Value Creation in European Equity Carve-Outs 383500526X, 9783835005266

The total of equity carve-outs (i.e. IPOs of subsidiary firms) in Europe over the last 20 years amounts to approx. 90 bi

467 35 16MB

English Pages 312 [331] Year 2006

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Value Creation in European Equity Carve-Outs
 383500526X, 9783835005266

Citation preview

Nikolas Pojezny

Value Creation in

European Equity Carve-Outs With a foreword by Prof. Ulrich Hommel, Ph.D.

Deutscher Universit~its-Verlag

Bibliografische Information Der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet iJber abrufbar.

Dissertation European Business School Oestrich-Winkel, 2006 D 1540

1. Auflage Oktober 2006 Alle Rechte vorbehalten 9 Deutscher Universit~its-Verlag I GWV Fachverlage GmbH,Wiesbaden 2006 Lektorat: Brigitte Siegel/Britta GShrisch-Radmacher Der Deutsche Universit~its-Verlag ist ein Unternehmen von Springer Science+Business Media. www.duv.de Das Werk einschliel~lich aller seiner Teile ist urheberrechtlich gesch~itzt. Jede Verwertung aul~erhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verla.gs unzul~issig und strafbar. Das gilt insbesondere ~r Vervielffiltigungen, Ubersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Die Wiedergabe von Gebrauchsnamen, Handelsnamen,Warenbezeichnungen usw. in diesem Werk berechtigt auch ohne besondere Kennzeichnung nicht zu der Annahme, dass solche Namen im Sinne der Warenzeichen- und Markenschutz-Gesetzgebung als frei zu betrachten w~iren und daher von jedermann benutzt werden diJrften. Umschlaggestaltung: Regine Zimmer, Dipl.-Designerin, Frankfurt/Main Druck und Buchbinder: Rosch-Buch, Schel~litz Gedruckt auf s~iurefreiem und chlorfrei gebleichtem Papier Printed in Germany ISBN-10 3-8350-0526-X ISBN-13 978-3-8350-0526-6

v

Foreword Over the past two decades, equity carve-outs (ECOs) have become an increasingly popular form of corporate restructuring in Europe. Individual business segments are separated from the parent conglomerate company, and a minority stake is listed on the stock exchange. The parent company thus retains economic control over the subsidiary, while simultaneously creating more transparency for capital markets, restructuring its investment portfolio and creating the option to either reintegrate or completely sell off the subsidiary at a later stage. The attractiveness of ECOs as research objects is largely due to their dualistic nature as both means of parent company financing and corporate restructuring. While comprehensive academic literature on ECOs already exists, studies mainly focus on the US market. The objective of the present study is to conduct corresponding analyses using a European sample, allowing the examination of both research issues on an intra-European level, and of the admissibility of analogies between US and European results. The author's goal is to cover all key aspects of the ECO literature with empirical analyses. This includes both financial (ex ante) and operating (ex post) considerations, as well as the analysis of the short- and long-term performance, and also covers related aspects such as the second event and the efficiency of internal capital markets. The relevant literature is comprehensively reviewed, and the specific research questions are properly derived on its basis. The author makes comprehensive use of existing empirical methodologies, employing alternative testing procedures to increase the robustness of results. While the economic interpretation of the analytical results is quite short in some cases, the author does ground the majority of the results in corporate finance theory. Overall the analyses are conducted very thoroughly and knowledgeably, and thus clearly demonstrate the author's competence in this field. The key results of the thesis are largely in line with intuition, both confirming results from previous studies, as well as adding a series of new insights. Operating and share price performance are found to be influenced by a number of event- and firm-specific characteristics, agreeing with economic common sense. Some of the results are surprising and noteworthy, e.g. an abnormally negative price reaction in the days following the initial ECO announcement (which, as in previous studies, is found to lead to positive announcement period returns). While some of the interpretations in the chapter on internal capital markets seem speculative and are therefore not able to answer the existing uncertainty regarding the efficiency of internal capital markets, the analysis is conducted very thoroughly and in great detail. The results regarding the second event decision complement existing literature by identifying some of the key drivers of the eventual wind-up of the ECO structure. The structure of the thesis is logically consistent and based on the international financial journal standards. The language and diction used complies with the high standards

VI

required in international publishing. The breadth of the analysis' scope by far surpasses that of a standard dissertation, without impairing the quality of the empirical analyses. The geographical focus of the thesis, the one-of-a-kind sample and the level of analytical meticulousness render this study unique. In total, this thesis is a tour de force, representing a milestone in European ECO literature.

Professor Ulrich Hommel, Ph.D. Rudolf-von-Bennigsen-FoerderFoundation Professor of Finance at the EUROPEAN BUSINESS SCHOOL

VII

Preface "Another thesis on equity carve-outs?" This was the tough albeit justified reaction of a former work colleague when I asked him for his opinion about the suitability of this topic as a subject for my doctoral thesis. The process of identifying a fitting subject is daunting: Ideally, there should be some prior research on the topic to prevent having to establish a new research field from scratch, but gaps in understanding should exist; the subject should be specialised enough allowing the researcher to make some tangible contribution to his discipline, without running the risk of producing results only marginally relevant to anyone but himself; 1 and finally, the topic should be of interest outside of academics, e.g., for market practitioners. As for requirement one, a series of papers and Ph.D. theses on equity carve-outs (ECOs) have been published in recent years. 2 Simultaneously, existing literature leaves some crucial questions regarding the potential value creation in ECOs unanswered. 3 As for requirement two, ECOs are a clearly defined, specialised form of corporate restructuring, but at the same time offer fascinating features allowing the formulation of research questions regarding an array of corporate finance concepts. As for requirement three, European firms have carried out at least 178 ECOs with a total volume of approx. C91 billion over the last 20 years. Companies in Europe continue to possess potential ECO candidates, and transaction volume will remain high in the foreseeable future. 4 ECOs thus promise to be an exciting research topic both for this thesis and for future studies. This thesis contributes to a better understanding of ECOs for firms, investors and academics: Firms learn under which conditions markets are likely to react positively to the announcement of an intended ECO, in which specific constellations of circumstances an ECO is likely to create value in the long term, and the potential impact of the ECO design on the firm's future growth, profitability and share price. Investors learn about the probable consequences of investing in firms engaged in an ECO, how to interpret the information provided by the parent firm regarding the ECO, and about profitable trading strategies involving the eventual reacquisition or complete sell-off of the partially floated subsidiary firm. Academics learn whether investors are able to differentiate efficient from inefficient internal capital markets, and whether ECOs tell us something about the efficiency of capital markets in general. In addition, academics may profit from a number of methodological insights presented in this study regarding both short- and long-term performance analysis. The pragmatically orientated reader may find some passages, in particular the methodological sections, tough going. The disadvantage of a detailed description of methodological issues is that readability of results may be hampered. The advantage is 1

2

On this point see Eco (2005), p. 16-22.

See sections 3.3, 4.3, and 5.3 for comprehensive literature reviews. 3 See section 1.2. 4 See section 1.1 and section 8.3.

VIII

that the relevant sections represent detailed overviews of the current state of knowledge regarding the methods used, and may be helpful for future researchers employing these methods. 5 Different reader groups will thus find value added in different sections. A helpful advice may be to begin with the end: Chapter 8 contains a summarizing conclusion of the key results of this study and may serve as a guide to individual sections of interest. As always, this thesis would not have been possible without ample support from various sides. I thank my supervisor, Prof. Ulrich Hommel, Ph.D., for giving me the opportunity to write this thesis, and for considerable support and valuable comments during its creation. Similarly, I thank Prof. Dr. Dirk Schiereck for acquiescing to co-supervise this thesis and providing a number of useful suggestions. I also thank Prof. Dr. Ralf Elsas for helpful discussions in the conceptual stage of the thesis, as well as for providing access to some crucial empirical data. I thank seminar participants of the 2006 Meeting of the Academy of Economics and Finance in Houston/Texas for constructive observations. Friends and former work colleagues have provided me with a series of comments and recommendations, as well as the appropriate amount of distraction over the last 18 months. My girl friend Carolin has been invaluable in supporting me and has proven an enviable amount of patience and tolerance when listening to my ups and downs while working on this project. Finally, and most importantly, I thank my parents, Dr. Jarmila and Peter Pojezny. Without their continuous manifold support over the last 28 years I would not be where I am today, nor in fact anywhere else. Thank you Mum and Dad.

Nikolas Pojezny

This was pointed out to me by Prof. Balik at the annual meeting of the Academy of Economics and Finance in Houston/Texas in February 2006.

IX

S u m m a r y of Contents List o f A b b r e v i a t i o n s ................................................................................................. X V I I List o f S y m b o l s ............................................................................................................ X I X List o f Tables ............................................................................................................... X X I List o f Figures ........................................................................................................... X X I I I List o f A p p e n d i c e s ..................................................................................................... X X V 1

2

3

4

5

Introduction ...............................................................................................................

1

1.1

Significance o f research object ......................................................................... 1

1.2

Current k n o w l e d g e and research gap ................................................................ 2

1.3

K e y research question and structure o f thesis ................................................... 4

Definitions and theoretical foundations .................................................................... 7 2.1

Description o f research object ........................................................................... 7

2.2

R e a s o n s for e n g a g i n g in an equity carve-out .................................................. 15

2.3

Efficient m a r k e t hypothesis ............................................................................. 18

Short-term price p e r f o r m a n c e o f E u r o p e a n equity ca r v e- o u t s ................................ 34 3.1

Abstr a c t ........................................................................................................... 34

3.2

Introduction ..................................................................................................... 34

3.3

Literature r e v i e w ............................................................................................. 36

3.4

Data and m e t h o d o l o g y .................................................................................... 40

3.5

Empirical results ............................................................................................. 54

3.6

Va r ious extensions .......................................................................................... 72

3.7

C o n c l u s i o n ...................................................................................................... 80

L o n g - t e r m operating p e r f o r m a n c e o f E u r o p e a n equity carve-outs ......................... 82 4.1

Abstract ........................................................................................................... 82

4.2

Introduction ..................................................................................................... 82

4.3

Literature r e v i e w ............................................................................................. 84

4.4

G e ne r a l m e t h o d o l o g y ...................................................................................... 86

4.5

Data and specific analyses .............................................................................. 88

4.6

Empirical results ............................................................................................. 93

4.7

C o n c l u s i o n .................................................................................................... 111

L o n g - t e r m price p e r f o r m a n c e o f E u r o p e a n equity c a r v e- o u t s .............................. 113 5.1

Abstr a c t ......................................................................................................... 113

5.2

Introduction ...................................................................................................

113

X

6

7

8

5.3

Literature r e v i e w ........................................................................................... 115

5.4

General m e t h o d o l o g y .................................................................................... 120

5.5

Data and specific analyses ............................................................................ 125

5.6

Empirical results ........................................................................................... 135

5.7

C o n c l u s i o n .................................................................................................... 160

W h a t do we l e a m about internal capital m a r k e t s f r o m equity carve-outs? ........... 162 6.1

Abstract ......................................................................................................... 162

6.2

Introduction ................................................................................................... 162

6.3

Literature r e v i e w ........................................................................................... 166

6.4

Data and specific analyses ............................................................................ 177

6.5

Empirical results ........................................................................................... 185

6.6

C o n c l u s i o n .................................................................................................... 193

D e t e r m i n a n t s o f the nature o f the s e c o n d event in E u r o p e a n equity carve-outs... 194 7.1

Abstr a c t ......................................................................................................... 194

7.2

Introduction ................................................................................................... 194

7.3

Literature r e v i e w ........................................................................................... 195

7.4

Data and univariate analyses ......................................................................... 199

7.5

Logit regression analyses .............................................................................. 212

7.6

C o n c l u s i o n .................................................................................................... 217

C o n c l u s i o n ............................................................................................................ 219 8.1

S u m m a r y o f k e y findings .............................................................................. 219

8.2

R e c u r r ing t h e m e s and their implications ....................................................... 222

8.3

A v e n u e s for future research .......................................................................... 227

A p p e n d i x ....................................................................................................................... 231 B i b l i o g r a p h y .................................................................................................................. 287

xI

Table of Contents List o f Abbreviations ................................................................................................. X V I I List o f S y m b o l s ............................................................................................................ X I X List o f Tables ............................................................................................................... X X I List o f Figures ........................................................................................................... X X I I I List o f Appendices ..................................................................................................... X X V 1

2

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

Significance o f research object ......................................................................... 1

1.2

Current k n o w l e d g e and research gap ................................................................ 2

1.3

K e y research question and structure o f thesis ................................................... 4

Definitions and theoretical foundations .................................................................... 7 2.1

Description o f research object ........................................................................... 7

2.1.1

Corporate restructuring as the overarching concept .................................. 7

2.1.2

Definition o f an equity carve-out .............................................................. 8

2.1.3

Differentiating an equity carve-out from other forms o f restructuring ..... 9

2.1.4

Primary and secondary equity carve-outs ............................................... 10

2.1.5

Sample identification .............................................................................. 11

2.1.6

Descriptive statistics ............................................................................... 12

2.2

Reasons for engaging in an equity carve-out .................................................. 15

2.2.1

Motivations o f parent firms ..................................................................... 15

2.2.2

Sources o f value creation ........................................................................ 16 Efficient market hypothesis ............................................................................. 18

2.3

3

2.3.1

History o f the efficient market hypothesis .............................................. 19

2.3.2

Basic concepts o f the efficient m a r k e t hypothesis .................................. 20

2.3.3

Definition o f an a n o m a l y ........................................................................ 22

2.3.4

Types o f anomalies ................................................................................. 23

2.3.5

A c a d e m i c c o m m u n i t y ' s reaction to potential anomalies ........................ 26

2.3.5.1

Questioning the a n o m a l y .................................................................... 26

2.3.5.2

Subsuming the a n o m a l y ...................................................................... 28

2.3.5.3

Explaining the a n o m a l y ....................................................................... 28

2.3.6

Potential biases ........................................................................................ 30

2.3.7

Conclusion .............................................................................................. 32

