, Journal of Financial Economics 43 (2), 153-193.
Fama, Eugene F., and Kenneth R. French, 2007, Dissecting Anomalies, Journal of Fi-nance, forthcoming.
Fama, Eugene F., and James D. MacBeth, 1973, Risk, Return, and Equilibrium: Empiri-cal Tests, Journal of Political Economy 81 (3), 607-36.
Fee, C. Edward, and Shawn Thomas, 2004, Sources of Gains in Horizontal Mergers: Evidence from Customer, Supplier, and Rival Firms, Journal of Financial Eco-nomics 74 (3), 423-460.
Frazzini, Andrea, 2006, The Disposition Effect and Underreaction to News, Journal of Finance 61 (4), 2017-2046.
Frazzini, Andrea, and Owen A. Lamont, 2006, The Earnings Announcement Premium and Trading Volume, Working Paper, University of Chicago and Yale Univer-sity.
Fuller, Kathleen, Jeffry Netter, and Mike Stegemoller, 2002, What Do Returns to Ac-quiring Firms Tell Us? Evidence from Firms That Make Many Acquisitions, Journal of Finance 57 (4), 1763-1793.
Funke, Christian, Timo Gebken, Lutz Johanning, and Volker Floegel, 2008, The Dy-namics within Merger Waves - Evidence from Industry Merger Waves of the 1990s, Working Paper, European Business School.
Gervais, Simon, Ron Kaniel, and Dan H. Mingelgrin, 2001, The High-Volume Return Premium, Journal of Finance 56 (3), 877-919.
Hasbrouck, Joel, 1985, The Characteristics of Takeover Targets: Q and Other Measures, Journal of Banking and Finance 9 (3), 351-362.
104
Hirshleifer, David, 2001, Investor Psychology and Asset Pricing, Journal of Finance 56 (4), 1533-1597.
Hirshleifer, David, and Siew Hong Teoh, 2003, Limited Attention, Information Disclo-sure, and Financial Reporting, Journal of Accounting and Economics 36 (1-3), 337-386.
Hong, Harrison, Terence Lim, and Jeremy C. Stein, 2000, Bad News Travels Slowly: Size, Analyst Coverage, and the Profitability of Momentum Strategies, Journalof Finance 55 (1), 265-295.
Hong, Harrison, and Jeremy C. Stein, 1999, A Unified Theory of Underreaction, Mo-mentum Trading, and Overreaction in Asset Markets, Journal of Finance 54 (6), 2143-2184.
Hong, Harrison, Walter Torous, and Rossen Valkanov, 2007, Do Industries Lead Stock Markets?, Journal of Financial Economics 83 (2), 367-396.
Hou, Kewei, 2007, Industry Information Diffusion and the Lead-Lag Effect in Stock Returns, Review of Financial Studies 20 (4), 1113-1138.
Hou, Kewei, and Tobias J. Moskowitz, 2005, Market Frictions, Price Delay, and the Cross-Section of Expected Returns, Review of Financial Studies 18 (3), 981-1020.
Hou, Kewei, Lin Peng, and Wei Xiong, 2006, A Tale of Two Anomalies: The Implica-tion of Investor Attention for Price and Earnings Momentum, Working Paper, Ohio State University, Baruch College, and Princeton University.
Huberman, Gur, and Tomer Regev, 2001, Contagious Speculation and a Cure for Can-cer: A Nonevent That Made Stock Prices Soar, Journal of Finance 56 (1), 387-396.
Ikenberry, David L., Graeme Rankine, and Earl K. Stice, 1996, What Do Stock Splits Really Signal?, Journal of Financial and Quantitative Analysis 31 (3), 357-375.
Ikenberry, David, Josef Lakonishok, and Theo Vermaelen, 1995, Market Underreaction to Open Market Share Repurchases, Journal of Financial Economics 39 (2-3), 181-208.
Jackson, Andrew, and Timothy Johnson, 2006, Unifying Underreaction Anomalies, Journal of Business 79 (1), 75-114.
Jaffe, Jeffrey F., 1974, Special Information and Insider Trading, Journal of Business 47 (3), 410-428.
