Selected Essays in Empirical Asset Pricing

Christian Funke Selected Essays in Empirical Asset Pricing

GABLER EDITION WISSENSCHAFT EBS Forschung Schriftenreihe der European Business School (EBS) International University · Schloss Reichartshausen Herausgegeben von Univ.-Prof. Ansgar Richter, PhD

Band 69

Die European Business School (EBS) – gegründet im Jahr 1971 – ist Deutschlands älteste private Wissenschaftliche Hochschule für Betriebswirtschaftslehre im Universitätsrang. Dieser Vorreiterrolle fühlen sich ihre Professoren und Doktoranden in Forschung und Lehre verpflichtet. Mit der Schriftenreihe präsentiert die European Business School (EBS) ausgewählte Ergebnisse ihrer betriebs- und volkswirtschaftlichen Forschung.

Christian Funke

Selected Essays in Empirical Asset Pricing Information Incorporation at 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 1997 1st 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 system or transmitted, mechanical, photocopying or otherwise without prior permission of the copyright holder. Registered and/or industrial names, trade names, trade descriptions etc. cited in this publication are part of the law for trade-mark protection and may not be used free in any form or by any means even if this is not specifically marked. Cover design: Regine Zimmer, Dipl.-Designerin, Frankfurt/Main Printed on acid-free paper Printed 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 research 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 understanding of capital markets, market designs, trading processes, and asset pricing models, a superior handling of large databases, and efficient programming skills. Christian 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 companies 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 average announcement reactions continue to do well in the future, while industries that experience negative average announcement reactions continue to do poorly. The evidence suggests that capital markets underreact to the industry-wide information provided by merger announcements. In the third essay, Christian documents return predictability across stocks, specifically 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 capital 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 Research 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 support provided by the German Academic Exchange Service (DAAD) and the EBS Department 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 appreciation 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 uncomplicated 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 freedom 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 dissertation to her and to my family, especially to my grandmother Gisela and to my grandfather 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

2

Introduction ........................................................................................................... 1 1.1

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

Information Signaling and Competitive Effects of M&A: Long-Term Performance of Rival Companies ........................................................................ 9 2.1

Introduction .................................................................................................. 10

2.2

Related Literature......................................................................................... 13

2.3

Data and Methodology................................................................................. 15

2.4

Results .......................................................................................................... 24

2.5

Conclusion ................................................................................................... 36

Appendix ............................................................................................................... 38 3

4

5

Predictability of Industry Returns After M&A Announcements................... 39 3.1

Introduction .................................................................................................. 40

3.2

Data and Methodology................................................................................. 43

3.3

Results .......................................................................................................... 48

3.4

Conclusion ................................................................................................... 65

Predictability of Supplier Returns After Large Customer Price Changes.... 67 4.1

Introduction .................................................................................................. 68

4.2

Related Literature......................................................................................... 71

4.3

Data and Methodology................................................................................. 75

4.4

Results .......................................................................................................... 83

4.5

Conclusion ................................................................................................... 96

Conclusion............................................................................................................ 99

References ................................................................................................................... 101

XI

Table of Contents

List of Tables and Figures........................................................................................... XIII List of Abbreviations ....................................................................................................XV List of Symbols ..........................................................................................................XVII 1

2

Introduction ........................................................................................................... 1 1.1

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

Information Signaling and Competitive Effects of M&A: Long-Term Performance of Rival Companies ........................................................................ 9 2.1

Introduction .................................................................................................. 10

2.2

Related Literature......................................................................................... 13

2.3

Data and Methodology................................................................................. 15

2.4

2.5

2.3.1

Data .................................................................................................. 15

2.3.2

Event Identification.......................................................................... 16

2.3.3

Long-Term Performance.................................................................. 21

Results .......................................................................................................... 24 2.4.1

Buy-and-Hold Abnormal Returns .................................................... 24

2.4.2

Calendar-Time Portfolios................................................................. 27

2.4.3

Size Groups ...................................................................................... 31

2.4.4

Cross-Sectional Regressions ............................................................ 33

Conclusion ................................................................................................... 36

Appendix ............................................................................................................... 38 3

Predictability of Industry Returns After M&A Announcements................... 39 3.1

Introduction .................................................................................................. 40

3.2

Data and Methodology................................................................................. 43 3.2.1

Daily Announcement Effects ........................................................... 43

3.2.2

Monthly Returns .............................................................................. 46

XII

3.3

Results .......................................................................................................... 48 3.3.1

3.3.2

3.3.3 3.4 4

3.3.1.1

Persistence in Intra-Industry Effects ................................. 48

3.3.1.2

Predictability of Monthly Industry Returns ...................... 50

Industry Portfolio Investment Strategies.......................................... 52 3.3.2.1

Raw and Adjusted Returns................................................ 52

3.3.2.2

Subperiod Analysis ........................................................... 55

3.3.2.3

Size Groups ....................................................................... 56

3.3.2.4

Liquidity Effects ............................................................... 58

3.3.2.5

Strength of Intra-Industry Effects ..................................... 59

3.3.2.6

Three-Factor Risks ............................................................ 61

The Cross-Section of Individual Stock Returns............................... 63

Conclusion ................................................................................................... 65

Predictability of Supplier Returns After Large Customer Price Changes.... 67 4.1

Introduction .................................................................................................. 68

4.2

Related Literature......................................................................................... 71

4.3

Data and Methodology................................................................................. 75

4.4

4.5 5

Industry Announcement Effects....................................................... 48

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

Results .......................................................................................................... 83 4.4.1

Large Abnormal Customer Price Changes..................................... 83

4.4.2

Contemporaneous Supplier Reactions ........................................... 85

4.4.3

Within- and Cross-Industry Effects ............................................... 86

4.4.4

Size Groups .................................................................................... 88

4.4.5

Time-Period Subsamples ............................................................... 90

4.4.6

Robustness Tests ............................................................................ 91

4.4.7

Cumulative Abnormal Return Regressions.................................... 93

Conclusion ................................................................................................... 96

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

ap

Time-series regression intercept for portfolio p

ARi ,t

Abormal return of firm i on day t

bp

Time-series regression factor loading on market factor for portfolio p

BHARi ,[ t ,T ]

Buy and hold abnormal return for firm i from point in time t until T

BHARi ,[ 1, 20 ]

Buy and hold abnormal return for firm i and 20-day post-event window

CARi ,[ 1, 20 ]

Cumulative abnormal return for firm i and 20-day post-event window

CAR[ 2, 2 ]

Cumulative abnormal return over 5-day event window

CARi ,t 1

Average 5-day cumulative abnormal return of equal-weighted industry portfolio i in the prior month t-1

CUSTAR j ,t

Abnormal return of customer j on event day t

H p,t

Time-series regression error-term for portfolio p in month t

H i,t

Cross-sectional regression error-term for event i at point in time t

hp

Time-series regression factor loading on HML for portfolio p

HML

Fama and French (1993) value factor (high minus low)

R f ,t

Risk-free return over time interval t

Ri ,t

Return of firm i over time interval t

R j ,t

Return of control firm j over time interval t

Rm.t

Return of equal-weighted CRSP market index over time interval t

R s ,t

Return of size quintile s over time interval t

R p ,t

Return of event portfolio p at month t after portfolio formation

R2

Regression coefficient of determination (R squared)

sp

Time-series regression factor loading on SMB for portfolio p

SMB

Fama and French (1993) size factor (small minus big)

SUPPARi ,t

Abnormal return of supplier i on event day t

SUPPSIZEi ,t

Natural logarithm of supplier i market value of equity on event day t

WITHININDi , j ,t 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 asset pricing. All three essays conduct original empirical research using U.S. capital market data. The first two essays both investigate the importance of mergers and acquisitions (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 between firms across industries and investigates the information transmission 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 efficiently do capital markets incorporate new information into the stock price discovery process. Are capital markets efficient or can we empirically observe violations of market efficiency in the information incorporation process? This question is a central issue in the asset pricing literature. In general, according 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 anomalies 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 anomalies 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 remained 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 represented 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 psychological reasons for the mispricing of assets. Therefore, the argument goes, equilibrium prices reflect the weighted average of rational and irrational traders, given the fact that limits of arbitrage exist which prevent rational arbitrageurs from exploiting mispricing 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, recently cross-industry links have generated increasing attention, as they appear to represent 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 understanding of the efficiency of capital markets by providing new evidence on the information 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: LongTerm 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 information signaling effects of the deal outweigh the negative competitive effects. However, 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 information 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 empirical evidence on the long-term wealth effects of corporate acquisitions on directly affected rival firms. This question is important to understand how capital markets incorporate 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 rival firms? Are the positive information-signaling effects and negative competitive effects of M&A transactions efficiently incorporated into rival stock prices? I examine this question using a sample of 2,511 deals announced between January 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 longterm 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 Announcements” 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 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 competitive effects of M&A activity change relatively 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 performance. To the author’s best knowledge, long-term industry returns after merger announcements 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 effects of the deal. Research Question 2:

Are there systematic 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 between 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 announcement of an M&A transaction depends on the average reaction to merger announcements 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 previous 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 bookto-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 cannot explain the returns of the industry portfolio investment strategies. In addition, FamaMcBeth regressions show that the average industry CAR of the previous month is significantly 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 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 longstanding 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. 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 horizon. 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 valuable 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 Frazzini (2006) also on a daily basis in an event study framework using extreme attention-grabbing customer price changes? 3.2: How long does it take, exactly, for customer stock return information to be incorporated 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 abnormal 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 significantly positive (negative) CARs for up to 20 days after the event. However, the major 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 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 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 acquisition targets. 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 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 capital markets do not immediately incorporate the effects of M&A announcements into rival 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 effects of M&A, evidence based on short-term event studies shows that target shareholders 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 returns 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 examine 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 performance of related rivals. Given the findings on significant abnormal long-term performance 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 corporate acquisitions on directly affected rival firms. This question is important to understand how capital markets incorporate the information contained in M&A announcements 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 December 31, 2005. Based on return correlations measured over a year prior to the announced 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 reactions due to the positive information-signaling or negative competitive effects, respectively, 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 shortterm 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, experience 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 information-signaling (‘gaining rivals’) experience significantly positive buy-and-holdabnormal 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 difference 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 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)). 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 return 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 significant. Overall, the results indicate that capital markets do not immediately incorporate the positive information signaling and negative competitive effects of M&A announcement 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 percentile 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 conjecture 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 acquirers 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 competition from the newly merged firms. Overall, researchers have found that the positive information signaling effects outweigh the negative competitive effects of M&A transac-

1

2

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). 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 longterm 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 underperformance 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 announcements 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 returns 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 valueincreasing 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 analyze 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 longhorizon 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 December 31, 2005, which have a transaction value of at least $100 million and where the acquirer is seeking to purchase at least 50% of shares. In addition, I require both acquirer and target firm to be publicly listed U.S. companies, excluding foreign acquirers and private firms. This results in an initial sample of 2,836 deals. I assign each transaction to one of 12 industry sectors5 based on the target’s SIC code.6 Secondly, my stock return data are from CRSP. I only use stocks identified as common equity (share code 10 and 11), excluding all certificates, ADRs, SBIs, REITs, etc. In order to minimize the influence of market microstructure effects and to ensure that the results are not driven by the tiniest micro-cap stocks, I exclude all firms which have a closing stock price of less than 5$ and/or which have a market value of equity

4

5

6

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. 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 equipment, telecom, utilities, wholesale and retail, healthcare, finance, and other. 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 tí1 with CRSP returns from July of year t to June of year t+1. Detailed definitions of the variables can be found in the appendix 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 announcement 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 initial analysis, I define the 30 firms as the rivals of the target which have the highest correlation 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

8

9

I thank Kenneth French for providing the NYSE breakpoints on his website. I can report that my findings are not sensitive to these exclusions. 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. In the following analysis I focus on results using the 30 most-correlated rivals to guarantee a sufficient 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 effects or heavy negative competitive effects, I rely on the short-term reactions of the respective 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 percentage 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

11

12

The mean (median) correlation of the 30 rivals is 0.257 (0.222). However, as the minimum correlation is only 0.008, I require a minimum correlation of 0.15. This removes about 13% of potential rivals. This screen ensures that I do not spuriously include lowly correlated rivals in the analysis. 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 transaction (none, one month, two months). 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 announcements, 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 announcement date as:14 2

CAR[ 2, 2 ]

¦ (R

i ,t

 Rm ,t ) ,

(2.1)

t 2

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

14

15

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. 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. 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 announcements, 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 percentage 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 brackets. Year

Number of Deals

As % of All Deals

Average Rival CAR in %

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

93 90 85 74 58 31 41 52 73 106 149 167 259 259 260 199 131 70 101 111 102

3.70 3.58 3.39 2.95 2.31 1.23 1.63 2.07 2.91 4.22 5.93 6.65 10.31 10.31 10.35 7.93 5.22 2.79 4.02 4.42 4.06

0.338 0.304 -0.323 -0.146 0.426 -0.054 -0.059 -0.699 -0.456 0.195 0.086 0.211 0.183 0.178 -0.098 0.231 -0.345 -0.762 0.078 0.412 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 experiencing CARs between -5% and +5%. What is more interesting, 14.7% of rivals experience 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 concentrate 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 correlation 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.

12% 12%

Historical Distribution of Rival Cumulative Abnormal Returns Historical Distribution of Rival Cumulative Abnormal Returns 10,1% 9,7% 10,1% 9,2% 9,7% 9,2% 8,5% 8,5% 7,6% 7,6% 6,9% 6,9%

8% 8%

5,7% 5,7%

6% 6% 4% 4%

5,4% 5,4% 4,1% 4,1% 3,1% 3,1% 2,3% 1,9% 2,3% 1,5% 1,9% 1,1% 0,9% 1,5% 0,7% 1,1% 0,9% 0,7%

4,1% 4,1%

3,1% 2,3% 3,1% 1,8% 2,3% 1,4% 2% 1,8% 1,4% 2% 0,6% 0,9% 1,0% 1,0% 0,6% 0,9% 0% 0%

-1 -1 2% 2% to -1 t -1 -1 1%o -1 1% 1% to1% -1to -10 -1 0%-10 % 0% to% -9to -9 -9 % -t9% % % o -8to - 8% -8 %-t8% % o -7to - 7% -7 %-t7% % o -6to - 6% -6 %-t6% % o -5to - 5% -5 %-t5% % o -4to - 4% -4 %-t4% % o -3to - 3% -3 %-t3% % o -2to - 2% -2 %-t2% % o -1to - 1% -1 %- 1t% % o o 0 0% 0t% 0% to% t1 o 1% % 1% 1t% o o 2% 2t% 2% 2t% o o 3% 3t% 3% 3t% o o 4% 4t% 4% 4t% o o 5% 5t% 5 5% t% o t6 o 6% %6% 6% to o 7% 7t% 7% 7t% o o 8% 8t% 8% 8t% o 9%to 9% 9% t9o% 10to 10 10 % 10 % % t o% 11to 11 11 %11 % % to% to 12 12 % %

Probability Probability

10% 10%

CAR Range CAR Range

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 absolute 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%), 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 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 Very Large Large Very Large Losses Gains Large Losses Gains

Average Rival CAR in % Very Large Large Large Losses Gains Losses

Very Large Gains

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

22 25 65 29 5 21 35 37 35 33 62 89 178 320 283 360 232 100 50 48 50

122 123 226 141 65 81 110 131 116 168 226 244 507 757 739 599 374 205 160 151 163

178 174 228 123 113 83 110 72 95 202 269 326 695 838 695 742 344 197 190 230 234

36 43 52 34 20 27 42 27 37 70 86 144 228 361 345 425 171 71 67 65 50

-13.501 -14.744 -16.110 -13.100 -12.421 -16.223 -13.965 -13.272 -13.375 -14.600 -14.127 -17.080 -16.607 -14.921 -15.796 -18.604 -17.520 -16.810 -13.616 -13.771 -13.406

-8.030 -8.591 -9.694 -8.167 -7.260 -9.282 -9.208 -8.794 -9.022 -8.505 -8.938 -10.784 -10.487 -10.444 -10.556 -14.145 -13.581 -11.976 -9.230 -9.003 -8.877

8.262 8.857 8.674 9.203 8.367 9.396 10.194 10.199 9.934 9.408 9.555 10.661 9.630 10.961 12.165 13.673 12.445 10.123 9.551 8.745 8.416

13.792 14.898 14.116 14.977 14.464 15.125 14.999 15.580 14.372 13.224 15.135 15.027 14.932 15.867 17.197 18.249 17.548 15.140 14.228 13.361 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 difference 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 efficiency and the model of expected returns employed. Bearing this caveat in mind, I carefully proceed in selecting the best available empirical methods for the evaluation of rival 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 methodology has substantially higher statistical power to detect abnormal performance than the calendar-time method.17 The BHAR approach is also known as the characteristicbased 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

BHARi ,[ t ,T ]

– (1  R

i ,t

t 1

16

17

T

)  – (1  R j ,t ) ,

(2.2)

t 1

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). The use of the BHAR methodology is substantiated by the fact that its statistical problems only become 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), returns for the additional matching firms are spliced in from that point in time going forward. 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 neglected 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, crosssectional 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). However, I choose to use a calendar-time portfolio approach instead to evaluate the robustness 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 Stafford (2000). The approach involves calculating calendar-time portfolios for firms experiencing 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 constructed comprising all firms with an event within the previous 12 months. As the number 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 resulting 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 returns 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: R p ,t  R f ,t

a p  b p ( Rm ,t  R f ,t )  s p ( SMBt )  h p ( HMLt )  H p ,t

(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-tomarket (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 account for heteroskedasticity due to the changing number of portfolio firms and possible autocorrelation, I use Newey and West (1987) heteroskedasticity- and autocorrelationcorrected standard errors with three monthly lags to calculate t-statistics.19

2.4

Results

2.4.1 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 gaining rivals and negative long-horizon post-event BHARs for losing rivals. Rivals who

18

19

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. 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 returns 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 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 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 announcement. 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) return 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 fiveday 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. 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 Gaining Rivals (1)

BHAR in % (t-statistic) N

0.383 (1.51) 6,111

0.207 (0.47) 6,048

0.431 (0.65) 5,934

1.101 (1.19) 5,818

2.844 (2.27) 5,708

Losing Rivals (2)