Short-term price performance o f European equity carve-outs ................................ 34

XII

3.1

Abstract ........................................................................................................... 34

3.2

Introduction ..................................................................................................... 34

3.3

Literature r e v i e w ............................................................................................. 36

3.3.1

E v i d e n c e from the U S ............................................................................. 36

3.3.2

E v i d e n c e from G e r m a n y and E u r o p e ...................................................... 38

3.4

Data and m e t h o d o l o g y .................................................................................... 40

3.4.1

Calculation o f a b n o r m a l returns .............................................................. 41

3.4.2

O L S a s s u m p t i o n s .................................................................................... 44

3.4.3

Estimation and event w i n d o w ................................................................. 46

3.4.4

A n n o u n c e m e n t dates ............................................................................... 47

3.4.5

Choice o f significance test ...................................................................... 47

3.5

Empirical results ............................................................................................. 54

3.5.1

Results o f a b n o r m a l return calculation ................................................... 54

3.5.2

N e g a t i v e a b n o r m a l returns following the a n n o u n c e m e n t ....................... 56

3.5.3

Cross-sectional regression analysis ......................................................... 60

3.5.3.1

E x p l a n a t o r y variables .......................................................................... 60

3.5.3.2

Control variables ................................................................................. 67

3.5.4 3.6

M o d e l s ..................................................................................................... 69 Various extensions .......................................................................................... 72

3.6.1

A s s u m p t i o n on m a r k e t p a r a m e t e r stability ............................................. 72

3.6.2

Multiple dates .......................................................................................... 74

3.6.2.1

Three additional dates o f interest ........................................................ 74

3.6.2.2

C o n t a m i n a t i o n o f a n n o u n c e m e n t dates ............................................... 77

3.6.3 3.7

I m p a c t on n o n - a n n o u n c i n g firms ............................................................ 78 C o n c l u s i o n ...................................................................................................... 80

L o n g - t e r m operating p e r f o r m a n c e o f E u r o p e a n equity carve-outs ......................... 82 4.1

Abstract ........................................................................................................... 82

4.2

Introduction ..................................................................................................... 82

4.3

Literature r e v i e w ............................................................................................. 84

4.4

Ge ne r a l m e t h o d o l o g y ....................................................................................... 86

4.5

Data and specific analyses .............................................................................. 88

4.6

Empirical results ............................................................................................. 93

4.6.1 4.6.1.1

Empirical results for p a r e n t firms ........................................................... 93 B a r b e r / L y o n (1996) m e t h o d o l o g y ...................................................... 93

XIII

4.6.1.2

Lie (200 l) m e t h o d o l o g y ...................................................................... 95

4.6.2

Empirical results for subsidiary firms ..................................................... 96

4.6.3

Explanations for patterns in operating performance ............................... 97

4.6.3.1

Market timing or earnings m a n a g e m e n t ............................................. 97

4.6.3.2

Analysis o f potential earnings m a n a g e m e n t ....................................... 98

4.6.4

Explanation o f cross-sectional results for parent firms ......................... 101

4.6.4.1

Independent variables ........................................................................ 101

4.6.4.2

Growth as dependent variable ........................................................... 105

4.6.4.3

Profitability as dependent variable .................................................... 106

4.6.4.4

Interpretation o f results ..................................................................... 107

4.6.5

Explanation o f cross-sectional results for subsidiary firms .................. 108

4.6.5.1

Growth as dependent variable ........................................................... 108

4.6.5.2

Profitability as dependent variable .................................................... 108

4.6.5.3

Interpretation o f results ..................................................................... 109

Conclusion .................................................................................................... 111

4.7

L o n g - t e r m price performance o f European equity carve-outs .............................. 113 5.1

Abstract ......................................................................................................... 113

5.2

Introduction ................................................................................................... 113

5.3

Literature review ........................................................................................... 115

5.4

General m e t h o d o l o g y .................................................................................... 120

5.5

Data and specific analyses ............................................................................ 125

5.5.1

B H A R and selection o f b e n c h m a r k s ..................................................... 126

5.5.2

B H A R and statistical significance ........................................................ 128

5.5.3

Calendar time portfolio m e t h o d ............................................................ 134

5.6

Empirical results ........................................................................................... 135

5.6.1

Post-event performance o f parent firms ................................................ 135

5.6.1.1

B H A R s across various time periods ................................................. 136

5.6.1.2

Statistical significance o f B H A R s ..................................................... 137

5.6.1.3

Calendar time m e t h o d ....................................................................... 138

5.6.1.4 5.6.2

Interpretation o f results ..................................................................... 139 Pre-event performance o f parent firms ................................................. 141

5.6.2.1

Two hypotheses related to pre-event performance ........................... 141

5.6.2.2

B H A R s across various time periods ................................................. 143

5.6.2.3

Interpretation o f B H A R results ......................................................... 144

XIV

5.6.2.4

Testing the first hypothesis: Market timing ...................................... 146

5.6.2.5

Additional test o f the m a r k e t timing hypothesis ............................... 146

5.6.2.6

Testing the second hypothesis: STPP and pre-event L T P P .............. 148 Explanation o f cross-sectional results for parent firms ......................... 149

5.6.3 5.6.3.1

Two hypotheses related to post-event p e r f o r m a n c e .......................... 149

5.6.3.2

Testing the first hypothesis: L T O P causes L T P P ............................. 150

5.6.3.3

Testing the second hypothesis: STPP causes L T P P .......................... 151

5.6.3.4

Multivariate case ............................................................................... 152

5.6.3.5

Interpretation o f results ..................................................................... 152 Post-event performance o f subsidiary firms ......................................... 153

5.6.4 5.6.4.1

Statistical significance o f B H A R s ..................................................... 153

5.6.4.2

Calendar time method ....................................................................... 154

5.6.4.3

Interpretation o f results ..................................................................... 155 Explanation o f cross-sectional results o f subsidiary firms .................... 156

5.6.5 5.6.5.1

Variables in regression ...................................................................... 157

5.6.5.2

Interpretation o f results ..................................................................... 158

Conclusion .................................................................................................... 160

5.7 6

W h a t do we l e a m about internal capital markets from equity carve-outs? ........... 162 6.1

Abstract ......................................................................................................... 162

6.2

Introduction ........................................................... ........................................ 162

6.3

Literature review ........................................................................................... 166

6.3.1

I C M s in the context o f equity carve-outs .............................................. 167

6.3.2

Internal capital markets in the literature .............................. , ................. 168

6.3.2.1

A r g u m e n t s for inefficiency o f I C M s ................................................. 169

6.3.2.2

A r g u m e n t s for efficiency o f I C M s .................................................... 171

6.3.3

Biases in studies on I C M s and the conglomerate discount ................... 173

6.3.4

I C M s and the conglomerate discount across time and g e o g r a p h y ........ 176 Data and specific analyses ............................................................................ 177

6.4 6.4.1

Sources o f data ...................................................................................... 177

6.4.2

Size measures ......................................................................................... 179

6.4.3

Efficiency measures .............................................................................. 182 Empirical results ........................................................................................... 185

6.5 6.5.1

6.5.1.1

Investors' perspective on I C M s ............................................................ 186 Measures o f I C M size from the year prior to the E C O ..................... 186

XV

6.5.1.2

Measures o f I C M efficiency from the year prior to the E C O ........... 187

6.5.1.3

Measures o f changes in I C M size ..................................................... 188

6.5.1.4

Measures o f changes in I C M efficiency ............................................ 188

6.5.1.5

Interpretation o f results ..................................................................... 189 Conditions o f I C M efficiency ............................................................... 190

6.5.2 6.6 7

Conclusion .................................................................................................... 193

Determinants o f the nature o f the second event in E u r o p e a n equity carve-outs... 194 7.1

Abstract ......................................................................................................... 194

7.2

Introduction ................................................................................................... 194

7.3

Literature review ........................................................................................... 195

7.4

Data and univariate analyses ......................................................................... 199

7.4.1

Data sources .......................................................................................... 199

7.4.2

Impact o f individual parameters on the second event ........................... 200

7.4.2.1

Frequency o f second events .............................................................. 200

7.4.2.2

Time until second event .................................................................... 202

7.4.2.3

Stake retained in initial E C O ............................................................ 203

7.4.2.4

Over/undervaluation o f subsidiary firm ............................................ 203

7.4.2.5

Leverage and coverage o f parent firm .............................................. 206

7.4.2.6

I C M size and efficiency .................................................................... 207

7.4.2.7

Industry ............................................................................................. 208

7.4.2.8

Region ............................................................................................... 210

7.5 7.5.1

Logit Model 1: Sell o f f vs. reacquisition .............................................. 212

7.5.2

Logit Model 2: Second event vs. no second event ................................ 215

7.5.3

Interpretation o f results ......................................................................... 216

7.6 8

Logit regression analyses .............................................................................. 212

Conclusion .................................................................................................... 217

Conclusion ............................................................................................................ 219 8.1

S u m m a r y o f key findings .............................................................................. 219

8.2

Recurring themes and their implications ....................................................... 222

8.3

A v e n u e s for future research .......................................................................... 227

Appendix ....................................................................................................................... 231 Bibliography .................................................................................................................. 287

XVII

List of Abbreviations APAR

announcement period abnormal return

AR

abnormal return

ATCF

after tax cash flow

AVA

absolute value-added

BA

book value of assets

BHAR

buy-and-hold abnormal return

BHRR

buy-and-hold raw return

bn

billion

BTM

book-to-market

CAPM

capital asset pricing model

CAR

cumulative abnormal return

CF

cash flow

CR

compound annual return

CRSP

Center for Research in Security Prices

DCF

discounted cash flow

EBIT

earnings before interest and taxes

EBITDA

earnings before interest, taxes, depreciation and amortization

EBT

earnings before taxes

ECM

external capital market

ECO

equity carve-out

EGLS

estimated generalised least squares

EMH

efficient market hypothesis

ew

equal-weighting

FECE

firm excess capital expenditure

GDP

gross domestic product

GLS

generalised least squares

HML

high-minus-low

He

headquarters

ICM

internal capital market

LTOP

long-term operating performance

MAM

market adjusted model

MM

market model

XVIII

MRM

mead adjusted returns model

MSCI

Morgan Stanley Capital International

MTB

market-to-book

NACE

Nomenclature G6n6rale des Acfivit6s Economiques dans l'Union Europ6ene

NPV

net present value

NYSE

New York Stock Exchange

OLS

ordinary least squares

PE

price-to-earnings

PECE

profitability-weighted excess capital expenditure

ROA

return on assets

ROCAA

return on cash-adjusted assets

ROMVA

return on market value of assets

ROS

return on sales

RVA

relative value-added

s.a.

sine anno (without year)

SDC

Securities Data Company

SEO

seasoned equity offering

SIC

Standard Industrial Classification

SMB

small minus big

STPP

short-term price performance

TA

total assets

TAC

total accruals

vw

value-weighting

WACC

weighted average cost of capital

WLS

weighted least squares

XIX

List of S y m b o l s expected value E

expectation operator

C

Euro

i

used to index firms or securities (unless otherwise indicated)

n

number of firms in the sample (unless otherwise indicated)

P

price of a security information set at time t

(3

standard deviation

P

price

r

return

t

used to index time

T

number of days in the estimation period

V

variance-covariance matrix excess return

XXI

List of Tables Table 1: Descriptive statistics .........................................................................................

14

Table 2: CARs across various event day windows ......................................................... 54 Table 3: Sample firms used in event study ..................................................................... 55 Table 4: Tests for hypotheses regarding [+2] day window abnormal returns ................ 59 Table 5: CARs across countries ...................................................................................... 61 Table 6: Country classification as explanatory variable for CARs ................................. 61 Table 7: Regional classification as explanatory variable for CARs ............................... 63 Table 8: Stated motivation as explanatory variable for CARs ........................................ 64 Table 9: Difference between companies with and without announcement ..................... 65 Table 10: Industry classification as explanatory variable for CARs ............................... 67 Table 11: Cross-sectional regression of CARs ............................................................... 70 Table 12: Average reaction of non-announcing companies: Grand subgroup averages. 79 Table 13: Number of p-values in excess of Bonferoni-adjusted alpha level .................. 80 Table 14: Total accruals/total assets by relative event year .......................................... 100 Table 15: Parent firm B H A R across various periods .................................................... 136 Table 16: Number of sample companies in calendar time method ............................... 138 Table 17: Calendar time method for parent firms ......................................................... 139 Table 18: Pre-event parent firm B H A R across various periods .................................... 143 Table 19: Regression of LTPP on 12-month pre-event LTPP ...................................... 146 Table 20: Change in valuation levels of subsidiary firm's and other industries ........... 147 Table 21: OLS-regression of LTPP on STPP ............................................................... 15 l Table 22: Subsidiary firm BHAR across various periods ............................................. 153 Table 23: Calendar time method tbr subsidiary firms .................................................. 154 Table 24: Average time until second event ................................................................... 202 Table 25: Stake retained after initial ECO and second event ........................................ 203 Table 26: Over/undervaluation of subsidiary firm and second event ........................... 205 Table 27: Leverage, coverage and second event ........................................................... 207 Table 28: Country financial development and second event ........................................ 211

XXIII

List of Figures Figure 1: C A R s around event d a t e - [-10;+10] day w i n d o w .......................................... 55 Figure 2: Correlation o f abnormal returns: Trend reversal ............................................. 58 Figure 3: A b n o r m a l returns as a function o f the post-event stake .................................. 66 Figure 4: Difference between pre- and post-event beta .................................................. 72 Figure 5: Parent firm post-event B H A R ....................................................................... 137 Figure 6: Parent firm pre-event B H A R ......................................................................... 143