105
Jarrell, Gregg A., James A. Brickley, and Jeffry M. Netter, 1988, The Market for Cor-porate Control: The Empirical Evidence since 1980, Journal of Economic Per-spectives 2 (1), 49-68.
Jegadeesh, Narasimhan, 1990, Evidence of Predictable Behavior of Security Returns, Journal of Finance 45 (3), 881-898.
Jegadeesh, Narasimhan, and Jason Karceski, 2004, Long-Run Performance Evaluation: Correlation and Heteroskedasticity-Consistent Tests, Working Paper, Emory University and University of Florida.
Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency, Journal of Finance 48 (1), 65-91.
Jensen, Michael C., and Richard S. Ruback, 1983, The Market for Corporate Control: The Scientific Evidence, Journal of Financial Economics 11 (1-4), 5-50.
Kahneman, Daniel, 1973. Attention and Effort (Experimental Psychology), Prentice Hall, Englewood Cliffs (NJ).
Kothari, S. P., and Jerold B. Warner, 1997, Measuring Long-Horizon Security Price Performance, Journal of Financial Economics 43 (3), 301-339.
Kothari, S. P., and Jerold B. Warner, 2007, Econometrics of Event Studies, in B. Espen Eckbo, ed.: Handbook of Corporate Finance: Empirical Corporate Finance, El-sevier/North Holland, Amsterdam/Oxford.
Lee, Charles M. C., and Bhaskaran Swaminathan, 2000, Price Momentum and Trading Volume, Journal of Finance 55 (5), 2017-69.
Lintner, John, 1965, The Valuation of Risky Assets and the Selection of Risky Invest-ments in Stock Portfolios and Capital Budgets, Review of Economics and Statis-tics 47 (1), 13-37.
Lo, Andrew W., and A. Craig MacKinlay, 1990, When Are Contrarian Profits Due to Stock Market Overreaction?, Review of Financial Studies 3 (2), 175-205.
Loderer, Claudio, and Kenneth Martin, 1992, Postacquisition Performance of Acquiring Firms, Financial Management 21 (3), 69-79.
Loughran, Tim, and Jay R. Ritter, 1995, The New Issues Puzzle, Journal of Finance 50 (1), 23-51.
Loughran, Tim, and Jay R. Ritter, 2000, Uniformly Least Powerful Tests of Market Ef-ficiency, Journal of Financial Economics 55 (3), 361-389.
106
Loughran, Tim, and Anand M. Vijh, 1997, Do Long-Term Shareholders Benefit from Corporate Acquisitions?, Journal of Finance 52 (5), 1765-1790.
Lyon, John D., Brad M. Barber, and Chih-Ling Tsai, 1999, Improved Methods for Tests of Long-Run Abnormal Stock Returns, Journal of Finance 54 (1), 165-201.
Mandelker, Gershon, 1974, Risk and Return: The Case of Merging Firms, Journal of Financial Economics 1 (4), 303-335.
Massa, Massimo, Zahid Rehman, and Theo Vermaelen, 2007, Mimicking Repurchases, Journal of Financial Economics 84 (3), 624-666.
Masulis, Ronald W., Cong Wang, and Fei Xie, 2007, Corporate Governance and Ac-quirer Returns, Journal of Finance 62 (4), 1851-1889.
Menzly, Lior, and Oguzhan Ozbas, 2006, Cross-Industry Momentum, Working Paper, Vega Asset Management and University of Southern California.
Merton, Robert C., 1973, An Intertemporal Capital Asset Pricing Model, Econometrica 41 (5), 867-887.
Merton, Robert C., 1987, A Simple Model of Capital Market Equilibrium with Incom-plete Information, Journal of Finance 42 (3), 483-510.
Michaely, Roni, Richard H. Thaler, and Kent L. Womack, 1995, Price Reactions to Dividend Initiations and Omissions: Overreaction or Drift?, Journal of Finance50 (2), 573-608.
Mitchell, Mark L., and Erik Stafford, 2000, Managerial Decisions and Long-Term Stock Price Performance, Journal of Business 73 (3), 287-329.