BHAR in % (t-statistic) N

-0.164 (-0.60) 5,386

-0.225 (-0.43) 5,349

-2.494 (-2.96) 5,276

-4.532 (-4.31) 5,193

-8.146 (-6.29) 5,096

Difference (1)-(2)

BHAR in % (t-statistic)

0.547 (1.47)

0.432 (0.64)

2.925 (2.76)

5.633 (4.03)

10.990 (6.09)

Panel B: Very Large Return Events Heavily Gaining Rivals (1)

BHAR in % (t-statistic) N

0.354 (0.71) 2,388

-0.057 (-0.07) 2,353

0.767 (0.59) 2,301

1.259 (0.69) 2,249

4.614 (1.87) 2,198

Heavily Losing Rivals (2)

BHAR in % (t-statistic) N

-0.369 (-0.69) 2,069

-0.512 (-0.48) 2,052

-2.689 (-1.78) 2,024

-5.486 (-2.55) 1,992

-12.895 (-4.80) 1,946

Difference (1)-(2)

BHAR in % (t-statistic)

0.722 (0.99)

0.455 (0.34)

3.455 (1.74)

6.745 (2.41)

17.509 (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 average BHAR of 12.895%. The average BHAR difference of 17.509% is highly statistically 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 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 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 announcement (month 0). 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 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 estimated over the same 250-day window as the correlations. Long-term performance is evaluated using buyand-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 Event-Time Buy-and-Hold Abnormal Rival Returns

BHAR in % BHAR in %

20 20 16 16 12 12 8 8 4 4 0 0 -4 -4 -8 -8 -12 -12 -16 -16 -20 -20

-1 -1

0

0

1

1

2

2

3

3

4

4

5

5

6

6

Event Month Event Month

7

7

8

8

9

9

10 10

11 11

12 12

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 statistically 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 portfolio 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 substantially 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 further 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 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 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 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. Returns are in monthly percent, t-statistics are shown in parentheses. Holding Period

1

3

6

9

12

Gaining Rivals (1)

1.861 (3.61)

1.772 (4.16)

1.522 (3.70)

1.482 (3.65)

1.423 (3.54)

Losing Rivals (2)

0.701 (1.36)

0.810 (1.70)

0.742 (1.57)

0.846 (1.78)

0.794 (1.71)

Long-Short (1)-(2)

1.194 (2.83)

0.963 (3.65)

0.780 (4.04)

0.636 (3.52)

0.629 (4.03)

Panel A: Raw Returns

Panel B: Size-BM Adjusted Returns Gaining Rivals (1)

1.105 (2.62)

0.598 (2.52)

0.414 (2.54)

0.326 (2.32)

0.257 (2.09)

Losing Rivals (2)

-0.448 (-1.27)

-0.103 (-0.47)

-0.336 (-1.86)

-0.261 (-1.52)

-0.381 (-2.55)

Long-Short (1)-(2)

1.584 (2.89)

0.701 (2.37)

0.751 (3.63)

0.587 (3.41)

0.638 (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 horizon: 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 periods the claendar-time adjusted returns of 5.282% and 7.655%, respectively, are comparable 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 announcements for rival firms is robust to the issue of cross-sectional event correlation, as evidenced by the high significance of the long-short portfolio returns at all post-event horizons. In addition, calendar-time portfolios provide for another possibility: the use of an alternative risk-adjustment methodology based on calendar-time multi-factor regressions. 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 significantly 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 exposure 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 horizon, a deduction which I confirmed in unreported subsamples.

30

Table 2.5: Calendar-Time Portfolio Three-Factor Alphas This table shows three-factor alphas, coefficients, t-statistics, and R² values from a monthly time-series regression of R P ,t  R f ,t a P  bP ( Rm ,t  R f ,t )  s P ( SMB t )  h P ( HML t )  H P ,t , 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 window 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 Panel A: Gaining Rivals 1 3 6 9 12 Panel B: Losing Rivals 1 3 6 9 12 Panel C: Long-Short 1 3 6 9 12

aˆ P (%)

bˆP (Rm-Rf)

sˆP (SMB)

hˆP (HML)

R² (%)

0.849 (2.59) 0.619 (3.01) 0.273 (1.79) 0.248 (2.04) 0.173 (1.57)

1.107 (17.25) 1.125 (24.42) 1.213 (36.18) 1.202 (38.91) 1.225 (40.66)

0.628 (6.52) 0.599 (8.03) 0.598 (11.80) 0.585 (11.51) 0.538 (8.42)

-0.220 (-1.75) -0.032 (-0.32) 0.077 (1.15) 0.059 (0.95) 0.073 (1.32)

64.39

-0.456 (-1.53) -0.455 (-1.99) -0.599 (-2.87) -0.482 (-2.10) -0.554 (-2.62)

1.273 (13.63) 1.327 (18.34) 1.397 (19.52) 1.383 (19.24) 1.378 (20.12)

0.382 (2.25) 0.449 (3.41) 0.485 (3.70) 0.474 (3.51) 0.494 (3.73)

-0.159 (-1.07) -0.056 (-0.45) 0.026 (0.21) 0.013 (0.11) 0.075 (0.63)

63.80

1.297 (3.49) 1.073 (3.71) 0.873 (4.03) 0.729 (3.75) 0.726 (4.60)

-0.157 (-1.53) -0.202 (-2.58) -0.183 (-2.97) -0.182 (-3.18) -0.154 (-3.00)

0.239 (1.13) 0.150 (1.18) 0.113 (1.08) 0.111 (1.13) 0.044 (0.57)

-0.075 (-0.41) 0.025 (0.16) 0.051 (0.37) 0.046 (0.40) -0.002 (-0.03)

2.60

78.86 89.48 90.82 92.12

77.59 83.80 81.95 82.82

5.63 8.94 9.84 7.62

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 immediately incorporate the positive information signaling and negative competitive effects 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 direction 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 perspective, in November 2005 (the last pre-event month) the breakpoint between microcap 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

23

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. 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 quantitatively 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 Groups 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 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 capitalization 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. 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. 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

-0.549 (-0.75) 997 0.859 (1.16) 928 -1.408 (-1.35)

-0.818 (-0.61) 987 0.858 (0.57) 914 -1.676 (-0.83)

-3.023 (-1.60) 958 -0.584 (-0.18) 902 -2.438 (-0.66)

2.164 (0.74) 929 -2.565 (-0.75) 885 4.729 (1.06)

8.548 (2.04) 904 -7.136 (-1.46) 861 15.684 (2.45)

0.605 (1.22) 1,875 -0.066 (-0.12) 1,659 0.671 (0.92)

1.052 (1.23) 1,846 0.864 (0.81) 1,646 0.188 (0.14)

2.444 (1.89) 1,813 -3.200 (-2.32) 1,624 5.644 (2.99)

2.272 (1.22) 1,776 -5.861 (-2.83) 1,590 8.133 (2.93)

3.051 (1.23) 1,733 -10.086 (-4.46) 1,554 13.137 (3.87)

0.540 (1.75) 3,239 -0.562 (-1.66) 2,799 1.103 (2.41)

0.036 (0.07) 3,215 -1.223 (-2.05) 2,789 1.259 (1.57)

0.324 (0.40) 3,163 -2.704 (-2.94) 2,750 3.027 (2.47)

0.116 (0.11) 3,113 -4.395 (-3.78) 2,718 4.511 (2.88)

1.049 (0.75) 3,071 -7.345 (-5.35) 2,681 8.394 (4.27)

Panel A: Microcap Rivals Gaining Rivals (1) Losing Rivals (2) Difference (1)-(2)

BHAR in % (t-statistic) N BHAR in % (t-statistic) N BHAR in % (t-statistic)

Panel B: Small Rivals Gaining Rivals (1) Losing Rivals (2) Difference (1)-(2)

BHAR in % (t-statistic) N BHAR in % (t-statistic) N BHAR in % (t-statistic)

Panel C: Big Rivals Gaining Rivals (1) Losing Rivals (2) Difference (1)-(2)

BHAR in % (t-statistic) N BHAR in % (t-statistic) N BHAR in % (t-statistic)

33

A different picture evolves for gaining rivals. For micro-cap rivals in Panel A, the average 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 likelihood 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 postevent 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 informationsignaling and negative competitive effects of M&A announcements. Any variables predicting 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 longterm negative returns to rivals following announcements accompanied by negative competitive 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 according to the Compustat segment files. More focused firms face a higher future takeover 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 segment dummy should not have any significant influence in the BHAR regressions of losing 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 Martin (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 leverage, 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 returns (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 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. 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 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. The segment 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 calculated 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. Method Segment Dummy Ln(Market Cap) Tobin’s Q PPE Cash Leverage Profitability R² (%) Observations

Panel A: Gaining Rivals (I) Pooled (II) FamaMacBeth 0.369 (0.10) -1.118 (-1.16) 1.211 (1.80) 1.589 (0.32) 12.391 (1.24) 0.332 (0.73) 40.847 (2.80) 0.59 4,763

5.768 (1.30) -0.761 (-0.79) 2.106 (2.52) 7.427 (0.99) 7.072 (0.69) 0.423 (0.46) -11.662 (-0.64) 6.99 21

Panel B: Losing Rivals (I) Pooled (II) FamaMacBeth 2.336 (0.81) 0.767 (0.84) -0.533 (-1.73) -4.290 (-0.73) -3.554 (-0.45) 1.103 (2.15) 17.624 (1.51) 0.29 4,335

0.512 (0.17) 1.284 (1.22) -1.158 (-1.39) 1.104 (0.14) -5.830 (-0.53) 0.949 (1.82) -3.452 (-0.38) 7.53 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) regression 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 approach, 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 indicates 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

25

26

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 significance 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. 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. 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 significantly 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 regression 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 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 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 longterm 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 adjustment 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 approach 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 likelihood, resulting in a higher positive return drift after the initial reaction to the M&A announcement. 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 announcements 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 investment 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 income 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 tí1 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 tí1. Tobin’s Q is the ratio of market assets to book assets, where the market value of assets is total assets minus book equity plus market equity. As before, market equity is the CRSP market capitalization at the end of December of year tí1.