XXV

List of Appendices Appendix 1: Overview of various restructuring and financing mechanisms ................ 232 Appendix 2: Derivation of sample ................................................................................ 233 Appendix 3: Sample by year and country ..................................................................... 234 Appendix 4: Sample by year (relative to IPOs) ............................................................ 235 Appendix 5: MSCI Europe 1984-2004 ......................................................................... 236 Appendix 6: Sample by year (money volume) ............................................................. 237 Appendix 7: Sample by industry (relative to benchmark universe) .............................. 238 Appendix 8: Sample by year and industry .................................................................... 239 Appendix 9: Sample by country (relative to IPOs) ....................................................... 241 Appendix 10: Existing studies on ECO announcement period returns - US ................ 242 Appendix 11: Existing studies on ECO announcement period returns - Non-US ....... 243 Appendix 12: Hypotheses on value creation ................................................................. 244 Appendix 13: Abnormal return on individual days around event ................................. 245 Appendix 14: CARs around event date - [-50;+20] day window; ................................ 246 Appendix 15: Abnormal returns per year ...................................................................... 247 Appendix 16: Abnormal returns per period .................................................................. 248 Appendix 17: CARs and significance - Betas from pre- and post-event period ........... 249 Appendix 18: CARs on four different dates ................................................................. 250 Appendix 19: CARs of excluded companies ................................................................ 251 Appendix 20: Average reaction of non-announcing companies- Subgroups ............... 252 Appendix 21: ROA rank of parent firms ...................................................................... 253 Appendix 22: Size rank of parent firms ........................................................................ 253 Appendix 23: ROA/size rank combinations ................................................................. 254 Appendix 24: MTB rank of parent firms ...................................................................... 254 Appendix 25: Benchmark universe composition .......................................................... 255 Appendix 26: Parent firm abnormal profitability- Barber/Lyon (1996) method ......... 256 Appendix 27: Parent firm abnormal growth- Barber/Lyon (1996) method ................. 257 Appendix 28: Algorithms used and firms included in sample ...................................... 258 Appendix 29: Parent firm abnormal profitability- Lie (2001) method ........................ 259 Appendix 30: Subsidiary firm abnormal profitability - Barber/Lyon (1996) method ..260 Appendix 31: Subsidiary firm abnormal growth - Barber/Lyon (1996) method ......... 261 Appendix 32: Explanation of parent firm growth .......................................................... 262

XXVI

Appendix 33: Explanation of parent firm profitability ................................................. 263 Appendix 34: Explanation of subsidiary firm growth .................................................. 264 Appendix 35: Explanation of subsidiary firm profitability ........................................... 265 Appendix 36: Annual portfolio composition for three-factor model ............................ 266 Appendix 37: Parent firm post-event B H A R - Significance of results ......................... 267 Appendix 38: Parent firm pre-event B H A R - Significance of results .......................... 268 Appendix 39: Regression of STPP on 12-month pre-event LTPP ................................ 269 Appendix 40: Significant regression coefficients LTPP / LTOP (growth) ................... 270 Appendix 41: Significant regression coefficients LTPP / L T O P (profitability) ........... 271 Appendix 42: Explanation of parent firm post-event LTPP ......................................... 272 Appendix 43: Subsidiary firm post-event BHAR - Significance of results .................. 273 Appendix 44: Subsidiary firm post-event abnormal BHAR ......................................... 274 Appendix 45: Explanation of subsidiary firm post-event LTPP ................................... 275 Appendix 46: Illustrative example for Billet/Mauer (2000) ICM measure .................. 276 Appendix 47: Explanation of abnormal returns using ICM measures .......................... 277 Appendix 48: Explanation of abnormal returns using the change in ICM measures .... 278 Appendix 49: Regression of CARs on cash flow correlation ....................................... 279 Appendix 50: Event day abnormal returns vs. cash flow correlation ........................... 280 Appendix 51: Nature of second events in previous studies .......................................... 281 Appendix 52: ICM size and efficiency and second event ............................................. 282 Appendix 53: Same/cross-industry ECOs and second event ........................................ 283 Appendix 54: Industry association and second event ................................................... 284 Appendix 55: Logit model I: Sell-off vs. reacquisition ................................................ 285 Appendix 56: Logit model II: Second event vs. no second event ................................. 286

1

Introduction

To appreciate the relevance of the present thesis, the introductory chapter highlights the practical significance of the research object, as well as its appeal to academics (section 1.1). The current status of knowledge regarding the research object is outlined, and key gaps in understanding are identified (section 1.2). The chapter concludes with an outline of the key research questions and a description of the structure of the thesis (section 1.3).

1.1

Significance of research object

Corporate restructuring in general and portfolio restructuring in particular have become an important part o f corporate life in Europe. The total volume of exchange-listed disinvestments in Europe from 1990 to 1998 amounted to approx. C100 billion. 6 The general motive for these transactions is (or should be) the desire to create value for company shareholders by focussing on core businesses, disposing of poorly performing divisions, eliminating negative synergies between unrelated business segments, creating pure-play companies easier to evaluate for investors, and reducing the debt burden. 7 Equity carve-outs (ECOs) are a popular instrument in a firm's portfolio restructuring toolbox. This study identifies 178 ECOs in 13 European countries in the time period from 1/1/1984 to 31/12/2004, with a total money volume of approx, t~91 billion, s This compares to a total initial public offering (IPO) volume in the same countries and over the same time period of approx. (~580 billion. 9 A series of recent high-profile transactions in Europe have brought ECOs onto the radar screen o f both firms and academics. ~~ From an academic's point of view, ECOs are fascinating because they combine elements of two distinct corporate restructuring mechanisms. ~ First, an ECO is similar

See Glatzel (2003), p. 4. Converted to • at the average of daily exchange rates from 1/1/1990 to 31/12/1998 (approx. US$/t~ 1.25). 7 See Gaughan (2002), p. 395-403. 8 The ECO volume is calculated as the market capitalisation of the subsidiary on the first day of trading, multiplied by the percentage stake sold by the parent company. All data points required for this calculation could be identified for 151 out of the 178 sample firms. The total ECO volume is based on the total identified ECO volume, plus an estimate of the average ECO size multiplied by 27 (i.e., 178151). 9 The IPO volume estimate is based on all capital-raising transactions listed in SDC for the 13 European countries in which ECOs have been identified, with a transaction date from 1/1/1984 to 31/12/2004, IPO flag marked as 'yes', and excluding repeatedly listed transactions. The US$ amount is converted to C at day-end exchange rates as of the date of the respective transaction. 10 Examples include the floatation of T-Online by Deutsche Telekom (2000), Epcos and Infineon by Siemens (1999 and 2000, respectively), Deutsche Postbank by Deutsche Post (2004) in Germany, Kemira GrowHow by Kemira (2004) in Finland, Pages Jaunes by Wanadoo (2004) in France, Terna by Enel (2004) in Italy, Cintra by Grupo Ferrovial (2004) in Spain, Converium by Zurich Financial Services (2001) in Switzerland, and Burberry by GUS (2002) in the UK. 11 See chapter 2.1.2 for a detailed definition of ECOs and other corporate restructuring activities.

to financial restructuring mechanisms like IPOs and seasoned equity offerings (SEOs) in that a subsidiary firm goes public, and cash is raised. Second, an ECO is similar to portfolio restructuring mechanisms like spin-offs and divestitures in that the composition of the parent firm's assets changes, and control over the subsidiary firm is transferred from the parent firm to shareholders. This dual nature renders ECOs a uniquely interesting research object, providing a range of potential research questions. ECOs can also serve as 'natural experiments' for addressing more general questions about the level of efficiency of internal capital markets. 12 ECOs thus represent a prominent transaction mechanism for European firms, allowing them to (partially) exit non-core business areas, and to use the released capital for investment into core businesses, the repayment of existing debt, or for other financing purposes. Academics are attracted by the dual nature of ECOs as both a portfolio and a financial restructuring mechanism. Hence, both the volume of and interest in ECOs will continue to be substantial.

1.2

Current knowledge and research gap

Previous research has established some important knowledge regarding ECOs. Announcements of intended ECOs on average lead to positive abnormal returns. 13 Two sets of explanations are offered: First, according to the divestiture gains hypothesis, value gains arise because the business focus of both parent and subsidiary firm increases following the ECO, cash proceeds can be used to retire debt, the carved-out entity is able to separately finance its investment projects and is more likely to be taken over, the information availability regarding the subsidiary firm's performance increases, investors are more inclined to invest into the new pure-play stock, and managers' contracts can be designed more efficiently. TM Second, according to the asymmetric information hypothesis, issuing shares in the subsidiary firm signals an undervaluation of the larger parent firm assets and an overvaluation of the smaller subsidiary firm assets. Investors use this information and buy shares in the parent firm, leading to positive returns. 15 Announcement period returns are higher on average when parent and subsidiary firms are from different industries 16, when pre-event informational asymmetry is high 17, when subsidiary firm assets are greater than non-subsidiary firm assets TM, when the ECO is conducted as a primary (rather than a secondary) offering 19, and when the parent firm

12 See chapter 6. 13 See chapter 3 for a literature review. 14 See Schipper/Smith (1986), p. 169-175 and Vijh (2002), p. 164-165. 15 See Nanda (1991) and Slovin/Sushk~Ferraro (1995) for a detailed description of the asymmetric information hypothesis. 16 See Vijh (2002), p. 177. 17 See Elsas/L/Sffler (2005), p. 15. 18 See Vijh (2002), p. 155. 19 See Kaserer/Ahlers (2000), p. 562-564.

uses the proceeds to repay debt 2~ Parent and subsidiary firm operating performance improves in the first year following the EGO 21, with the subsidiary firm's operating performance declining again in the following years. 22 Carved-out entities tend to be the high-growth divisions of the parent firms. 23 In many cases ECOs are temporary structures, and the parent firm either reacquires or completely sells off its partially floated subsidiary firm in later years. 24 An ECO thus creates a real option, allowing the parent firm to profit from the resolution of uncertainties. 2s Also, investors seem partially able to anticipate the second event: ECO announcements followed by an eventual take-over of the carved-out entity by a third party produce higher abnormal returns than announcements not followed by a takeover. 26

Simultaneously, existing research fails to answer several aspects of value creation in ECOs. Regarding short-term price performance, it is unclear whether announcement returns differ across time and geography, whether there is a systematic price reaction at additional dates during the ECO process, and whether non-announcing firms with similar ECO candidates experience abnormal reactions to other firms' ECO announcements. Regarding long-term operating performance, it is unclear why the subsidiary firm's operating performance, after peaking around the time of the ECO, deteriorates in later years. The parent firm's operating performance has not been analysed in a multi-year window around the event. Evidence of earnings management around classical IPOs suggests that a similar phenomenon may occur in ECOs, which has not been analysed so far. Also, it is unclear whether ECO characteristics can explain the cross-sectional distribution of performance results: Answering this question will help firms to design ECOs more efficiently. 27 Regarding long-term price performance, evidence is mixed. Parent firms are found to underperform 28, to perform in line with 29, and to outperform 3~ benchmark firms.

20

See Allen/McConnell (1998), p. 165. See Hulburt/Miles/Woolridge (2002), p. 95-99. 22 See Powers (2003), p. 32. 23 See Powers (2003), p. 40. 24 See Klein/Rosenfeld/Beranek (1991), p. 450. 25 See Kranenburg/Perotti/Rossetto (2004) for a description of ECOs as real options. 26 See Hulburt (2003), p. 30. 27 For example, it is unclear how the stake retained relates to subsequent operating performance. While Boone/Haushalter/Mikkelson (2003) find that parent firm operating performance improves only when the entire stake is carved out, and subsidiary firm operating performance is unaffected, Powers (2003) finds a negative relation between subsidiary firm operating performance and the percentage of shares sold. 28 See Madura/Nixon (2002), p. 172. Negative performance is exclusive to distressed parents. 29 See Vijh (1999), p. 285-290. 30 See Anslinger/Carey/Fink/Gagnon (1997), p. 166.

21

Similarly, subsidiary firms are found to underperform IPO 31 and benchmark firms 32, to perform in line with IPO 33 and benchmark firms 34, and to outperform benchmark firms 35. Discrepancy in results is driven by differing test designs resulting from the disagreement about an appropriate methodology. Further, it is unclear whether ECO characteristics can explain the cross-sectional distribution of results. Also, the relationship between long-term operating performance and long-term price performance has not been analysed. Regarding internal capital markets (ICMs), there exist empirical studies and theoretical models supporting the notion both of efficient and of inefficient ICMs. Discrepancy in results suggests that the relative level of ICM efficiency depends on specific firm characteristics, rather than being low or high in general as suggested by some of the models. Previous studies use spin-offs and asset sales to analyse ICMs, while ECOs have not been used to that avail. It is unclear whether investors are able to discern efficient from inefficient ICMs, and what the conditions are required for an ICM to be efficient. Regarding the second event, the determinants of the decision of whether to reacquire or completely sell off the subsidiary firm have not been analysed in detail. It is unclear whether factors such as the parent firm's leverage, the subsidiary firm's valuation level, industry association, and the institutional setting of the parent firm's home country influence the decision. This study addresses these open issues.