Mitchell, Mark, Todd Pulvino, and Erik Stafford, 2004, Price Pressure around Mergers, Journal of Finance 59 (1), 31-63.
Moeller, Sara B., Frederik P. Schlingemann, and Rene M. Stulz, 2004, Firm Size and the Gains from Acquisitions, Journal of Financial Economics 73 (2), 201-228.
Moeller, Sara B., Frederik P. Schlingemann, and Rene M. Stulz, 2007, How Do Diver-sity of Opinion and Information Asymmetry Affect Acquirer Returns?, Review of Financial Studies 20 (5), 2048-2078.
Moskowitz, Tobias J., and Mark Grinblatt, 1999, Do Industries Explain Momentum?, Journal of Finance 54 (4), 1249-1290.
Mossin, Jan, 1966, Equilibrium in a Capital Asset Market, Econometrica 34 (4), 768-783.
107
Newey, Whitney K., and Kenneth D. West, 1987, A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix, Econo-metrica 55 (3), 703-708.
Nisbett, Richard E., and Lee Ross, 1980. Human Inference: Strategies and Shortcom-ings of Social Judgement, Prentice Hall, Englewood Cliffs (NJ).
Park, Jinwoo, 1995, A Market Microstructure Explanation for Predictable Variations in Stock Returns Following Large Price Changes, Journal of Financial and Quan-titative Analysis 30 (2), 241-256.
Pashler, Harold E., 1998. Attention, Psychology Press, Philadelphia (PA).
Peng, Lin, 2005, Learning with Information Capacity Constraints, Journal of Financial and Quantitative Analysis 40 (2), 307-329.
Peng, Lin, and Wei Xiong, 2006, Investor Attention, Overconfidence and Category Learning, Journal of Financial Economics 80 (3), 563-602.
Pritamani, Mahesh, and Vijay Singal, 2001, Return Predictability Following Large Price Changes and Information Releases, Journal of Banking and Finance 25 (4), 631-656.
Purnanandam, Amiyatosh K., and Bhaskaran Swaminathan, 2004, Are Ipos Really Un-derpriced?, Review of Financial Studies 17 (3), 811-848.
Rau, P. Raghavendra, and Theo Vermaelen, 1998, Glamour, Value and the Post-Acquisition Performance of Acquiring Firms, Journal of Financial Economics49 (2), 223-253.
Rosenberg, Barr, Kenneth Reid, and Ronald Lanstein, 1985, Persuasive Evidence of Market Inefficiency, Journal of Portfolio Management 11 (3), 9-16.
Ross, Stephen A., 1976, The Arbitrage Theory of Capital Asset Pricing, Journal of Economic Theory 13 (3), 341-360.
Shahrur, Husayn, 2005, Industry Structure and Horizontal Takeovers: Analysis of Wealth Effects on Rivals, Suppliers, and Corporate Customers, Journal of Fi-nancial Economics 76 (1), 61-98.
Sharpe, William, 1964, Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk, Journal of Finance 19 (3), 425-442.
Shleifer, Andrei, and Robert W. Vishny, 1997, The Limits of Arbitrage, The Journal of Finance 52 (1), 35-55.
108
Song, Moon H., and Ralph A. Walkling, 2000, Abnormal Returns to Rivals of Acquisi-tion Targets: A Test of the 'Acquisition Probability Hypothesis.', Journal of Fi-nancial Economics 55 (2), 143-171.
Spiess, D. Katherine, and John Affleck-Graves, 1995, Underperformance in Long-Run Stock Returns Following Seasoned Equity Offerings, Journal of Financial Eco-nomics 38 (3), 243-267.
Spiess, D. Katherine, and John Affleck-Graves, 1999, The Long-Run Performance of Stock Returns Following Debt Offerings, Journal of Financial Economics 54 (1), 45-73.
Stulz, Rene M., 1990, Managerial Discretion and Optimal Financing Policies, Journal of Financial Economics 26 (1), 3-27.
White, Halbert, 1980, A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity, Econometrica 48 (4), 817-838.
Zhang, X. Frank, 2006, Information Uncertainty and Stock Returns, Journal of Finance61 (1), 105-136.