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 announcement reactions continue to do well in the future, while industries that experience negative average announcement reactions continue to do poorly. 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. Profitability has strengthened over time and seems to exist also for the largest 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 Locarek-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 workshop 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 significantly at the announcement of a transaction, studies that measure the intra-industry effects 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 consolidation 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 competitive effects of M&A activity change relatively 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 performance. To the author’s best knowledge, long-term industry returns after merger announcements have not yet been

27

28

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. Eckbo (1983) and Eckbo (1985) find positive intra-industry effects for the announcement of horizontal 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 announcement of horizontal deals.

41

studied. Therefore, the objective of this essay is to investigate whether there are systematic 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 between 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 announcement of an M&A transaction depends on the average reaction to merger announcements 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 impact 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 announcements. Paralleling the results for CARs I find that returns depend on the average industry announcement reaction in the previous month. Average monthly size and book-tomarket (B/M) adjusted returns are significantly positive 0.266% after months with average 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 industry-wide information provided by merger announcements does not seem to be incorporated 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 crosssection of industry returns in the current month. Hence, I examine industry portfolio investment strategies in order to investigate the profitability of this cross-sectional pattern 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 statistically 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 indicates 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 relevant for smaller companies, although it remains statistically significant for all sizebased 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. However, 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 hypothesis. 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 average 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 announcements 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 rebalancing 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 Section 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 Database.29 I consider all deals that were announced between January 1, 1985 and December 31, 2002 with an inflation-adjusted transaction value of at least $25 million.30 I assign each transaction to one of 20 industry groups, defined as in Moskowitz and Grinblatt (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 estimate 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

30

31

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. 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. 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 classification.

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

CARi ,[ 2, 2 ]

¦ (R

i ,t

 Rm ,t ) ,

(3.1)

t 2

where Ri,t is industry i’s daily return on date t and Rm,t is the return for the equalweighted 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 reaction 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 million, 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 consistent 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 observed. 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.

Number of Deals

754 970 930 1,197 1,040 657 579 707 922 1,138 1,369 1,704 2,140 2,349 2,011 1,824 1,196 1,096

22,583

Year

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

Total sample

100.00

3.34 4.30 4.12 5.30 4.61 2.91 2.56 3.13 4.08 5.04 6.06 7.55 9.48 10.40 8.90 8.08 5.30 4.85

As % of Total Number

0.048

0.074 0.068 -0.034 0.001 0.008 -0.046 0.019 0.127 0.054 0.121 0.065 0.057 0.167 -0.003 -0.028 -0.008 0.064 0.165

Average Industry CAR (%)

Panel A: Transaction value > $25mn

16,483

600 776 736 947 783 460 404 496 621 787 932 1,175 1,541 1,669 1,500 1,400 868 788

Number of Deals

100.00

3.64 4.71 4.47 5.75 4.75 2.79 2.45 3.01 3.77 4.77 5.65 7.13 9.35 10.13 9.10 8.49 5.27 4.78

As % of Total Number

0.052

0.084 0.072 -0.034 -0.002 -0.013 -0.042 -0.002 0.123 0.044 0.094 0.067 0.095 0.185 -0.006 -0.044 0.032 0.096 0.162

Average Industry CAR (%)

Panel B: Transaction value > $50mn

11,351

461 587 541 692 553 276 258 311 385 512 610 745 1,074 1,123 1,076 1,005 603 539

Number of Deals

100.00

4.06 5.17 4.77 6.10 4.87 2.43 2.27 2.74 3.39 4.51 5.37 6.56 9.46 9.89 9.48 8.85 5.31 4.75

As % of Total Number

0.053

0.095 0.084 -0.071 0.001 -0.023 -0.083 -0.006 0.089 0.050 0.103 0.080 0.127 0.182 -0.010 -0.032 0.024 0.085 0.220

Average Industry CAR (%)

Panel C: Transaction value > $100mn

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). For each year, the table shows the number of deals, the percentage of the total number of deals during that specific year, as well as the average industry reaction to M&A announcements during that year. The 20 Industries are defined as in Moskowitz and Grinblatt (1999). CAR is the cumulative abnormal return to the equal-weighted industry portfolio for a five-day event window [-2,2] around the announcement date. Average Industry CAR denotes the average CAR to the industry portfolio at the announcement of deals that take place within that industry.

Table 3.1: Annual Distribution of M&A Transactions and Average Industry Announcement Effects

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-tomarket 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 independently sorting stocks into size and B/M quintiles using NYSE breakpoints. I then intersect those sorts to create 25 portfolios of similar size and B/M. Stocks within each industry 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 statistics. 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 35

36

37

See, e.g., Banz (1981), Fama and French (1992, 1993, 1996), and Daniel and Titman (1997). To avoid look ahead bias caused by back filled Compustat data, I include firms only once two complete years of history are available. In addition, I exclude firms with negative book value of equity. 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 characteristics better capture cross-sectional variation in mean returns than factor loadings. 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 returns 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 reported 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 average 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.

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

Industry

SIC Codes

Mining Food Apparel Paper Chemical Petroleum Construction Prim. Metals Fab. Metals Machinery Electrical Eq. Transport Eq. Manufacturing Railroads Transport Utilities Dept. Stores Retail

10-14 20 22-23 26 28 29 32 33 34 35 36 37 38-39 40 41-47 49 53 50-52, 54-59 60-69 other

19. Financial 20. Other Average

F-statistic (identical means) (p-value)

Average Number of Stocks (Minimum) 273.5 (149) 109.6 (91) 83.3 (43) 47.2 (33) 320.6 (199) 26.1 (21) 37.9 (23) 73.0 (54) 105.4 (59) 368.5 (261) 449.8 (375) 94.2 (75) 376.1 (307) 12.0 (8) 102.4 (76) 205.0 (132) 39.3 (29) 576.3 (429) 1165.7 (863) 1329.1 (957) 289.8 (209.2)

Raw Returns (%) 0.942 1.091 0.572 1.108 1.467 1.277 1.015 0.766 1.114 1.165 1.386 0.874 1.274 1.226 1.014 1.028 0.715 0.917

Adjusted Industry Returns CAR (%) (%) $25m -0.130 0.152 -0.086 0.031 -0.511 -0.043 -0.109 -0.002 0.387 0.059 -0.030 -0.036 -0.211 -0.050 -0.627 -0.136 -0.185 0.009 -0.072 0.080 0.238 0.040 -0.194 -0.021 0.161 0.181 -0.050 0.043 -0.120 0.052 -0.176 -0.071 -0.168 0.139 -0.163 -0.012

Industry CAR (%) $50m 0.145 0.019 -0.063 -0.045 0.062 -0.021 -0.038 -0.119 0.001 0.066 0.033 -0.004 0.175 0.021 0.046 -0.060 0.154 -0.030

Industry CAR (%) $100m 0.063 0.004 -0.060 -0.008 0.003 -0.001 -0.037 -0.089 0.052 0.037 0.041 -0.007 0.144 0.046 0.024 -0.011 0.179 -0.042

1.318 1.118 1.069

-0.015 0.079 -0.099

0.054 0.053 0.026

0.061 0.061 0.023

0.068 0.064 0.024

0.296 (0.999)

1.124 (0.318)

1.382 (0.124)

1.380 (0.125)

1.064 (0.382)

48

Finally, Table 3.2 reports average industry CARs for the three different M&A transaction samples with minimum transaction values of $25m, $50m, and $100m, respectively. 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 industry 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 returns 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 transactions. 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 continues and become negative at the end. This pattern is due to increasingly negative competitive effects towards the end of a consolidation 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 competitive effects of M&A activity change relatively 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 consolidation processes.