1.3

Key research question and structure of thesis

Three key research questions form the framework for the entire thesis. 9

What are the sources of value creation in European equity carve-outs, both in the short term and in the long term? 9 What can researchers learn about the efficiency of internal capital markets with the help of equity carve-outs? 9 What factors determine the eventual wind-up of ECO structures through reacquisitions or sell-offs? The structure of the thesis is based on these research questions and the identified research gaps related to the short-term price performance following the announcement 31 See Prezas/Tarimcilar/Vasudevan (2000), p. 130-134. 32 See Madura/Nixon (2002), p. 172. Negative performance is exclusive to subsidiaries carved out of distressed parent firms, 33 See Schikowsky/Schiereck/V61kle/Voigt (2005). 34 See Vijh (1999), p. 285-290. 35 Annema/Fallon/Goedhard (2002) find that subsidiaries gaining full independence outperform the S&P500, while subsidiaries remaining under the control of the parent underperform the S&P500 in the two years after the event. Powers (2003) finds a positive performance in his sample of 181 ECOs in the three years after the event, but results are due to positive first year performance. In each of the years 2-5, sample firms underperform.

of the ECO (chapter 3), long-term operating and price performance of parent and subsidiary firms following an ECO (chapter 4 and chapter 5), the efficiency of internal capital markets using ECOs as research objects (chapter 6), and the nature and determinants of the second event following the initial ECO (chapter 7). Each of these chapters is structured similarly: Following an introductory paragraph detailing the specific research question 36, the relevant literature is reviewed, followed by a description of the data and methodology used, a presentation of the empirical results and their economic interpretation, and a summarizing conclusion. The following paragraphs provide an overview of the specific issues addressed in each chapter. Chapter 2 has two objectives. First, it defines the research object and differentiates it from similar portfolio and financial restructuring mechanisms. Second, the efficient market hypothesis is presented, and potential violations are discussed. Market efficiency is the yardstick against which abnormal price performance is assessed. Chapter 3 analyses the short-term price performance of firms announcing an ECO. An event study framework is used to address several questions. First, are positive abnormal returns constant across time and countries? If capital markets become more efficient across time, the changing relative value of internal vs. external capital markets may impact the level of abnormal returns. Similarly, abnormal returns may differ as a function of cross-country varying development states of capital markets. Second, is there a pattern in returns across three additional dates (first rumour date, bookbuilding date, first trading date) on which markets receive information about the impending ECO? Since the last two dates are known in advance, any systemic pattern could yield profitable trading strategies. Third, do abnormal returns differ on 'clean' vs. 'contaminated' announcement dates? Companies may choose to link ECO announcements to other news, causing part of the generally claimed announcement period returns to be attributable not to the ECO itself. Fourth, do non-announcing companies with future ECO candidates show abnormal price reactions to ECO announcements by other firms? Investors could either sell the non-announcing firms, because they have not yet performed an ECO, or buy them as future ECO candidates. Chapter 4 analyses long-term operating performance (LTOP) of parent and subsidiary firms. According to the divesture gains hypothesis, operating performance should improve following the ECO. Firms could also engage in earnings management (similar to previous findings for IPOs and SEOs), by using discretionary accounting mechanisms to render themselves more attractive to capital markets, followed by a decline in earnings as the measures are reverted. LTOP is analysed using growth and profitability measures. The cross-section of LTOP is analysed in a multivariate regression framework as a function of various event and firm characteristics to identify the sources of positive operating performance development.

In addition to the introductory section each main chapter is preceded by a journal-type abstract succinctly summarizingthe key findings of the respectivechapter.

Chapter 5 analyses long-term price performance (LTPP) of parent and subsidiary firms. Extant studies offer contradicting results because of a lacking consensus on how to appropriately assess abnormal LTPP. Consequently a large variety of methodologies are applied to assure robustness of results. In addition to post-event LTPP, pre-event LTPP is also analysed. Two hypotheses are tested: First, is there a relationship between preevent LTPP and short-term price performance? A positive short-term price performance could be merely a reaction to a negative pre-event LTPP; alternatively a positive shortterm price performance could be positively linked to pre-event LTPP, indicating that the latter serves as a signalling mechanism reducing informational asymmetry between parent firm and future investors. Second, the relationship between pre- and post-event LTPP is analysed to assess whether managers market-time the ECO to occur in periods of high relative prices and valuation levels. The cross-section of LTPP is analysed in a multivariate regression framework as a function of various event and firm characteristics to identify the sources of positive price development. Chapter 6 addresses two specific questions regarding the efficiency of internal capital markets (ICM) using ECOs as a research object. First, how do investors view ICMs? Announcement period returns are regressed on ICM size and efficiency measures, controlling for other factors influencing abnormal returns as previously identified. If investors view ICMs negatively, parent firms with larger and less efficient ICMs should experience higher announcement period returns. In addition to absolute ICM measures, change measures are used to assess whether changes in ICM size and efficiency are related to announcement period returns. Second, what are the conditions for ICM efficiency? The mixed results in existing literature on ICM efficiency suggest that firmspecific factors are likely to influence relative ICM efficiency. Literature offers two contradictory views on when ICMs are efficient, ultimately differing in their conclusion on whether firms with related or unrelated business segments are more likely to have efficient ICMs. These hypotheses are tested by regressing announcement period returns on the correlation between parent and subsidiary firm cash flows. Chapter 7 addresses two key questions regarding the second event: First, what is the frequency of reacquisitions vs. sell-offs in Europe, relative to the US? If there is tradeoff between internal and external capital markets, and the latter are more developed in the US than in Europe, ICMs are more valuable in Europe, and reacquisitions (i.e., reestablishments of partially closed ICMs) should be more frequent in Europe. Second, what are the determinants of the second event decision? Hypotheses on the impact of relative subsidiary firm valuation levels, the parent firm's debt burden, ICM size and efficiency, the parent firm's industry and country, and the development state of financial markets are developed and tested. Chapter 8 summarises and discusses the key findings of the empirical analyses in the present study, identifies recurring themes, and highlights areas for future research.

2

Definitions and theoretical foundations

This chapter has two objectives. First, it defines the research object (section 2.1) and analyses empirically identified and theoretical reasons of why firms engage in an ECO (section 2.2). Second, it presents the basic concepts of the efficient market hypothesis and reviews evidence for and against market efficiency (section 2.3).

2.1

Description of research object

This section proceeds by first placing the specific research object into a general context (section 2.1.1). ECOs are then defined in detail by listing typical characteristics (section 2.1.2), by differentiating them from similar portfolio and financial restructuring mechanisms (section 2.1.3), and by distinguishing two forms of ECOs (section 2.1.4). The construction of the sample is described (section 2.1.5), and some summary descriptive statistics are provided (section 2.1.6).

2.1.1

Corporate restructuring as the overarching concept

Corporate restructuring can broadly be defined as "a major change in the composition of a firm's assets combined with a major change in its corporate strategy ''3v. It has attracted the attention of scholars in various fields, including financial, management and organizational research. The general goal of corporate restructuring, as evidenced by many company announcements, is to create shareholder value. 3s According to Bowman/Singh (1993), corporate restructuring comprises three different dimensions: First, portfolio restructuring refers to material changes in the firm's asset holdings through acquisitions, divestitures, liquidations, spin-offs and equity carve-outs. Second, financial restructuring refers to material changes in a firm's capital and ownership structure through public-to-private transactions (e.g., leveraged buyouts), private-topublic transactions (IPOs), leveraged recapitalizations and debt-to-equity swaps. Third, organizational restructuring refers to material changes in the firm's structure and the redesign of hierarchies. As pointed out by Heugens/Schenk (2004), organizational restructuring often follows in the wake of portfolio or financial restructuring. 39 A survey of existing studies by Bowman/Singh/Useem/Bhadbury (1999) finds that financial restructuring yields the most tangible returns, while portfolio restructuring also yields positive but on average lower returns. The value effects of organizational restructuring are more mixed and dependent on the specific circumstances. As indicated by Bowman/Singh (1993), an ECO seems most often associated with portfolio restructuring. However, ECOs are also a type of IPO (namely of the subsidiary

37 Hoskisson/Turk (1990), p. 459. 38 See Bowman/Singh (1993), p. 6. 39 See Heugens/Schenk (2004), p. 88.

firm), i.e., a mechanism of financial restructuring for the parent firm of the carved-out subsidiary firm. This dual nature renders ECOs both a unique measure for parent and subsidiary firms, and an interesting and complex research topic for academics.

2.1.2

Definition of an equity carve-out

Schipper/Smith (1986) define an equity carve-out (ECO) as "the initial public offering of some of the stock of a wholly owned subsidiary ''4~ Similarly, Vijh (1999) states that "in an equity carve-out, a parent firm raises money by selling part or all of the equity in a wholly owned subsidiary to the public ''41. For the purpose of this study, I define an ECO as follows:

An equity carve-out (ECO) is the initial public offering by an exchange-listed parent firm of shares in a majority-controlled legally separated subsidiary firm to the public. The individual elements of the definition are explained next. First, an ECO is an initial public offering, i.e., a going public, of a subsidiary firm. Therefore practitioners also refer to ECOs as 'sub-IPOs'. A going-public is defined using five criteria42: A privately held company is converted into a (totally or partially) publicly held company; the equity is offered to a broad range of investors; it is possible for the first time for investors to buy shares in the company; the shares are offered in a secondary market, allowing regular trading to take place; and the transaction leads to a cash inflow for either the issuing company, the selling shareholders, or both. Second, the parent firm is exchange-listed. This requirement is important for this study because one of the research objectives is to analyse the long-term value impact of an ECO. Value is measured as the share price development of the parent firm. Third, the subsidiary firm must be majority-controlled, i.e., the parent firm's ownership of the voting capital must be in excess of 50%. This is more relaxed than the 'whollyowned' requirement by Vijh (1999), and takes into consideration that in some cases there may exist (non-listed) minority positions in the subsidiary firm. This definition is in line with Kaserer/Ahlers (2000). 43 Fourth, the subsidiary firm must be a separate legal entity. This separate legal entity has either existed prior to the announcement of the ECO, or is created in the process leading up to the public offering. Fifth, the offer is made to the public. Generally, this takes the form of an offer to the general public. However, it does not preclude the parent firm from favouring existing shareholders in the allocation of shares. Indeed, there are demands for the introduction 40 Schipper/Smith (1986), p. 154. 41 Vijh (1999), p. 274. 42 See Mettler (1990), p. 19. 43 See Kaserer/Ahlers (2000), p. 552.

of a pre-emptive right for existing parent firm shareholders allowing them to buy shares in the subsidiary firm before non-parent firm shareholders are allowed to do SO. 44 The aim is to prevent non-parent firm shareholders from profiting from positive initial trading day returns, at the (opportunity) cost of parent firm shareholders who have not received subsidiary firm shares in the allocation process. In Germany, however, the consensus among legal scholars up to recently was that neither existing law nor past judicial decisions allow deducing a general claim by parent firm shareholders to receive preferential treatment in the allocation of shares. Recently this view has become more contentious. 45 Sixth, the definition does not make a statement about the stake held by the parent firm following the ECO: Usually, the parent firm will either retain a majority or a minority stake. In some cases, it may also decide to carve out the entire stake.

2.1.3

Differentiating an equity carve-out from other forms of restructuring

Appendix 1 summarises some of the key differences between various forms of portfolio and financial restructuring. An ECO is different from all other forms of portfolio and financial restructuring in that it combines aspects of both of these measures, while most other mechanisms have a strong tendency to be classified as either restructuring or financing. This dual nature of ECOs implies the necessity to take into consideration the prime motivation of the parent firm in carrying out the ECO when analysing short and long-term performance. An ECO differs from a spin-off in at least three aspects46: First, in a spin-off existing parent firm shareholders receive shares in the subsidiary firm as a special dividend, whereas in an ECO these shares are sold to new shareholders. Second, a spin-off generally does not result in a cash flow to either parent or subsidiary firm, whereas an ECO leads to cash inflows for either the parent firm, the subsidiary firm, or both. Third, a spin-off usually results in a complete separation, whereas in an ECO the parent firm in most cases retains a stake in the carved-out entity. 47 An ECO differs from a seasoned equity offering (SEO) in at least two aspects48: First, in an SEO a parent firm sells its own shares, whereas in an ECO shares of its subsidiary firm are sold. Second, shares of the parent firm have been trading before an SEO, whereas shares of the subsidiary firm have not been trading before an ECO.

44

See Paul (2003), p. 527-529. 45 See Wunderlich (2003), p. 23-26 for more details. 46 See Michaely/Shaw (1995), p. 5-6. 47 In this study's sample, only 13 of the 178 sample parent firms sell their entire stake in the subsidiary in the initial ECO. 48 See Schipper/Smith (1986), p. 153-154.

10

A sell-off is usually a complete divestiture of a business unit. 49 In contrast to an ECO, the selling parent firm does not retain a stake in the business unit, the cash flow always goes to the parent firm, and there is usually no immediate trading in the secondary market if the buyer consolidates the acquired asset as its own subsidiary. A tracking stock 5~ is a stock issued by a parent firm whose value is contingent on the development of the business of a particular subsidiary, which is not legally split from the parent firm as it is in an ECO. Tracking stock is issued by a parent firm when it wants to retain the synergies between itself and the subsidiary firm, but also wants to allow its shareholders to profit from a potentially stronger development of the subsidiary relative to the parent firm. It also allows the compensation of subsidiary firm managers to be tied to the performance of the subsidiary firm, rather than to the whole group. 51 Tracking stocks have so far mostly been issued in the US. 52 The only European tracking stock issue, floated by Alcatel on its opto-electronics business unit in October 2000, was re-converted into ordinary parent firm shares in April 2003. 53 In a split-up, the parent firm transfers all of its assets unto two or more new firms, and parent firm shareholders receive shares in each of these new firms. The old parent firm ceases to exist. In a split-off, parent firm shareholders tender their shares against new 54 shares in a subsidiary firm. Both of these types of restructuring are rather rare events. The choice of which portfolio or financial restructuring measure to employ in a specific situation depends, in addition to the considerations listed above, on a series of countryspecific regulatory issues and tax considerations. Given the multi-country nature of the sample, a detailed analysis of these questions is beyond the scope of this study. 55

2.1.4

Primary and secondary equity carve-outs

ECOs can be structured in three basic ways, leading to different allocations of the capital raised. 56 First, a subsidiary firm may increase its capital without the participation of the parent firm, leading to a cash inflow for the subsidiary firm (primary ECO). Second, the parent firm may sell part of or its entire stake in the subsidiary firm, leading to a cash inflow for the parent firm (secondary ECO). Third, a mix of both primary and secondary ECO is feasible. The choice of how to structure an ECO depends on the 49 See Gaughan (2002), p. 397-403. 50 Synonyms for tracking stock are targeted stock, alphabet stock and letter stock (see Bauer (1992), p.

33).

51 See Billett/Mauer (2000), p. 1459. 52 General Motors issued the first tracking stocks in 1984 for its EDS subsidiary. See Gaughan (2002), p. 423. 53 See Alcate12004 20F, p. 13. 54 See Hite/Owers (1983), p. 411, footnote 3 and Schipper/Smith (1983), p. 483, footnote 2. 55 See Glatzel (2003), p. 111-123, for a description of tax and other regulatory effects of equity carveouts, spin-offs and tracking stocks in the US and in Germany. 56 See Kaserer/Ahlers (2000), p. 540-541.

11

objective of the parent firm: A primary ECO is usually preferred if the parent firm aims to increase the capital base of its subsidiary to allow it to exploit its growth potential, while at the same time retaining a stake in the carved-out entity. A secondary ECO is usually preferred if a parent firm aims to increase its own capital base, and/or if its main aim is to divest the subsidiary firm. In this study's sample, 48 (27.0%) firms perform a primary offering, 50 (28.1%) firms perform a secondary offering, 54 (30.3%) perform a mixed offering, and for the remaining 26 (14.6%) firms the variable could not be determined, mostly because the IPO prospectus was not available.

2.1.5 Sample identification Since there is no central database explicitly listing ECOs, they are identified using a variety of sources. First the SDC database is searched for all European capital-raising occurrences between 1/1/198457 and 31/12/2004 (n=20,067). Appendix 2 details how companies are deleted: All cases not marked as IPOs (n=12,794), with no or a transaction value below US$10 million (n=2,389), marked as follow-on offerings (n=86), and where the issuing entity is either indicated to be a state body or a financial investor (n=651) are deleted, leaving 4,147 cases. All transactions where the parent firm and the issuing company have the same name (n=3,015) are deleted. 5s The remaining 1,132 cases are complemented by all publicly available lists of German and European ECOs. 59 For all cases newslines (LexisNexis and Factiva) are checked to determine whether they qualify as an ECO. This procedure identifies 178 ECOs in 13 European countries. Appendix 3 shows the composition of the sample by announcement year and by the parent firm's home country. The European setting is chosen because one of the main concerns with previous German and European studies is small sample size. Also, some of the empirical analyses in this study are fairly data-intensive, leading to a reduction in the actual sample size used in the respective analyses due to limited data availability. This reduction would likely decimate e.g., a German-only sample to a statistically undesirable size. Additionally, while the scope of the study limits the possibilities of analysing country-specific differences in detail, different measures of the development of each country's capital market and institutional settings are used as explanatory variables in some of the analyses.

The starting point is chosen arbitrarily, but also taking into consideration that the first ECO in Germany occurred in 1984. 58 The relevant database fields are "Ultimate parent company" and "Issuer". The assumption is that if the same name is listed in both fields, the transaction represents a classical IPO, rather than an ECO, in which case the issuer name is the name of the subsidiary company (irrespective of whether the ECO is conducted as a primary or a secondary offering). 59 These sources comprise lists in Pellens (1993), Hasselmann (1997), Kaserer/Ahlers (2000), Langenbach (2001), L6ffler (2001), Stienemann (2003), Bfihner (2004), Wagner (2004) and Elsas/L6ffler (2005).

57

12

2.1.6

Descriptive statistics

Appendix 4 shows the total number of ECOs in each year of the sample period, and compares it to the total number of IPOs in the same period. The comparison of ECO and IPO numbers shows that years of relatively high IPO activity also tend to be years of relatively high ECO activity. The number of ECOs trends upward in a cyclical fashion: A first peak of five ECOs occurs in 1989, followed by a subsequent drop coinciding with the decline in the MSCI Europe in the second half of 1990 as shown in Appendix 5. A second peak occurs in 1993 and 1994 with 13 ECOs, followed by a drop in 1995, again paralleling the general stock market development. The numbers then trend strongly upward until the height of the stock market bubble in 2000, when 35 ECOs are announced. Incidentally, 2000 is also the year with the highest volume of global M&A activity in history. 6~ The subsequent drop coincides with the drop in many stock price indices in 2001, and ECO numbers remain low until the end of the sample period. Appendix 6 shows the total money volume (in Cbn) of ECOs in each year of the sample period. The major upward trend until 2000 is confirmed, though the interpretation in individual years may differ. This is a result of considerable size differences in individual ECOs. There is also an upward trend in ECO volume in 2004, which does not show in the number of ECOs, due to a few above-average size ECOs in 2004. Thus, not surprisingly, the frequency of ECOs seems linked to general stock market sentiment and overall IPO activity. Appendix 7 shows the distribution of the sample firms across the 18 industries represented in the sample, where industry is defined on the two digits NACE code level. 61 The table also compares the sample firms' percentage distribution across the industries with the distribution of the benchmark universe, consisting of all listed firms for which industry codes are available on Datastream in the 13 European countries represented in the study. Overall, the sample ECOs are distributed similarly across the industries, compared to the overall benchmark universe. ECOs are slightly underrepresented in the food & beverage (NACE code 35) and the personal & household goods (NACE code 37) industries, and over-represented in the telecommunication (NACE code 65) and utilities (NACE code 75) industries. Overrepresentation in these two industries could result from the respective industry deregulation occurring across many European countries over the last two decades: Companies facing new competition are looking for ways to become more competitive themselves, and a potential way of achieving this is to restructure their portfolio of activities. To further analyse this finding, the upper part of Appendix 8 shows the distribution of the sample firms in each year across the 18 industries. The lower part of the table shows the deviation of the actual to the expected number of sample firms in According to Bloomberg, global M&A deal volume in 2000 was approx. US$2.8 trillion. The 2 nd best year is 2005, with a global M&A deal volume of approx. US$2.7 trillion (Bloombergfunction: MA). Defining industry on the two digits industry code level is standard in empirical literature; see, e.g., Barber/Lyon (1996). Industry codes are based on the NACE (Nomenclature G6n6rale des Activit6s Economiques dans l'Union Europ6ene) system, rather than the SIC (Standard Industrial Classification) system, which is popular in US papers. First, NACE seems more appropriate, given the European nature of the sample. Second, for European companies, availability of NACE codes in Datastream is far superior to availability of SIC codes.

13

each cell, where the expected number is calculated as (sum across rows)*(sum across columns)/total sum. Three points are noteworthy: First, the number of ECOs in the utilities industry in each year roughly corresponds to the expected numbers. Thus, while overall the utilities industry is over-represented in the ECO sample, there are no individual years of overrepresentation, relative to the number of ECOs in each year. Second, there are more ECOs in the telecommunication industry (NACE code 65) in 1998, 1999 and 2000 than expected. This may be the result of the liberalization process in this industry, which was largely completed by 1998. 62 It is also likely to be a reflection of the new economy area, when for example France Telecom and Deutsche Telekom both carved out their Internet subsidiaries (Wanadoo and T-Online, respectively) because the parent firms believed that these assets were not correctly valued when they were part of a larger conglomerate. This explanation is supported by the fact that there are considerably more ECOs than expected in the technology industry (NACE code 95) in 2000. Third, as a mirror image of the previous finding, the number of ECOs in the construction & materials (NACE code 23) and the industrial goods & services industries (NACE code 27) is lower than expected in 2000. ECOs in the same two industries are over-represented in some of the years prior to 2000. Thus, deregulation and the new economy period lead to the observed deviation of the distribution of ECO sample firms from the distribution of all benchmark firms. However, as indicated, the level of deviation seems limited. There are two consequences from these descriptive statistics. First, the analysis of value creation in ECO must consider the industry of the involved firms to account for possible industryrelated effects, e.g., in the explanation of the cross-sectional distribution of performance measures following the ECO. Second, the motivation of the parent firm for the transaction should also be taken into account to reflect the various reasons (each with potentially different value implications) why firms engage in an ECO. 63 Appendix 9 shows the distribution of sample firms across the 13 countries in the study, as well as the number of ECOs per country as a percentage of all IPOs in that respective country. The majority of the sample firms (n=69) are German. There are two possible explanations. First, this could be evidence for a sample selection bias, caused by the inclusion of all ECO cases from existing academic studies largely using German samples. 64 Second, the high frequency of German ECOs in the sample may reflect underlying economic reasons. For example, it could be caused by the conglomerate nature of many German companies, leading to a higher potential for unbundling activities. Also, firms in other countries may prefer other forms of portfolio restructuring. In the UK, for example, spin-offs can be structured tax-free 65, whereas in Germany the criteria for a tax-free treatment of a spin-off have been much harder to 62 See Mayer-Schoenberger/Strasser (1999), p. 574. 63 These two considerations are reflected in the later analyses by matching benchmark firms (among other) on industry association, and by including a dummy variable encoding the motivation of the parent firm for the ECO as stated in the ECO announcement. 64 See section 2.1.5 for details on how the sample was identified. 65 See UBS Investment Research (2005), p. 9.

14

meet during most of the sample period. 66 This view is supported by the fact that while on a European level, 79% of all exchange-related disinvestments carried out between 1990 and 1998 have been spin-offs and only 16% have been ECOs 67, there have been only three spin-offs (vs. 69 ECOs) in Germany in the 1984 to 2004 period. Further, the comparison of ECOs as a percentage of all IPOs shows that while this percentage is highest for Germany (approx. 8.2%), the difference to the next two largest values does not seem dramatic (approx. 7.2% for Spain and approx. 6.2% for Switzerland). The mean value for all European ECOs identified in this sample is approx. 3.4%. Thus, even if a selection bias exists, its impact is likely to be limited. Table 1 shows some key descriptive statistics for the ECO sample. Four points are noteworthy: First, parent firms on average continue to hold a majority stake in the carved-out entity, with a mean (median) stake sold of 40.3% (34.3%), and a mean (median) stake held by the parent firm in the carved-out entity of 50.7% (55.0%). Parent firms apparently prefer to hold on to a controlling stake, suggesting that the benefits associated with an ECO may also be realised when the parent firm continues to be the majority owner. Second, sample firms display considerable differences in size, as evidenced by the 10%- and 90%-percentile values for parent and subsidiary firm assets. This finding emphasises the necessity to use size variables when analysing ECOs to control for these differences.

Mean

Median

10th percentile

90th percentile

Pre-event stake Post-event stake Stake sold

90.7% 50.7% 40.3%

100.0% 55.0% 34.3%

62.1% 10.0% 11.0%

100.0% 77.1% 81.9%

Gross proceeds (~ million) Relative gross proceeds

513 33.0%

149 8.9%

11 1.0%

1,702 63.6%

Subsidiary assets (E million)

2,330

156

5

4,663

34,991 44.8% 23.2%

2,982 8.5% 6.0%

56 0.3% -2.2%

68,726 59.5% 44.8%

Parent assets (E million) Relative size Underpricing Table 1: Descriptive statistics

Third, the relative importance of the ECO for parent firms differs, with average (median) relative gross proceeds 68 of 33.0% (8.9%) at the announcement date, and relative asset size 69 of 44.8% (8.5%) in the year of the announcement. Depending on the relative size, ECOs may thus represent a 'non-event' for some parent firms. Fourth,

See Glatzel (2003), p. 114-117, for a description of tax treatment for equity carve-outs and spin-offs in Germany. 67 See Glatzel (2003), p. 4. 68 Relative gross proceeds are defined as subsidiary to parent firm market capitalization on the first day of trading, multiplied by the percentage stake sold. 69 Relative asset size is defined as subsidiary to parent firm total assets.

15

there is evidence of underpricing 7~ with average (median) underpricing amounting to 23.2% (6.0%). The average value is similar to the level of underpricing found for the US by Welch/Ritter (2002), who document an average first day return of 18.8% for their sample of 6,249 IPOs in the 1980 to 2001 period. 7~ The average underpricing in the current sample is slightly higher, potentially because of the high number of ECOs in the Internet bubble period: Blgttchen/Gutschlag (1999) document that underpricing for all German IPOs in 1997 and 1998 on average is 45%, whereas it is only 17% for the sub-set of firms listed in the 'Amtlicher Handel' and the 'Geregelter Markt', the old economy segments of the Frankfurt Stock Exchange. The finding of a general underpricing suggests that informational asymmetries, which are assumed to be a potential explanation for the underpricing of IPOs 72, may also play a role in ECOs. Additional explanations for the underpricing phenomenon relating to principal-agent issues (e.g., managers interested in a low initial offer price to profit from relatively higher trading prices73) may also be relevant for ECOs. These problems may even be exacerbated in ECOs, because parent firm management constitutes an extra layer between the subsidiary firm and potential investors and thus an additional source of principal-agent issues. TM Underpricing may also be in the interest of parent firms intending to list multiple subsidiaries: High initial investor gains are likely to increase the probability of successful future ECOs.

2.2

Reasons for engaging in an equity carve-out

This section reviews some of the empirically determined motivations for an ECO which parent firms state when announcing their intended transaction (section 2.2.1). Since it may not be in the interest of the parent firm to announce all of the reasons for an ECO (e.g., if an ECO is carried out to sell an overvalued subsidiary firm), some theoretical considerations regarding the sources of value creation for the parent firm are also described (section 2.2.2). Together these two perspectives produce a holistic view of the reasons why firms engage in an ECO.

2.2.1 Motivations of parent firms The motivations for an ECO can be manifold. The final objective of these motivations (as of any company's decision) should be the increase of shareholder value. A company may conclude on the basis of a strategic review that a certain business segment does not any longer fit into its overall long-term business strategy and hence decide to dispose of Underpricing is defined as first day closing price divided by offer price, minus 1. Ibbotson/Sindelar/Ritter (1988) document a similar level of underpricing (16.4%) for an earlier US sample of IPOs in the time period from 1960 to 1987 (see Ibbotson/Sindelar/Ritter (1988), p. 37) 72 See Rock (1986), Welch (1992) and Drobetz/Kammermann/W~ilchli(2003). 73 See Loughran/Ritter (2004), p. 6. 74 See Wunderlich (2003), p. 43-44.