49

I investigate this issue by examining the intra-industry effects of M&A announcements 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 portfolio 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 zvalues 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 Activity Industryi,t-1 (2)

Average Industry CARi,t-1 < 0 (3)

Difference ((1)-(3))

Panel A: Transaction value > $25mn CAR i, t (%)

N

0.048 (5.47)

0.173 (14.79)

0.030 (0.65)

-0.095 (7.08)

0.268 (15.13)

[0.035] (6.99)

[0.140] (18.85)

[-0.0124] (0.34)

[-0.0919] (9.48)

[0.232] (19.69)

22,583

11,541

1,015

10,027

Panel B: Transaction value > $50mn CAR i, t (%)

N

0.052 (5.05)

0.176 (13.34)

0.055 (1.26)

-0.097 (5.90)

0.274 (13.10)

[0.036] (6.35)

[0.137] (15.73)

[-0.0019] (0.65)

[-0.0801] (7.28)

[0.217] (15.88)

16,483

8,351

1,093

7,039

Panel C: Transaction value > $100mn CAR i, t (%)

N

0.053 (4.31)

0.160 (9.61)

0.049 (1.16)

-0.070 (3.52)

0.231 (8.93)

[0.033] (5.25)

[0.127] (11.75)

[-0.0113] (0.82)

[-0.0764] (4.73)

[0.203] (11.42)

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 transaction 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 returns 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) industry portfolios for the period of February, 1985 to December, 2002. Size-B/M Adjusted Industry Return i, t denotes 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 Industry CARi,t-1 > 0 (1)

No M&A Activity Industryi,t-1 (2)

Average Industry CARi,t-1 < 0 (3)

Difference ((1)-(3))

Panel A: Transaction value > $25mn Size-B/M Adjusted Industry Return i, t (%) N

-0.098 (2.07)

0.247 (3.24)

-0.061 (0.61)

-0.477 (6.40)

0.725 (6.79)

[-0.158] (4.33)

[0.082] (1.93)

[-0.199] (2.00)

[-0.424] (7.22)

[0.506] (6.70)

4,300

1,584

1,158

1,558

-0.098 (2.07)

0.266 (3.37)

-0.116 (1.35)

-0.454 (5.62)

[-0.158] (4.33)

[0.101] (2.31)

[-0.218] (3.25)

[-0.382] (6.35)

4,300

1,442

1,451

1,407

Panel B: Transaction value > $50mn Size-B/M Adjusted Industry Return i, t (%) N

0.720 (6.38) [0.483] (6.30)

Panel C: Transaction value > $100mn Size-B/M Adjusted Industry Return i, t (%) N

-0.098 (2.07)

0.252 (3.00)

-0.142 (1.89)

-0.390 (4.43)

0.643 (5.28)

[-0.158] (4.33)

[0.124] (2.45)

[-0.254] (4.06)

[-0.344] (5.27)

[0.467] (5.61)

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 returns from January, 1985 to December, 2002.

52

month’s average industry announcement reaction changes the picture completely. Mirroring Table 3.3, size-B/M adjusted returns are a significantly positive 0.266% following 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 implies an annualized return of 8.64%. So far, my evidence indicates a profitable investment opportunity as the industry-wide information provided by merger announcements does not seem to be incorporated 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 crosssection 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 predictability of industry returns by the average industry announcement effects can be exploited 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 average CAR in industry i in portfolio formation month t (negative reaction, no M&A, positive 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 positive 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 selffinancing 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 tstatistics are given for the self-financing long-short strategy. The results for the onemonth 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 average returns to the long-short strategy suggests that the apparent industry return predictability 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 different 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 calculated 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 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 in parentheses for the self-financing longshort strategy. Months after portfolio formation

Average Industry CARi t > 0 (1)

No M&A Activity Industryi,t (2)

Average Industry 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. Compared 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 returns 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 returns 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 markets 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 January, 1985 to December, 1993. Panel B shows the performance of the portfolios for January, 1994 to December 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 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). 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 longshort strategy. Months after portfolio formation

Average Industry CARi t > 0 (1)

No M&A Activity Industryi,t (2)

Average Industry 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 period I find a statistically significant size-B/M adjusted return of 0.529% (6.35% annualized) 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 quintiles.

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 rollingportfolio method as in as Jegadeesh and Titman (1993) and Fama (1998). Panels A through E show average 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 portfolio formation

Average Industry CARi t > 0 (1)

No M&A Activity Industryi,t (2)

Average Industry 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 portfolio formation

Average Industry CARi t > 0 (1)

No M&A Activity Industryi,t (2)

Average Industry 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, especially 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 investment 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 similar 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 significance, 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 different 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 portfolios, 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 portfolio formation

Average Industry CARi t > 0 (1)

No M&A Activity Industryi,t (2)

Average Industry 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 industries 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 portfolio, 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 industries 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 adjusted returns for these alternative rolling industry investment portfolios.41 Indeed, the profitability of the long-short strategy improves for all holding periods. At the onemonth horizon, returns increase from 0.748% in Table 3.5 to 0.881% (10.57% annualized), and at the 12-month horizon they increase from 0.242% to 0.311% (3.73% annualized). 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 horizons 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

40

41

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 industries are 2.20, 2.37, and 2.11, respectively. 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. 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 important 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 strategy 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: R p ,t  R f ,t

a p  b p ( Rm ,t  R f ,t )  s p ( SMBt )  h p ( HMLt )  H p ,t ,

(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-tomarket portfolio returns from the same calendar months.42 Note that the portfolio returns 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 position in all industries with negative average industry CAR (Panel C). t-statistics are calculated 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 formation. This is due to a significantly positive intercept of 0.537% for the portfolio taking a long position in all industries with positive average industry CAR, and a significantly negative 0.611% for the portfolio taking a short position in all industries with negative average industry CAR in the previous month.

42

43

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. t-statistics are qualitatively unchanged if I use Newey and West (1987) autocorrelation- and heteroskedasticity-corrected standard errors at various lags instead of the White (1980) correction.

62

Table 3.9: Industry Portfolio Three-Factor Alphas This table shows month-by-month alphas, coefficients, t-statistics, and R² values from a monthly timeR p ,t  R f ,t a p  b p ( Rm ,t  R f ,t )  s p ( SMB )  h p ( HML )  H p ,t regression where the left-hand side series 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

aˆ P (%)

bˆP (Rm-Rf)

sˆP (SMB)

hˆP (HML)

R² (%)

N

Panel A: Long-Short Industry Portfolio 1

0.719 (3.06)

-0.108 (-2.07)

0.182 (1.24)

-0.094 (-0.81)

7.24

215

3

0.017 (0.07)

-0.063 (-1.20)

0.328 (2.44)

0.000 (-0.00)

13.86

213

6

-0.074 (-0.39)

-0.087 (-2.02)

0.269 (2.98)

-0.058 (-0.53)

13.66

210

12

-0.001 (-0.01)

0.020 (0.38)

0.117 (1.25)

-0.176 (-1.60)

9.72

204

Panel B: Positive Industry CAR Portfolio 1

0.537 (3.32)

0.927 (23.84)

0.830 (9.57)

0.309 (4.46)

84.29

215

3

0.194 (1.19)

0.964 (23.63)

0.926 (9.98)

0.342 (3.90)

84.58

213

6

0.030 (0.21)

0.976 (22.51)

0.853 (11.39)

0.383 (4.76)

85.99

210

12

0.192 (0.92)

1.008 (20.12)

0.776 (6.96)

0.273 (2.80)

76.70

204

Panel C: Negative Industry CAR Portfolio 1

-0.611 (-3.45)

1.036 (22.24)

0.656 (5.89)

0.411 (4.06)

78.97

215

3

-0.250 (-1.48)

1.028 (24.17)

0.607 (6.09)

0.350 (3.39)

79.95

213

6

-0.320 (-1.74)

1.064 (30.51)

0.593 (6.29)

0.449 (5.10)

79.87

210

12

-0.226 (-1.47)

0.990 (26.95)

0.668 (11.32)

0.457 (6.31)

83.15

204

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 conventional levels of confidence.44 This confirms the earlier conclusion that the industry return 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 announcement. 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 returns? 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 determine the impact of average industry announcement CARs controlling for factors known to predict stock returns. Each month from 1985 to 2002, I regress the crosssection of individual stock returns over one, three, six and 12-month horizons on a constant 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 negative and the six-month intercept is significant at the 10 percent level of confidence.

64

cludes: market ȕ,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)), several 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 ȕ, size and book-to-market as regressors allows to capture their ability, if any, to explain the cross-section of expected stock returns. The one-month past return controls for liquidity and microstructure effects documented 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 predictive ability of the average industry CAR is somewhat less reliable as t-statistics decrease substantially, but the influence remains positive and the coefficients are still significant, 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 evidence on differences in long-term industry returns after observing the initial intraindustry effects and provides strong evidence affirming the robustness of the results.