70 71

16

it (example: Sulzer/Elma Electronics75), combined with the desire to exit a loss-making business (example: Siemens/Infineon76). A company may also wish to develop either its own business or the business of its subsidiary firm but may be lacking the capital to do so, and decide to obtain the required financing from external capital markets (example: Rheinmetall/Aditron 77 and B.U.S./Befesa 78, respectively). ECO proceeds may be used to repay debt of the parent (example: Hoechst/SGL Carbon 79) or of the subsidiary firm (example: Babcock Borsig/Nordex8~ A company may intend to diversify its investor base nationally (example: Triumph Adler/Zapf CreationS~), or internationally if its subsidiary firm is operational in a different country (example: RWE/Consol Energy82). A company may doubt that its subsidiary is valued properly by capital markets as part of the parent group, and aim for a valuation more in line with the subsidiary's industry peers (example: Mobilcorn/Freenet83). The carved-out firm may also become more flexible in its dealings with other companies and customers (example: Merkantildata/Hands84). Employees may be more motivated if they are allocated publicly traded shares and/or options on such shares (example: Goldzack/Bankhaus Heinrich Gontard85). A company may have to carve out a subsidiary firm to comply with regulatory requirements (example: Oerlikon-Btihrle/Pfeiffer Vacuum 86) or because it wants to insulate itself from liability claims in another legislation (example: Leif Hoegh/Bona Shipholding87).

2.2.2

Sources of value creation

Why would an ECO create value? In a Miller-Modigliani world with perfect capital markets 8s, the value of a firm would only depend on the net present value of the firm's projects, and not on how the firm is structured financially. This section provides a brief overview of the theoretical reasons why an ECO could beexpected to create value, if the assumptions of a perfect world do not hold.

75

See Neue Ziircher Zeitung, November 1 1996, p. 31. See B6rsen-Zeitung, November 5 1998, p. 1. 77 See vwd, Dezember 16 1999. 78 See B6rsen-Zeitung, February 18 1998, p. 6. 79 See Focus Magazin, April 3 1995, p. 296. 80 See Frankfurter Allgemeine Zeitung, March 20 2001, p. 28. 81 See Frankfurter Allgemeine Zeitung, February 4 1999, p. 26. 82 See B6rsen-Zeitung, November 20 1998, p. 6. 83 See Die Welt, August 10 1999. 84 See AFX, August 24 2004. 85 See BSrsen-Zeitung, October 10 1998, p. 6. 86 See Neue Ztircher Zeitung, March 15 1996. 87 See Journal of Commerce, October 21 1992, p. 1B. 88 I.e., a world without taxes, costs of financial distress and transaction costs, where all market participants act rationally and in possession of complete and cost-free information. See Miller/Modigliani (1958). 76

17

First, an ECO may allow a firm to raise relatively cheaper capital. If the cost of equity can be proxied by the inverse of the P/E ratio (i.e., the E/P ratio) 89, and a subsidiary firm is more highly valued by capital markets than the parent firm as reflected by the respective P/E ratios, then issuing equity through the subsidiary firm rather than through the parent firm may increase firm value by lowering the WACC. Consistently, Schipper/Smith (1986) find that the carved-out entities have significantly higher P/E ratios than their former parent firms. 9~ Higher valuation multiples may occur in certain industries over certain periods of time. 91 Similarly, if the subsidiary firm is operational in another country in which capital markets value its profits more highly than parent firm profits, then issuing equity through the subsidiary firm in this country can increase firm value. As an example, a number of US companies have floated their subsidiaries in Japan in the 1980s and 1990s, potentially to profit from higher local valuation levels. 92 Second, an ECO can serve to overcome informational asymmetries between the parent firm and investors. By floating the subsidiary firm, supply of information increases because of the publication of the IPO prospectus 93 and the legally required increase in transparency of the subsidiary as a publicly traded company (e.g., the publication of annual, semi-annual and in some cases even quarterly accounts). 94 Demand for information increases as well, because existing and future investors require data to form their valuation views. Third, an ECO could be a positive signal by management to capital markets: As proposed by Nanda (1991), issuing equity in a subsidiary firm is a substitute for issuing equity in the parent firm. If a firm only issues equity when it is overvalued, then issuing equity in a subsidiary firm signals that the subsidiary firm's industry is overvalued, and by deduction the parent firm is (relatively) undervalued. Therefore an ECO may increase the parent firm's share price if investors take it as a signal of its undervaluation. Similarly, Myers/Majluf (1984) argue that management may forego certain NPVpositive investments if the capital required has to be raised externally: Issuing equity would send a negative signal to capital markets and indicate an overvaluation of the parent firm's equity. To prevent this underinvestment problem, Schipper/Smith (1986) suggest separating parent and subsidiary firm to allow separate financing for the profitable projects of the subsidiary firm. 95

89

See Hax/Hartmann-Wendels/von Hinten (1988), p. 695, who suggest that market values can be used to derive the rate of return on equity capital required by the markets. 9o See Schipper/Smith(1986), p. 172. 91 "Particularly successful carve-outs often occur in industries that are currently 'hot' in the equity markets" (Mathesius (2003), p. 121; there quoted from Lindenberg, E.B., Blaton, P.B., Abuaf, N., Thacher, K.L. (1991): "The Case for Carve-Outs - The Executive's Guide to Creating Value Through an Equity Carve-Out"). 92 See Fikre (1991), p. 61-64. 93 In Germany, companies going public are legally responsible for this document ("Prospekthaftung"), increasing the credibility of the information provided. 94 See Pellens (1993), p. 857. 95 See Schipper/Smith (1986), p. 169-170.

18

Fourth, if capital markets are incomplete and investors have heterogeneous expectations, then separating the subsidiary from the parent firm potentially increases the number of securities that cannot be recreated as a linear combination of existing securities. Investors are thus able to increase their diversification and to improve the risk-return profile of their holdings. 96 Practically speaking, the floatation of the subsidiary firm creates a pure-play company allowing investors to allocate funds to a single-segment firm, which may better suit their investment strategy. Fifth, an ECO replaces internal control mechanisms subject to principal-agent issues with assumedly more efficient external control mechanisms. External control mechanisms include capital markets, the legal and regulatory framework, and product and factor markets. 97 Sixth, an ECO allows employees to participate in the success of the subsidiary firm by either buying or being allocated shares and option on shares in the company. Allen (1998) documents how following a series of ECOs US company Thermo Electron has managed to retain all of its executives over a multi-year period. 98 Positive market reactions to the announcement of an ECO could also be the result not of value creation, but of a value transfer between various stakeholders. 99 On a theoretical basis, value may shift from debt to equity holders because equity value, calculated as a call option on the firm's underlying assets 1~176 may increase when its variance increases as a result of the new stand-alone nature of the carved-out entity. If the total value of the firm remains constant (based on the neoclassical assumption that firm value is determined by the net present value of the firm's projects), an increase in equity value comes at the expense of the value of debt. Empirically, Maxwell/Rao (2002) demonstrate how in their sample of 80 US spin-offs from the 1976 to 1997 period positive announcement period returns for parent firm shares of 3.6% are accompanied by negative returns of-0.9% for the corresponding debt securities. Overall wealth thus increases, but part of the gain for shareholders is explained by a transfer, rather than a creation, of value.

2.3

Efficient market hypothesis

An understanding of the efficient market hypothesis (EMH) is fundamental to assessing the question of whether ECOs produce abnormal returns both in the short and in the 96 See Hakansson (1982), p. 977. 97 See Jensen (1993), p. 850. For example, a conglomerate company may manufacture products which are not competitive but whose existence is cross=subsidized by other profitable projects. This may not be possible any longer when the subsidiary becomes a stand-alone company. 98 Allen (1998), p. 122. 99 Btihner (2004) demonstrates with a numerical example how value can be shifted from equity to debt holders or vice-versa, depending on the return standard deviation of the two separated entities in relation to the return standard deviation of the combined group. 100 See the seminal paper by Black/Scholes (1973) for further details.

19

long term. This section describes the origins (section 2.3.1) and the basic concepts (section 2.3.2) of the EMH. The notion of a financial anomaly is clarified (section 2.3.3), and some of these alleged violations of the EMH are reviewed (section 2.3.4), as well as the academic community's way of dealing with such violations (section 2.3.5). Given the volume of the relevant literature, this review cannot be exhaustive: Its purpose is to sufficiently sensitise the reader to the disagreement regarding the validity of the EMH. A brief analysis of psychological biases potentially at work in the dispute about the validity of the EMH (section 2.3.6) concludes the section.

2.3.1

History of the efficient market hypothesis

The roots of the EMH date back to the 19 th century. The basic ideas are usually credited to Louis Bachelier's Ph.D. thesis from 1900, "Thdorie de la Sp6culation ''~~

In his

work, Bachelier applies the tools of probability calculation to the development of prices of French government bonds, options and forward contracts. He seems to have coined the term "fair game ''1~ to describe a situation in which the next price movement cannot be systematically anticipated. His description of the movement of prices anticipates the basic notions of the random walk hypothesis 1~ i.e., the idea that asset prices fluctuate unpredictably. The origins of the E M H have been traced back by historians even further, to as early as the 1860s and 1870s. Jovanovic/Le Gall (2001) argue that the foundations for the EMH were described by Jules Regnault in his 1863 book "Calcul des chances et philosophie de la Bourse". They also mention that another French economist, Henri Lefevre, published a book in 1870 entitled "Trait6 des valeurs mobili6res et des operations de Bourse: Placement et speculation". Considering these earlier works, Preda (2004) concludes that Bachelier did not (as is sometimes implied when placing him at the beginning of the history of EMH) produce his work in genial isolation, but rather based it on a series of pre-existing related considerations. Bachelier's work has gone largely unnoticed for more than half of the 20 th century. There are some empirical works in the 1930s and 1940s on the random nature of stock

101

102 103

See Dimson/Mussavian (1998) and Kasper/Sullivan/Weithers (1991). A colourful description of Louis Bachelier's life and contribution to financial theory can be found in Mandelbrot/Hudson (2005), p. 77-91. See Mandelbrot/Hudson (2005), p. 87. Bachelier derived the equations for the price behaviour on the basis of a specific stochastical process. Five years later Einstein derived very similar equations describing the movements of gas molecules. The associated stochastical process became known as the Einstein-Wiener process of Brownian motion, with reference to Scottish botanist Robert Brown who first observed the seemingly random movement of gas molecules (see Cunningham (2000), p. 4, and Dimson/Mussavian (1998), p. 92). Probably owing to the human mind's preference for graphical images as opposed to abstract statistical concepts, the term "random walk" became popular as a potent visual representation for the statistical process. According to Mandelbrot/Hudson (2005), the concept of a random walk originated in a discussion in the natural science journal Nature in 1905, where Karl Pearson concluded that the most likely place to find a drunken person left to navigate himself is his original starting point.

20

and other asset class prices 1~ but it is only in 1964 that Bachelier's work is translated into English by Cootner (1964), and thus made accessible to mainstream financial academia. Shortly afterwards Samuelson (1965) formulates a microeconomic foundation of the random nature of price movements. He argues that any systematic expectation of a price change must already be incorporated in the current price, and hence, the next price movement cannot be determined in advance. Fama (1970) summarises the literature on price development and formalises the theory. From then on the idea of the random nature of price changes is referred to as the efficient market hypothesis.

2.3.2 Basic concepts of the efficient market hypothesis Fama (1970) defines an efficient market as a "market in which prices always 'fully reflect' available information ''1~ Based on a distinction made by Harry Roberts in an unpublished manuscript 1~ Fama differentiates between three forms of market efficiency: Weak form efficiency, which posits that current prices reflect all past information; semi-strong efficiency, which posits that current prices reflect all past and publicly available information; and strong efficiency, which posits that current prices reflect all past, publicly and privately available information. This taxonomy has remained popular until today. 1~ According to Fama (1970), the expected price of security i, given a certain information set, is

E(~i,t+l[*t )-- [l + E(~,t+ll*, )]* Pi,,, where E is the expectations operator, Pi,,+I is the price of security i at time t+ 1 (with the tilde indicating that this is a random variable at time t), @, is the information set which is fully reflected in the price at time t, ~,t+l is the percentage return for the security (again a random variable at t), and Pi,, is the price of security i at time t. The excess return zi,,+1, defined as the return above the equilibrium expected return, is

Zi,t+l

= F/,t+l-E("~ii,t+ll(~kt),

104 See Dimson/Mussavian (1998), p. 92. 105 Fama (1970), p. 383. 106 See Dimson/Mussavian (1998), p. 103. 107 Among others, many mainstream corporate finance textbooks use this classification, see, e.g., Ross/Westerfield/Jaffe (1999), p. 320-325 and Brigham/Gapenski/Ehrhardt (1999), p. 353-354.

21

where r/.t+1is the actual return, and E(~/,t+1* t ) i s

the equilibrium expected return,

calculated as EI,/)i,,+l~ , ) _ 1. The implication of the assumption that prices fully reflect Pi,t

all available information is that

0

An expected value for the excess return of 0 implies that it is impossible to continuously earn above-equilibrium profits, based on the knowledge of ~ , . Analogously, bets on changes in the price of security t are a "fair game ''~~ There are two common extensions to this basic model. First, one can assume that

E(pi,,+II*, )>- Pi.,, or equivalently E(~ii,t+lI~ t )~ O, which means that the price sequence for security i follows a submartingale 1~ As pointed out by Fama (1970), the implication is that any portfolio consisting of a combination of security i and cash cannot earn a higher return than simply buying security i and holding it for the period. The return will always be non-negative, even if the expected return on the security is negative 1~~ because the expected return on cash is zero. Second, the random walk model places an additional restriction on the development of the price sequence by assuming that future price changes are identically and independently distributed, and that this distribution is stationary through time. 11~ It thus makes a statement about the return-generating process, which can be made subject to empirical analysis: Any sign of serial correlation in prices would pose a violation of the random walk model. Also, the fact that the return-generating process is assumed to be known implies that there is value in knowing past returns: Because the distribution is assumed to be stationary through time, it is the best estimate for the future distribution of returns (whereas the actual past order of returns has no predictive power for the future order of returns).

108 Fama (1970), p. 385. 109 For a precise definition of a martingale, see Shreve/Chalasani/Jha (1997), p. 58. Intuitively, if a random variable follows a martingale process, the variable's current value is the best estimator for the variable's value in the next time period. With a submartingale process, the value of the variable tends to go up. 110 Which, e.g., can be the case in the CAPM model when the security is negatively correlated to the market, see Fama (1970), p. 386. 111 As pointed out by Fama (1970), since expected price changes can be non-zero, the random walk may be subject to a drift (see Fama (1970), p. 386).

22

A somewhat more relaxed definition of EMH is provided by Jensen (1978), who states that a "market is efficient with respect to information set | if it is impossible to make economic profits by trading on the basis of information set | ''~ 12 He goes on to define economic profits as "risk adjusted return net of all costs ''~3. Therefore, there is a band around the efficient price within which the actual price can oscillate, and the width of the band is determined by the sum of all transaction costs. This concept is similar to Basu's (2004) conclusion that US and UK stock markets are "not rational, but [...] minimally rational ''114 (italics in original): He finds that a trading strategy based on audit opinions and net operating assets does not produce abnormal profits when transactions costs are taken into account. The strongest empirical support for market efficiency comes from studies indicating that professional fund managers are generally not able to consistently produce abnormal positive returns. Malkiel (2005) finds that over return periods of ten years, 80% of active fund managers underperform the S&P 500. However, hedge fund managers on average do seem able to outperform the market on a pre-fee return basis, indicating the existence of price inefficiencies (Ackermann/McEnally/Ravenscraft (1999)). 115

2.3.3

Definition of an anomaly

In the context of capital markets research, the term 'anomaly' refers to violations of the EMH. Three issues need to be addressed from a science of philosophy point of view. First, how is the term anomaly used in scientific work? Second, is there potential for a linguistic bias arising out of that use? Third, does the EMH meet the required standards to qualify as a proper scientific theory? First, Kuhn (1970) uses the terms 'paradigm' and 'anomaly' in his description of how science evolves: "Discovery commences with the awareness of an anomaly, i.e., with the recognition that nature has somehow violated the paradigm-induced expectations that govern normal science. It then continues with a more or less extended exploration of the area of anomaly ''~16. Anomalies thus stand in contrast to paradigms, which a r e defined as "achievements that some particular scientific community acknowledges for a time as supplying the foundation for its further practice ''~17. In the course of scientific work, an anomaly can (but does not necessarily have to) be the cause of the rejection of a previously well-established paradigm, and be the first step in the creation of a new paradigm. A financial anomaly is defined by Brav/Heaton (2002) as "a documented pattern of price behaviour that is inconsistent with the predictions of traditional efficient 112 Jensen (1978), p. 96. 113 Jensen (1978), p. 96. 114 Basu (2004), p. 346. 115 However, hedge fund investors do not seem to profit from this as most of the excess return is used for paying the incentive and administrative fees (see Ackermann/McEnally/Ravenscraft (1999), p. 871.) 116 Kuhn (1970), p. 52. 117 Kuhn (1970), p. 10.

23

markets ''1~8 Therefore, in Kuhn's terms, EMH is the existing paradigm, whose prevalence is challenged by the anomalies, i.e., price patterns not predicted by the EMH. Second, Frankfurter/McGoun (2001) analyse the term 'anomaly' with reference to its use by Kuhn (1970), and find that he employs the term in line with the generally understood meaning as "an irregularity, a deviation from the common or natural order, or an exceptional condition ''119. They also point out that in its actual use in the finance literature it "becomes a pejorative term applied to something not just inconsistent with a paradigm, but with a deeper underlying ideology ''12~ The term also stands in contrast to the more positively associated adjective 'efficient '~21 Given these differing connotations, there is potential for a linguistic bias in the decision of which framework one favours, assuming one is faced with empirical data supporting both the EMH and alternative asset pricing theories. Third, as pointed out by Fama (1970), tests of market efficiency are always also tests of the underlying model assumed to produce normal returns. 122 Campbell/Lo/MacKinlay (1997) point out that this joint hypothesis problem makes it impossible for the EMH to be rejected. 123 According to Chalmers (2001) and in the spirit of Karl Popper, it is an essential criterion for a hypothesis to be falsifiable in order for it to qualify as a scientific theory. 124 In this sense, EMH is not a valid theory.

2.3.4 Typesof anomalies There is a bountiful literature on financial anomalies. While a detailed review is beyond the scope of this study, a survey suffices to demonstrate the seriousness of the challenge posed by the sum of these findings to the prevalent EMH paradigm. 125 One of the first documented anomalies is presented by Ball/Brown (1968), and later confirmed among others by Bernard/Thomas (1990): Both document a post-earnings announcement drift in the same direction as the earnings surprise, indicating that expectations for future earnings are not completely revised on the basis of present earnings surprises. Banz (1981), using a sample from the 1936 to 1975 period, finds that NYSE firms with a low market value of equity earn returns in excess of expected returns as predicted by the CAPM ('small firm/size effect'). 126 Basu (1977), using a US sample from the 1956 to 1969 period, finds that firms with low price-to-earnings (PE) 118 119 120

Brav/Heaton (2002), p. 575. Frankfurter/McGoun (2001), p. 410.

Frankfurter/McGoun (2001), p. 421. 121 Put saliently, given a choice, would you rather be efficient or anomalous? 122 See Fama (1970), p. 384. 123 See Campbell/Lo/MacKinlay (1997), p. 24. 124 See Chalmers (2001), p. 53. 125 Comprehensive reviews of financial anomalies include Jensen (1978), Fama (1998) and Schwert (2003). 126 See Schwert (1983) for a detailed review of the early literature on the small-firm effect.

24

ratios earn positive abnormal returns. Similarly, Fama/French (1988) find that firms with a high dividend yield earn positive abnormal returns, and Fama/French (1992), using a US sample from the 1962 to 1989 period, find that shares with high book-tomarket (BTM) ratios have significantly higher returns than shares with low BTM ratios. These findings on PE, dividend yield and BTM are often summarised as the 'value effect'. Various levels of serial correlation in prices have been noted. DeBondt/Thaler (1985) find that from 1926-1982 portfolios of NYSE companies which have underperformed the market in the previous three to five years perform better in the following three years than portfolios of companies which have previously outperformed the market. DeBondt/Thaler (1985) interpret this as evidence that investors overreact tonews, and market prices subsequently drift back to normal levels. On the other hand, Jegadeesh/Titman (1993) find that portfolios of NYSE companies which have underperformed the market in the previous twelve months continue to underperform in the following three to twelve months in the 1965 to 1989 period. Rouwenhorst (1998) confirms this result for a number of European markets. Thus, share prices are apparently characterised by a long-term trend reversal and a short-term momentum effect. A series of studies have pointed out anomalies based on accounting information. Sloan (1996) finds evidence that investors seem to overweight the information content of accruals and underweight the information content of cash flows when forming their expectations of future earnings. As accruals tend to be less persistent than cash flows, high-accrual firms will earn lower returns than low-accrual firms ('accrual anomaly'). Taffler/Lu/Kausar (2004) document an abnormally positive return to a strategy of shortselling shares in UK firms who have received a going-concern modified audit opinion in 1995-2000: Apparently, investors do not adequately take into account the warning signals issued by accountants. Hirshleifer/Hou/Teoh/Zhang (2004) find that shortselling shares in firms with high net operating assets (i.e., the cumulative difference between accounting income and free cash flow) 127, scaled by total assets, and buying shares in firms with low scaled net operating assets, produces positive excess returns over the next three years. They interpret this result as evidence for investors overweighting information regarding accounting profitability, and under-weighting information regarding cash profitability. Ackert/Athanasskos (1997) find that for their sample of 167 firms companies with a low dispersion of analysts' forecasts outperform companies with a high dispersion ('analyst forecast dispersion anomaly'). Ang/Ciccone (2001) confirm these findings for a larger US sample, and Dische (2002) finds similar evidence in a German sample. Many anomalies have been discovered with relation to certain time periods. French (1980) finds that S&P 500 composite returns are significantly negative over weekends in the 1953 to 1977 period ('weekend effect'). Other papers confirm positive average returns on Fridays and negative average returns on Mondays for the US, and negative 127

See Hirshleifer/Hou/Teoh/Zhang (2004), p. 299.

25

average returns on Tuesdays for some European and Asian countries 128, possibly indicating spill-over effects on these markets from the US. Related to this, Jaffe/Westerfield/Ma (1989) find that negative returns on Mondays tend to follow negative market returns in the prior week, while Monday returns are not different from zero when market returns were positive in the prior week ('twist-of-the-Monday effect'). Aggarwal/Leal (1996) confirm this effect for the Brazilian stock market. Keim (1983) finds that a large proportion of the abnormal gains of the small-firm effect occur in January ('January' or 'turn-of-the-year effect'). Potential explanations for this effect include realization of losses for tax purposes and windows dressing by fund managers. 129 The January effect can also be observed in debt markets (ChristieDavid/Chaudry (2000)). Ariel (1987) documents positive average returns for the first half of a calendar month ('turn-of-the-month effect'). Similarly, Kohers/Patel (1999) find that returns of the S&P (1960 to 1995) and NASDAQ (1972 to 1995) index during the first third of a month tend to exceed returns in the second third of a month, which in turn exceed returns in the final third of the month ('time-of-the-month effect'). Lakonishok/Smidt (1988) observe significantly higher returns on trading days preceding holidays for the Dow Jones Industrial Average (DJIA) index in the 1897 to 1986 period ('holiday effect'). The holiday effect is confirmed for Italy (Barone (1989)), Canada, Japan, Hong Kong and Australia (Cadsby/Ratner (1992)), and the UK (Arsad/Coutts (1997)). Bouman/Jacobsen (2002) find that a trading strategy based on the old saying "Sell in May and go away ''13~ (i.e., buying the market portfolio in October and selling it at the end of April) yields abnormal positive returns ("Halloween indicator") in 36 out of 37 countries in their sample. The Halloween indicator links time-related anomalies to another broad strand of research analysing the impact of weather and temperature on stock returns. In general, these papers argue that the weather is likely to influence investors' moods TM, and moods in turn impact the decisions taken by investors in the market. Saunders (1993) finds that returns on the DJIA index (1927 to 1989) and on the AMEX index (1962 to 1989) are significantly lower on days when New York is covered by clouds. Consistently, Hirshleifer/Shumway (2003) find that stock returns in 26 different countries in the 1982 to 1997 period are significantly higher on sunny days than on overcast days. Both studies support the intuitive idea that investors are more optimistic during good-weather periods, and hence tend to be buyers (rather than sellers) on those days. Kamstra/Kramer/Levi (2000) observe that index returns in the US, UK, Canada and Germany are below average on the Mondays following weekends with daylight savings time changes, following both weekends with a "gain" of one hour and 128

See Bildik (2004), p. 3, footnote two and three, for a detailed survey of this literature. See his footnote one for a list of other markets in which the weekend effect has been observed. See Christie-David/Chaudry(2000), p. 79-81 for more details on these potential explanations. 130 As Bouman/Jacobsen (2002) point out (p. 1618), the origins of the phrase are unclear, but it is widely used in the popular press. 131 See Cao/Wei (2005) for a detailed survey of literature linking a person's mood to his decision making process. 129

26

weekends with a "loss" of one hour. Yuan/Zheng/Zhu (2006) link stock returns to lunar phases: Stock returns tend to be higher around a new moon than around full moon. The authors link this to generally observed effects of the lunar cycle on human behaviour. 132 Kamstra/Kramer/Levi (2003) find that stock market returns are influenced by a seasonal affective disorder (SAD), a clinically recognised form of mood shift in winter months, in eight out of nine international markets. In particular, returns in autumn are more negative, and returns following the shortest day of the year are more positive than average returns, coinciding with the SAD cycle. Cao/Wei (2005) find a negative relation between stock returns and temperature in eight international stock market indices ("temperature anomaly"), which is robust towards including variables for other anomalies such as the January effect, cloud cover, and SAD.

2.3.5

Academic community's reaction to potential anomalies

How does the financial academic community deal with findings of anomalies? Three general approaches can be observed. First, the results of the respective empirical study are questioned on methodological grounds, and stability of results across time and geography is denied (section 2.3.5.1). Second, the anomaly may be subsumed under another (more accepted) anomaly (section 2.3.5.2). Third, if the finding passes these two screens, the anomaly is explained either as evidence of investor irrationality, or as reflecting underlying sources of risk requiring compensation in the form of returns which seem abnormal when the risk sources are not known (section 2.3.5.3).

2.3.5.1

Questioning the anomaly

Several anomalies have been found to be either unstable across time or geography, or when other variables are controlled for. Berk (1997) finds that the size effect is only found when using the market value of equity, rather than total assets, as a size proxy. There is little evidence for a long-term trend reversal once size and BTM are controlled for (Fama/French (1996)). Schwert (2003) finds no evidence of a small firm/size effect in the US after 1982 (around the time when the first research on this effect was published). Mehdian/Perry (2002) find that the January effect seems to have disappeared in the US after the 1987 market crash. Connolly (1989) finds no evidence for the weekend effect in the US market after 1975. Madureira/Leal (2001) find no evidence for the twist-of-the-Monday effect in Brazilian stock indices and individual stocks in the 1994 to 1998 period. The turn-of-the-month effect is not found in Japan, Canada and the UK (Jaffe/Westerfield (1989)). Coutts/Sheikh (2002) find no evidence for either the weekend, the holiday or the turn-of-the-month effect in the gold index on the Johannesburg stock exchange. Chong/Hudson/Keasey/Littler (2005) find that the size of the holiday effect has declined in the US, with its sign even reversed in the 1991

132

See Yuan/Zheng/Zhu (2006), p. 4.