45

Market ȕs are estimated by regressing the prior 36 months of excess returns for each stock on a constant 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 equalweighted 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: market ȕ (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. HoriIntercept zon

Eˆ

ln(Size)

ln(B/M)

ret-1:-1

ret-12:-2

ret-36:-13

Average Industry CARt-1

Average R² (%)

1

0.0228 (3.05)

0.0008 (0.48)

-0.0009 (1.46)

0.0000 (0.02)

-0.0661 (12.02)

0.0310 (1.25)

-0.1588 (4.75)

0.5036 (7.07)

5.34

3

0.0684 (3.27)

0.0017 (0.39)

-0.0026 (1.64)

-0.0006 (0.19)

-0.0061 (0.88)

0.0485 (0.90)

-0.4816 (4.20)

0.4917 (2.54)

5.56

6

0.1406 (4.06)

0.0041 (0.53)

-0.0054 (2.00)

-0.0010 (0.17)

0.0096 (0.88)

-0.0491 (0.43)

-0.9129 (4.11)

0.7572 (2.15)

5.79

12

0.2154 (4.59)

0.0091 (0.80)

-0.0081 (2.15)

-0.0020 (0.21)

0.0164 (1.16)

-0.2538 (1.54)

-1.3609 (3.98)

0.9855 (2.39)

5.90

3.4 Conclusion Most studies that measure the intra-industry effects of M&A transactions show that rivals experience significantly positive abnormal returns when the industry consolidates. In a recent study, however, 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, they decrease steadily and become negative at the end of such a consolidation process. Hence, time-varying competitive effects of M&A activity, depending on time-varying industry structure, are responsible for a significant difference in intraindustry reactions over time. However, as industry structure and therefore the competitive 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 positive average announcement reactions continue to do well at future merger announcements, while industries that experience negative average announcement reactions continue to do poorly. These differences in short-term announcement reactions directly impact 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 explained 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 returns 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, Gaston 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 longstanding 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 relationships on monthly returns. I want to know more precisely for how long return predictability 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 customer 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 shortlived. My results show that this is the case. I use publicly available customer information 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 investor 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 after 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 extreme return events. I surmise that my different finding arises as capital markets incorporate large abnormal customer price changes more quickly into supplier stock prices. Limited investor attention may be grabbed by such large price changes and shifted relatively 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 investor 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) contemporaneous 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 supplier 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 naming 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 information 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 primarily an industry phenomenon. I show that supplier return predictability also exists across industries and when excluding the smallest supplier decile. When I examine the importance 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 coverage 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 investor 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 capacity constraints in information processing. However, investor attention is only ‘partially limited’ when examining the large, attention-grabbing customer price changes investigated 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 overview 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 methodology. 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 research 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 requirements 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 investors to appropriately share the risks of these stocks (neglected stocks). More recently, Hong and Stein (1999) propose a model with two boundedly rational 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) develop a model investigating investors’ attention allocation, their ensuing learning behavior 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 phenomenon: 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 firmspecific information, especially negative information, diffuses only slowly across investors. 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 literature examining shifts in investor attention due to events such as earnings announcements 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 announcement premium’ is driven by small investor buying and the resulting price pressure 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 supply 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 Titman (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 address 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 incorporated into supplier stock prices?51 The event study approach also provides the opportunity to study the more extreme cases of return predictability: i.e., really large customer 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 horizon.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

52

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 accompanying asset pricing model specification problems. 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 center of the distribution.

75

return predictability due to particularly obvious limited investor attention? I will investigate 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 customersupplier 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 requires publicly listed firms to report information on operating segments in their shareholder reports.53 Most interesting for this study, firms are required to disclose information 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 certificates, ADRs, SBIs, REITs, etc. However, the customer identification process is rather tedious: customer identities 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 before 1998 and full names afterwards. In addition, the full names do not always correspond 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 identifier. 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 identifying 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 imposed 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 matching precision from the fact that I am reasonably close to the 11,484 unique customersupplier 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 volatility 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 abnormal 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 supplier stock price reactions due to limited investor attention, and not the effects of infrequent 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 provides 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

55

56

57

ports the results for more extreme price changes using four standard deviations as distance. Furthermore, I can report that my results remain qualitatively similar using an absolute cut-off of similar magnitude such as 8%, 10% or 12%. 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). Of course, we do not observe intraday dynamics in these price changes and the corresponding supplier 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. 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 picking 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 completely 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 valueweighted 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

Median

Min

Max

S.D.

Mean

Median

Min

Max

S.D.

640

564

118

1323

348

444

442

22

905

238

328

294

72

652

160

264

247

20

470

125

464

417

92

1090

250

348

339

18

784

200

23.5

23.7

15.0

28.8

3.7

23.2

22.9

16.8

32.4

4.6

Customer abnormal return (%) Events per customer

8.4

6.5

1.6

216.0

6.5

-8.8

-6.8

-88.7

-2.2

6.3

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 suppliers per customer Number of customers per supplier % of sales to customer

4.6

1.0

1.0

193.0

13.8

4.7

1.0

1.0 185.0

13.5

1.8

1.0

1.0

17.0

1.3

1.7

1.0

1.0

19.1

14.0

0.0

600.0

17.6

19.2

14.0

Number of events per year Number of suppliers per year Number of customers per year % of suppliercustomer in same industry

14.0

1.2

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 lowvolatility 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 customer and 17 unique customers per supplier. This indicates that some very large customer 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 following large abnormal customer price changes. I am interested in examining the speed of supplier price adjustments given limited attention to such customer stock return information at the monthly horizon as discovered by Cohen and Frazzini (2006). Therefore, 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 ARit 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: ARi ,t

(4.1)

Ri ,t  Rs ,t

where Ri ,t is the daily return for each supplier i in the sample and Rs ,t is the respective size quintile s return on day t . Post-event cumulative abnormal returns CARi ,[ 1, 20 ] for the 20-day window are then simply calculated as the sum: 20

CARi ,[ 1, 20 ]

¦ AR

(4.2)

i ,t

t 1

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

BHARi ,[ 1, 20 ]

– (1  R

i ,t

t 1

20

)  – (1  Rs ,t )

(4.3)

t 1

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 variation in the number of events per year, with as little as 118 events in a low-volatility period 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 variation ranging from as low as 5.08% in 1989 to as much as 17.64% in 2001. Interestingly,

58

59

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 sizematched control portfolio approach to capture any size effect in returns. I thank Ken French for supplying this data. My results are not susceptible to alternative control portfolios: I reach qualitatively similar conclusions using size quintiles or calculating control portfolio returns 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 table: 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 picking 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 customersupplier relationships are across industries as reported in Table 4.1.

Number of Events

118 338 223 236 434 758 971 178 643 518 406 580 837 823 827 1,182 1,181 1,125 891 1,323 439 547 399 392

15,369

Year

1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

All

8.44

7.13 6.26 8.09 7.85 6.31 5.83 6.07 8.62 5.08 6.29 8.67 7.43 7.17 7.65 7.20 8.46 8.21 8.79 10.31 12.12 17.64 11.70 10.00 7.42

AR (%)

48.76

30.06 35.32 43.32 36.02 49.92 56.20 62.63 43.51 52.79 41.65 38.54 42.48 45.57 43.10 49.27 55.19 55.69 54.26 60.09 53.33 26.02 28.40 31.27 43.94

Price ($)

Size ($m)

37,172

2,884 9,292 4,220 2,347 9,206 12,109 20,160 9,332 19,864 15,177 10,340 18,020 24,632 21,948 25,592 25,962 44,355 42,963 84,226 102,380 44,129 60,289 23,071 84,996

Customer Average

Panel A: Positive Events

0.55

0.21 0.34 0.54 0.72 0.36 0.28 0.10 0.47 0.17 0.30 0.75 0.04 0.42 0.25 0.75 0.37 0.28 0.22 0.79 1.02 3.52 1.26 0.60 0.53

AR (%)

22.15

23.41 20.83 21.73 17.75 21.32 24.24 22.59 19.37 22.80 18.82 22.00 21.93 21.10 19.82 23.07 20.20 22.84 21.31 25.61 28.47 17.93 18.01 20.30 21.32

Price ($)

Supplier Average

1,355

383 387 249 266 474 766 703 615 1,381 701 862 1,236 1,124 911 1,118 747 1,187 1,350 2,465 2,768 2,237 2,299 2,372 2,993

Size ($m)

10,652

22 147 184 301 291 446 511 161 292 486 308 327 469 490 715 739 829 780 567 905 380 689 437 176

Number of Events

-8.79

-5.86 -6.56 -7.03 -6.00 -5.69 -5.43 -6.64 -7.43 -5.66 -6.22 -7.44 -9.44 -7.03 -7.49 -6.97 -9.35 -8.10 -8.80 -10.75 -15.10 -14.48 -10.11 -10.10 -9.48

AR (%)

46.95

42.38 31.39 50.44 44.04 60.35 56.62 46.52 54.25 55.65 42.58 48.70 47.80 43.94 41.06 50.76 44.75 56.79 52.32 58.91 44.63 32.96 32.98 37.03 33.57

Price ($)

Size ($m)

37,479

3,339 2,568 10,283 10,456 12,017 19,106 12,404 16,976 18,717 18,466 19,471 19,847 19,509 19,343 32,334 30,875 36,990 43,091 66,525 88,611 44,389 66,260 69,793 33,777

Customer Average

Panel B: Negative Events

-0.61

0.17 0.03 -0.36 -0.50 -0.31 -0.01 -0.80 -0.63 -0.36 -0.08 -0.44 -0.31 -0.49 -0.46 -0.95 -0.66 -0.37 -0.51 -0.53 -1.29 -2.44 -0.57 -0.17 -0.80

AR (%)

21.70

17.90 21.04 19.82 19.37 20.94 23.26 18.78 17.23 20.55 20.99 20.53 19.86 20.98 20.97 23.19 21.30 23.61 21.34 23.80 26.99 20.94 19.44 20.56 19.15

Price ($)

Supplier Average

1,694

223 351 259 361 651 735 841 457 1,011 1,664 798 613 1,019 1,002 1,131 1,357 1,289 1,775 3,347 4,790 2,089 1,829 3,437 1,333

Size ($m)

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. For each year, the table shows the number of events, event day average abnormal return, stock price, and size (market capitalization) for customers and as well as their respective suppliers, for both positive (Panel A) and negative (Panel B) customer return shocks. The supplier abnormal return is the supplier raw return minus the respective equal-weighted size-decile portfolio.

Table 4.2: Annual Descriptive Statistics

82

83

4.4

Results

4.4.1 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, suppliers 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 significant 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 apparently relatively quickly shifted onto the importance of such customer stock return information 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 following the large customer price change.

60

61

I also checked for any continuing drift beyond the 20-day post-event window, but did not find any significant effects. 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 250day 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. Panel A: 3 S.D. Distance Events: Sample Size: Customer AR (%):

Panel B: 4 S.D. Distance

Positive 15,369 8.436 (160.16)

Negative 10,652 -8.792 (-143.27)

Positive 5,901 12.016 (101.58)

Negative 4,944 -12.289 (-108.65)

t+0

0.552 (14.61)

-0.613 (-14.28)

0.614 (9.18)

-0.733 (-10.88)

t+1

0.231 (7.02)

-0.192 (-4.59)

0.218 (3.86)

-0.265 (-3.97)

t[+1,+5]

0.236 (3.56)

-0.432 (-4.92)

0.452 (4.14)

-0.505 (-3.61)

t[+6,+10]

0.065 (0.97)

0.026 (0.31)

0.124 (1.07)

0.005 (0.04)

t[+11,+15]

0.175 (2.63)

0.022 (0.27)

0.084 (0.76)

0.207 (1.68)

t[+16,+20]

0.093 (1.43)

0.012 (0.15)

0.157 (1.49)

-0.048 (-0.37)

t[+1,+20]

0.569 (4.49)

-0.372 (-2.40)

0.817 (3.92)

-0.340 (-1.41)

Trading Day Supplier CAR (%)

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 significantly positive drift of 0.452% and a significantly negative drift of 0.505%, respectively. 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 information for supplier stock prices. 4.4.2 Contemporaneous Supplier Reactions

This first finding, however, warrants some detailed scrutiny. Intuitively, the contemporaneous 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) contemporaneous 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 expect a drift in supplier stock prices in the same direction as the customer event as investors 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 postevent day drift. On the contrary, the tertile with the lowest abnormal return, i.e., a negative 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 Reaction 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 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 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 Event Day Supplier AR:

Lowest Tertile

Sample Size: Customer AR (%):

Panel B: Negative Events

Tertile 2

Highest Tertile

Lowest Tertile

Tertile 2

Highest Tertile

5,123 8.425 (93.73)

5,123 7.595 (96.58)

5,123 9.288 (90.95)

3,551 -9.852 (-84.74)

3,550 -7.864 (-83.11)

3,551 -8.659 (-83.09)

t+0

-3.195 (-75.71)

0.124 (16.09)

4.727 (67.98)

-4.745 (-75.09)

-0.462 (-46.83)

3.369 (59.51)

t+1

0.370 (5.92)

0.103 (2.39)

0.219 (3.48)

-0.104 (-1.21)

-0.276 (-4.89)

-0.197 (-2.72)

t[+1,+5]

0.666 (5.43)

0.117 (1.25)

-0.076 (-0.61)

-0.099 (-0.58)

-0.439 (-3.55)

-0.758 (-4.82)

t[+6,+10]

0.146 (1.16)

0.028 (0.30)

0.020 (0.16)

0.258 (1.51)

-0.160 (-1.37)

-0.020 (-0.13)

t[+11,+15]

0.256 (2.07)

0.146 (1.56)

0.123 (0.98)

-0.048 (-0.30)

0.064 (0.58)

0.049 (0.35)

t[+16,+20]

0.076 (0.63)

0.117 (1.22)

0.086 (0.72)

-0.205 (-1.26)

0.199 (1.68)

0.044 (0.30)

t[+1,+20]

1.144 (4.81)

0.409 (2.30)

0.154 (0.65)

-0.094 (-0.31)

-0.336 (-1.50)

-0.685 (-2.51)

Trading Day Supplier CAR (%)

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: customer firms are substantially larger than supplier firms. Lo and MacKinlay (1990) report 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 information 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 observe return predictability for supplier-customer relationships across industries. Therefore, I split my sample into across industry (customer and supplier are in different industries) and within industry (customer and supplier are in the same industry) subsamples. 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 Across Industries Events: Sample Size: Customer AR (%): Trading Day Supplier CAR (%) 0

Pos. Neg. 11,676 8,122 8.289 -8.694 (138.46) (-126.67)

Within Industries Pos. Neg. 3,561 2,441 8.855 -9.070 (80.53) (-66.09)

0.434 (10.21)

-0.484 (-10.01)

0.954 -1.027 (11.39) (-10.89)

+1

0.173 (4.60)

-0.144 (-3.01)

0.422 (6.08)

[+1,+5]

0.110 (1.46)

-0.464 (-4.55)

[+6,+10]

0.070 (0.91)

[+11,+15]

Panel B: Excluding Smallest Size Decile Across Industries Pos. Neg. 9,252 6,487 8.349 -8.773 (120.17) (-112.39)

Within Industries Pos. 3,049 8.716 (73.90)

Neg. 2,064 -9.071 (-58.87)

0.471 (10.13)

-0.534 (-10.08)

0.977 (10.96)

-1.130 (-11.15)

-0.357 (-3.99)

0.162 (3.93)

-0.120 (-2.27)

0.391 (5.35)

-0.322 (-3.38)

0.642 (4.54)

-0.342 (-1.95)

0.181 (2.20)

-0.434 (-3.90)

0.588 (3.95)

-0.264 (-1.43)

0.060 (0.60)

0.041 (0.29)

-0.095 (-0.55)

0.041 (0.49)

0.037 (0.34)

0.113 (0.73)

-0.010 (-0.06)

0.101 (1.32)

0.065 (0.70)

0.442 (3.14)

-0.076 (-0.46)

0.066 (0.79)

0.073 (0.73)

0.404 (2.67)

-0.084 (-0.48)

[+16,+20]

0.101 (1.35)

-0.029 (-0.30)

0.097 (0.71)

0.142 (0.84)

0.131 (1.59)

-0.018 (-0.17)

0.112 (0.76)

0.170 (0.95)

[+1,+20]

0.381 (2.64)

-0.369 (-2.05)

1.223 (4.55)

-0.370 (-1.18)

0.419 (2.63)

-0.342 (-1.74)

1.217 (4.25)

-0.188 (-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 between 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 supplier size decile.63 The results do not change the conclusion from Panel A. Return predictability after positive events is stronger within-industries, but also statistically significant 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 shortrun 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

63

The difference in significance can be attributed to the substantially higher number of cross-industry events. 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 valueweighted 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 Tertile

Smallest Tertile

Tertile 2

Largest Tertile

Sample Size: Customer AR (%):

8,538 8.385 (122.95)

4,334 8.627 (81.33)

2,497 8.279 (63.85)

5,847 -8.662 (-112.09)

3,049 -9.157 (-71.38)

1,756 -8.590 (-57.17)

t+0

0.416 (7.97)

0.689 (9.39)

0.782 (10.09)

-0.432 (-7.23)

-0.766 (-10.07)

-0.950 (-9.19)

t+1

0.275 (5.91)

0.216 (3.55)

0.104 (1.56)

-0.192 (-3.31)

-0.276 (-3.57)

-0.047 (-0.49)

t[+1,+5]

0.203 (2.17)

0.286 (2.34)

0.262 (1.89)

-0.538 (-4.36)

-0.288 (-1.74)

-0.329 (-1.83)

t[+6,+10]

0.075 (0.80)

0.107 (0.86)

-0.041 (-0.29)

0.123 (1.01)

0.043 (0.28)

-0.326 (-1.73)

t[+11,+15]

0.158 (1.65)

0.318 (2.67)

-0.013 (-0.09)

0.012 (0.11)

-0.067 (-0.47)

0.206 (1.15)

t[+16,+20]

0.011 (0.11)

0.234 (1.99)

0.130 (0.95)

0.014 (0.12)

-0.266 (-1.75)

0.492 (2.71)

t[+1,+20]

0.446 (2.53)

0.945 (3.87)

0.337 (1.33)

-0.389 (-1.77)

-0.578 (-2.02)

0.044 (0.14)

Trading Day Supplier CAR (%)

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 stability 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 disappeared 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 indistinguishable 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 20day 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 conditions. 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 communication 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: Events: Sample Size: Customer AR (%):

Jan 1981 – Dec 1988

Jan 1989 – Dec 1996

Jan 1997 – Dec 2004

Positive 3,256 6.510 (78.84)

Negative 2,063 -6.233 (-71.16)

Positive 5,816 7.325 (101.57)

Negative 3,826 -7.555 (-81.75)

Positive 6,297 10.458 (109.42)

Negative 4,763 -10.894 (-108.37)

0.299 (5.25)

-0.392 (-4.35)

0.379 (7.70)

-0.525 (-8.78)

0.843 (11.32)

-0.779 (-10.63)

t+1

0.215 (3.95)

-0.684 (-7.40)

0.225 (4.99)

-0.160 (-2.77)

0.244 (3.90)

-0.005 (-0.07)

t[+1,+5]

0.191 (1.76)

-1.580 (-8.69)

0.321 (3.65)

-0.253 (-2.13)

0.180 (1.41)

-0.079 (-0.52)

t[+6,+10]

0.145 (1.41)

0.141 (0.80)

0.114 (1.24)

0.078 (0.68)

-0.022 (-0.17)

-0.065 (-0.44)

t[+11,+15]

0.209 (2.04)

-0.340 (-2.34)

0.286 (3.23)

0.204 (1.80)

0.056 (0.43)

0.032 (0.23)

t[+16,+20]

-0.035 (-0.35)

-0.091 (-0.62)

-0.046 (-0.50)

-0.073 (-0.61)

0.287 (2.31)

0.126 (0.87)

t[+1,+20]

0.510 (2.71)

-1.869 (-6.93)

0.675 (3.96)

-0.043 (-0.19)

0.501 (2.02)

0.013 (0.05)

Trading Day Supplier CAR (%) t+0

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 customer. 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 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. 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: Sample Size: Customer AR (%): (t-statistic) Supplier CAR (%) (t-statistic) 0

Positive 3,236 8.132 (79.87)

Negative 2,297 -8.675 (-71.41)

Positive 11,488 9.336 (138.64)

Negative 8,246 -9.449 (-125.99)

Positive 15,369 8.436 (160.16)

Negative 10,652 -8.792 (-143.27)

0.790 (9.57)

-0.750 (-7.49)

0.688 (14.85)

-0.689 (-13.93)

0.552 (14.61)

-0.613 (-14.28)

+1

0.244 (3.37)

-0.430 (-4.53)

0.254 (6.35)

-0.254 (-5.40)

0.231 (7.02)

-0.192 (-4.59)

[+1,+5]

0.188 (1.24)

-0.550 (-2.82)

0.224 (2.81)

-0.456 (-4.54)

0.199 (2.97)

-0.469 (-5.43)

[+6,+10]

0.161 (1.00)

-0.121 (-0.63)

-0.007 (-0.09)

-0.023 (-0.24)

0.050 (0.74)

-0.021 (-0.24)

[+11,+15]

0.243 (1.66)

-0.172 (-0.96)

0.140 (1.74)

-0.006 (-0.06)

0.152 (2.27)

-0.020 (-0.25)

[+16,+20]

-0.079 (-0.55)

-0.124 (-0.68)

0.116 (1.48)

0.014 (0.14)

0.062 (0.94)

-0.001 (-0.01)

[+1,+20]

0.513 (1.74)

-0.966 (-2.87)

0.472 (3.10)

-0.471 (-2.64)

0.392 (2.95)

-0.731 (-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 reports event-time regressions of the 20-day supplier CARs on a constant and supplier and customer characteristics. The base regression equation takes the form: CARi ,[ t 1,t  20 ]

E 0  E1 ˜ CUSTAR j ,t  E 2 ˜ SUPPARi ,t E 3 ˜ WITHININDi , j ,t  E 4 ˜ SUPPSIZE i ,t  H i ,t (4.4)

where CARi ,[ t 1,t 20 ] is the supplier cumulative abnormal return over the post-event window [t+1,t+20], CUSTAR j ,t is the event-day abnormal return for the respective customer, SUPPARi ,t is the event day supplier abnormal return, SUPPSIZE i ,t is the natural logarithm of its market value of equity, and WITHININDi , j ,t a dummy taking the value of one if supplier and customer are in the same industry and zero otherwise, all for supplier 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 respective 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 equalweighted 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 reported in parentheses below the coefficient estimates, which are displayed x 100. Panel A: Positive Events Dependent Variable: Supplier CAR[t+1,t+20]

Panel B: Negative Events

I

II

III

IV

CONSTANT

1.63 (1.04)

-0.12 (-0.07)

0.45 (0.14)

-1.25 (-0.38)

CUSTARt+0

3.94 (1.42)

2.19 (0.74)

2.92 (1.03)

0.88 (0.29)

SUPPARt+0

-13.63 (-3.12)

-13.81 (-3.15)

-14.75 (-3.36)

-14.89 (-3.39)

SUPPSIZE

-0.07 (-0.86)

-0.04 (-0.49)

-0.09 (-1.07)

-0.07 (-0.80)

0.11 (1.19)

0.02 (0.21)

0.13 (1.20)

0.01 (0.06)

0.72 (0.66)

0.92 (0.82)

1.33 (0.87)

1.33 (0.86)

3.05 (1.55)

2.30 (1.15)

4.54 (1.80)

3.88 (1.52)

Year dummies

No

Yes

No

Yes

No

Yes

No

Yes

Industry dummies

No

No

Yes

Yes

No

No

Yes

Yes

0.10

0.34

0.16

0.39

0.08

0.92

0.20

1.01

5.06 (0.00)

2.98 (0.00)

1.49 (0.01)

1.81 (0.00)

3.29 (0.01)

4.68 (0.00)

1.43 (0.02)

2.47 (0.00)

15,246

15,223

15,199

15,176

WITHININD

Adjusted R² (%) F-statistic (p-value) Degrees of freedom

65

I

II

III

IV

-2.50 -3.05 -4.43 -4.91 (-1.30) (-1.24) (-0.62) (-0.67) 2.02 (0.61)

5.64 (1.57)

4.33 (1.25)

7.17 (1.92)

-12.27 -13.51 -12.56 -13.83 (-2.52) (-2.78) (-2.57) (-2.85)

10,573 10,550 10,526 10,503

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. Adjusted 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 statistical significance in any of the four regression specifications. Nevertheless, the coefficient is consistently positive, i.e., the higher the large price change the higher the postevent supplier drift, a result already visible in Table 4.3 comparing Panel A (three standard 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 statistically significant for specification IV. Second, the coefficient on the event day supplier abnormal return is also consistent 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 evidence 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 heteroskedasticity-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 evidence 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 examined in an event study framework using daily stock returns and looking at large, attention-grabbing customer price changes. I show that for large positive (negative) customer 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 reaction, 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 attention. 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 widespread 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 uncovered by Cohen and Frazzini (2006) does not hold completely for the tails of the customer 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 investors. Overall, my evidence refutes only the most obvious form of limited investor attention, i.e., supplier return predictability after large, attention-grabbing customer price changes. Hence, limited investor attention may still play an important role in understanding 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 research should investigate empirical evidence on limited investor attention carefully and very closely to determine the precise extent to which any effects of limited investor attention 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 incorporation 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 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 examines explicit economic links between firms and investigates the information transmission 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 efficiently do capital markets incorporate new information into the stock price discovery process. Are capital markets efficient or can we empirically observe violations of market efficiency in the information incorporation process? In general, I find new evidence supporting the view of many researchers in behavioral 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 capital 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 Likelihood, Journal of Financial and Quantitative Analysis 27 (4), 575-589. Andrade, Gregor, Mark Mitchell, and Erik Stafford, 2001, New Evidence and Perspectives 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 Returns: The Empirical Power and Specification of Test Statistics, Journal of Financial 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, Review of Financial Studies, forthcoming. Basu, Sanjoy, 1983, The Relationship between Earnings' Yield, Market Value and Return for Nyse Common Stocks: Further Evidence, Journal of Financial Economics 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 Decreases, 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 CrossSection of Stock Returns, Working Paper, Yale, Wharton, and Stern. Daniel, Kent, and Sheridan Titman, 1997, Evidence on the Characteristics of Cross Sectional 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 Examination of Long-Term Abnormal Stock Returns and Operating Performance Following R&D Increases, Journal of Finance 59 (2), 623-650. Eckbo, B. Espen, 1983, Horizontal Mergers, Collusion, and Stockholder Wealth, Journal 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), 1575617. Fama, Eugene F., 1998, Market Efficiency, Long-Term Returns, and Behavioral Finance, 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 Pricing 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 Finance, forthcoming. Fama, Eugene F., and James D. MacBeth, 1973, Risk, Return, and Equilibrium: Empirical 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 Economics 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 University. Fuller, Kathleen, Jeffry Netter, and Mike Stegemoller, 2002, What Do Returns to Acquiring 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 Dynamics 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 Disclosure, 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, Journal of Finance 55 (1), 265-295. Hong, Harrison, and Jeremy C. Stein, 1999, A Unified Theory of Underreaction, Momentum 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), 9811020. Hou, Kewei, Lin Peng, and Wei Xiong, 2006, A Tale of Two Anomalies: The Implication 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 Cancer: A Nonevent That Made Stock Prices Soar, Journal of Finance 56 (1), 387396. 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 Corporate Control: The Empirical Evidence since 1980, Journal of Economic Perspectives 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, Elsevier/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 Investments in Stock Portfolios and Capital Budgets, Review of Economics and Statistics 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 Efficiency, 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 Acquirer 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 Incomplete 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 Finance 50 (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 Diversity 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), 768783.

107

Newey, Whitney K., and Kenneth D. West, 1987, A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix, Econometrica 55 (3), 703-708. Nisbett, Richard E., and Lee Ross, 1980. Human Inference: Strategies and Shortcomings 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 Quantitative 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), 631656. Purnanandam, Amiyatosh K., and Bhaskaran Swaminathan, 2004, Are Ipos Really Underpriced?, Review of Financial Studies 17 (3), 811-848. Rau, P. Raghavendra, and Theo Vermaelen, 1998, Glamour, Value and the PostAcquisition Performance of Acquiring Firms, Journal of Financial Economics 49 (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 Financial 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 Acquisition Targets: A Test of the 'Acquisition Probability Hypothesis.', Journal of Financial 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 Economics 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 Finance 61 (1), 105-136.