27

to 1997 period, subsequently re-establishing itself in the 1997 to 2003 period. 133 They interpret this as supporting the idea of traders attempting to take advantage of a wellpublicised anomaly, resulting in its temporary disappearance. Studies uncovering weather-related anomalies are criticised by Jacobsen/Marquering (2004) for potentially producing spurious results due to the weakly established link between weather and temperature on the one hand, and investment decisions on the other hand. Jacobsen/Marquering (2004) do not question the existence of seasonal anomalies, but they do cast doubt on the various explanations offered to explain this effect. Similarly, Goetzmann/Zhu (2005) find no link between cloud cover and individual trading account movements in five major US cities in the 1991 to 1996 period. Loughran/Schultz (2004) observe that trading in NASDAQ stocks is localised, i.e., influenced by conditions existing at the location of a firm's headquarters, such as blizzards and time zones, but cloud cover at the firm's headquarters does not seem to influence the firm's stock returns. Instability across time, as pointed out by Jones/Pomorski (2002), is likely to be caused 134 by the market becoming aware of the profit opportunity, and arbitraging it away. However, limitations on short sales and risk considerations may prevent or limit arbitrage opportunities: It is not sufficient for an arbitrageur to know that a security is mispriced, but he also has to feel certain that the market will realise this mispricing in an appropriate time frame. 135 Cao/Kolasinski (2005) find that short-sellers exploit the post-earnings announcement drift and the accrual anomaly; however, prices continue to drift following earning announcements, indicating that short-sellers do not completely arbitrage away the anomaly. Hudson/Keasey/Littler (2002) point out that investors and academics have different objectives, and therefore anomalies detected by the latter may not be of benefit to the former. In contrast, Lucey/Pardo (2005) demonstrate that the holiday effect c a n be exploited by investors in real life. Dimson/Marsh (1999) document how an anomaly, once published, can not only disappear but actually reverse in nature, potentially resulting from arbitrageurs attempting to take advantage of the anomaly, and collectively pushing prices in the opposite direction. Not all anomalies are instable. Jegadeesh/Titman (2001) find that the momentum effect has remained in existence in the 1990-1998 period, following their 1993 paper. Bouman/Jacobsen (2002) document the Sell-in-May-and-go-away effect despite the fact

133

In contrast, Brockman/Michayluk(1998) find that the holiday effect persists in the 1987-1993 period, which they interpret as surprising because the first finding of the holiday effect is from Lakonishok/Smidt (1988): Traders do not seem to have been able to profit from this effect in the 1987-1993 period even though it was first documented in 1987/1988. 134 See Jones/Pomorski (2002), p. 1. 135 For example, marking-to-market requirements in the futures market may lead to margin calls, requiring the arbitrageur to put up further capital if the market increases the level of mispricing prior to the expected price correction. This may force arbitrageurs out of the market even though their assessment of the mispricing eventually proves accurate. See Schwager (1992), p. 92-94, for a description of the experience of Randy McKay, a successful currency futures trader, regarding this issue.

28

that the saying has been known since at least 1964.136 Schwert (2003) finds that the January effect still exists in the US in the 1980 to 2001 period. Ali/Hwang/Trombley (2003) attribute the continued persistence of the BTM effect to the fact that high BTM stocks also tend to be high volatility stocks, making arbitrage (short-selling) more expensive and therefore limiting arbitrage opportunities. Bushee/Raedy (2003) explicitly model various trading restrictions and find that while size and trend reversal anomalies cannot be profitably exploited, trading strategies based on other anomalies such as cash flow to price, return momentum, the post-earnings announcement drift and accruals can produce abnormal profits even after consideration of trading restrictions.

2.3.5.2 Subsuming the anomaly If the finding of the anomaly is accepted, the finding can be a reflection of other wellknown anomalies: The small-firm effect seems to be mainly caused by the January effect (Keim (1983), Reinganum (1983)). Cadsby/Torbey (2003) find that the time-ofthe-month anomaly is largely based on the turn-of-the-month anomaly. Desai/Rajgopal/Venkatachalam (2004) find some evidence that the accrual anomaly is an expression of the value effect. Hence, it is conceivable that the plethora of research results uncovering potential anomalies is reducible to a smaller number underlying asset pricing rules. Research efforts in this direction appear valuable, potentially paving the way to a more fundamental understanding of (financial markets) reality. In their own modest way, such efforts parallel physics' search for a unified field theory: Just as it is the aim of physics to unify the known four forces 137 and the currently existing elementary particles zoo (as physicists refer to the growing number of basis components of matter) into a single theoretical framework, the current 'anomalies zoo' in finance is potentially reducible to a more fundamental framework.

2.3.5.3 Explaining the anomaly If the finding of the anomaly is accepted, and the anomaly cannot be subsumed under another existing anomaly, there are at least two possible explanations. Investors could be acting irrationally as suggested by Lakonishok/Shleifer/Vishny (1994), who find that the former wrongly extrapolate past growth rates into the future. Similarly, Barucci/Monte/Reno (2003) show that short-term momentum and long-term trend reversal can be explained if investors are assumed to act under bounded rationality with adaptive learning rules. Such assumptions form the basis of the burgeoning field of behavioural finance. Alternatively, the existence of apparently abnormal profits could be linked to not explicitly recognised sources of risk 138, for which investors demand an 136 See Bouman/Jacobsen (2002), p. 1619. 137 Strong force, weak force, electromagnetic force, and gravitational force, see Greene (2004), p327-329. 138 An alternative to explaining varying returns as a result of unrecognised risk sources is to assume that the market risk premium changes over time. Efforts to adapt the CAPM to incorporate time-varying

29

adequate return. The seemingly abnormal return would thus become a normal return, relative to an asset-pricing model which takes the previously unrecognised sources of risk into consideration. This risk-based explanation could thus salvage the rational asset-pricing paradigm and with it the EMH, and is favoured among others by Fama (1998). Fama/French (1993) demonstrate that the EMH is flexible enough to deal with some of the detected anomalies: They find that size and BTM seem to capture the explanatory power that PE and leverage ratios were found to have for the cross-section of stock returns. 139 They attempt to link both size and BTM to economic fundamentals: Small firms seem to have a lower ROA than large firms, and similarly high BTM firms seem to have a persistently low ROA. 14~ Assuming that a lower ROA makes a firm a more risky investment, a higher return for small and high BTM firms may merely represent a compensation for this higher risk. There is no fundamental reason why a similar process of adding further factors resulting from the search for anomalies could not produce a risk-based model of expected returns, thereby allowing to maintain the EMH paradigm. 141 The challenge for such extensions is to prevent the impression of merely being "a convenient offshoot of the anomalous findings that motivated the extension ''142. Fama/French (1993) try to prevent such criticism by linking size and BTM factors to economic fundamentals and explicit sources of risk, as described above. A related approach is to develop asset-pricing models based on macroeconomic factors, which also link the explanatory variables to economic fundamentals. An example of this are Hahn/O'Neill/Reyes (2004), who develop a six-factor model and find that the size of the value effect varies through time and is dependent on the specific value proxy used. More generally, research detecting anomalies is sometimes criticised for highlighting the anomalous at the cost of the ordinary. Lo/MacKinlay (1990) warn that the more often a single data set is used, the more likely it is that an anomalous pattern will be found. 143 This is the classical data-mining argument. Given the frequent use of established databases such as Compustat and Datastream, this admonition seems highly relevant. Sullivan/Timmermann/White (2001) find that adjusting p-values of known calendar-related anomalies for the existence of all potential calendar-based anomalies renders the former insignificant. Merton (1987) points out that there is a "natural individual predilection to focus, often disproportionately so, on the unusual ''144 Similarly, Schwert (2003) speculates that the focus on anomalies "could be a by-product

market returns have yielded various versions of 'conditional' CAPM models. However, Lewellen/Nagel (2004) find that while company betas do vary across time, the changes are not enough to explain the unconditional pricing errors (Lewellen/Nagel (2004), p. 3-5). 139 See Fama/French (1993), p. 4. 140 See Fama/French (1993), p. 7-8. 141 Carhart (1997) extends the three-factor model by a fourth factor not captured by size and BTM: momentum. The model of asset prices used in a particular study may need to be adapted for the specific purpose: For one of the research questions in this study (long-term price performance), Ang/Zhang (2004) find that using Carhart's (1997) four-factor model may lead to overfitting. 142 Schwert (2003), p. 964. 143 See Lo/MacKinlay (1990), p. 432. 144 Merton (1987), p. 104.

30

of the publication process, if there is a bias toward publication of findings that challenge existing theories ''~45. Thus, the sheer volume of research published on anomalies may cause the observer to formulate a view tilted towards a rejection of the EMH, even should in reality most financial markets be efficient (in the somewhat more relaxed Jensen-sense, i.e., not offering any consistent abnormal profit opportunities) most of the time.

2.3.6

Potential biases

Bowman/Buchanan (1995) point out reasons why market practitioners may continue to be non-believers in the EMH: Markets do not seem to sufficiently punish investment strategies which, while earning positive returns, are not risk-return efficient. The existence of investment strategies and managers outperforming the market is seen as evidence for market inefficiency, in disrespect of the Infinite Monkey Theorem 146. Similarly, the success of some research analysts in predicting future share prices is highlighted, while not taking into consideration their unsuccessful colleagues' performances. Finally, Bowman/Buchanan (1995) postulate the existence of "vested interests ''147 within the investment management industry, whose justification for existence hinges largely on the non-efficiency of markets. Bowman/Buchanan (1995) also raise the counterpart question: Why might academics continue to believe in the EMH in the face of mounting counterevidence? They leave the answer "for the reader to pursue ''148. The following paragraphs address this issue. It is a well-known result in empirical psychology that people cling to their opinions. This phenomenon, known as belief perseverance, persists in the face of evidence contrary to the initial opinion. Ross/Lepper/Hubbard (1975) conduct an experiment in which participants are randomly told they have performed either very well or very badly in discriminating true from fictitious suicide notes. Participants continue to believe in this randomly assigned ability even following an explicit debriefing about the nature of the experiment. 149 The phenomenon carries over to social theories: Anderson/Lepper/Ross (1980) present participants of their experiment with two case studies on the relationship between a fire-fighter's risk preference and his job performance. Even though the participants are subsequently told that the case studies 145

Schwert (2003), p. 941. The Infinite Monkey Theorem, initially proposed by French mathematician Emile Borel in his 1913 book "M6canique Statistique et Irr6versibilit6", pp.189-196, states that "a monkey hitting keys at random on a typewriter keyboard will almost surely eventually type every book in France's Biblioth6que nationale de France" (http://en.wikipedia.org/wiki/Infinite_monkey _theorem). Similarly, a large enough number of fund managers are very likely to produce a certain number of fund managers outperforming the market, even on a repeated basis. 147 Bowman/Buchanan (1995), p. 160. 148 Bowman/Buchanan (1995), p. 157, footnote three. 149 More precisely, 'outcome' debriefing (i.e., information about the random assignment of test scores) does not affect the participant's self-perception about their own abilities, while 'process' debriefing (i.e., information about the perseverance effect) helps to reduce the false self-perception.

146

31

are fictitious, participants who read that high risk preference predicts future success continue to believe in this relationship, while participants who read that high risk preference leads to future failure continue to believe in that relationship. This violates normative standards of decision-making, according to which both groups should hold the same beliefs once the counter-evidence is presented. In addition to this cognitive belief perseverance, Sherman/Kim (2002) present evidence for an affective perseverance, i.e., emotions felt towards an object or situation tend to be felt by participants in their study even when researchers attempt to convince participants of the opposite by rational arguments ("adding cognition ''~5~ in the succinct terminology of empirical psychologists). Given these experiments, one is tempted to find the following statement by one of the most prominent supporters of EMH, "I have been an advocate of the efficient market hypothesis for over 30 years ''15~, as at least susceptible to belief perseverance. More generally, if people continue to believe a statement they have just read even when they are told that it is wrong (as in the experiments described above), how can they be expected to change their view of an idea which they have held for many years? This seems plausible irrespective of whether the view is held for subjectively rational reasons, or whether the researcher has an emotional attachment to his belief, which could even reinforce the level of perseverance. In addition to belief perseverance, and in parallel to Bowman/Buchanan's (1995) suggestion about the investment management community, there may be "vested interests" in the academic community supporting the EMH. The foundations of EMH were developed in a climate in which natural sciences in general, and physics in particular, were seen as the role model for social sciences. As pointed out by Preda (2004), economists in the 19th century aimed to construct their models in a deterministic fashion similar to mechanics. 152 To be able to follow physics in casting the seemingly chaotic nature of the outer world into a series of elegant equations must have been tempting for the nascent science of financial markets and investments. This desire for a precise determinability of results may still prevail today among some academics. To give up the precise equations of market efficiency in exchange for seemingly speculative behaviour-based models of asset prices may seem an unattractive proposition. Finally, there may be another set of vested interests in the financial management community, in addition to those identified by Bowman/Buchanan (1995): Conceivably, market practitioners are not interested in academia turning away from the paradigm of market efficiency. It could well be their point of view that universities and business schools every year turn out students brought up to believe in efficient markets, serving a dual purpose: First, they are unlikely to attempt to commit financial heresy by Sherman/Kim(2002), p. 236. Malkiel (2005), p. 1. 152 See Preda (2004), p. 359. 150 151

32

systematically competing for abnormal returns, thereby lessening competition within the market. Second, they represent 'financial markets cannon fodder' by potentially acting in such a way as to create capital market anomalies153.

2.3.7

Conclusion

While the EMH has held as a widely accepted paradigm in the 1970s, the last two decades have produced a plethora of research results indicating that there may be more to return than market risk. The "mini-revolution ''154 in modern finance expected by Jensen (1978) at the end of the 1970s as a result of the first indications of potential violations of the EMH has tumed into a major storm on the efficiency stronghold of modem finance theory. In addition to the anomalies literature, another challenge to EMH is excess volatility, i.e., the finding that share prices fluctuate more wildly than predicted by standard market models. 155 Fama's (1998) contention that market efficiency is upheld because positive and negative abnormal retums occur at roughly the same frequency ~56 falls short of the whole picture: Settling for this interpretation of the anomalies literature bars the path to a better understanding of asset returns. Whether it is now a question of identifying additional risk factors demanding adequate retums, or whether a more behavioural financial modelling of returns is required, continues to pose a challenge for future research. Additional venues for research include chaos theories of financial markets 157 and explanations based on fractal geometry 158. Together, these approaches promise to produce a better understanding of asset prices and market behaviour. In the meantime, market efficiency should best be thought of "as a continuum, rather than the...yes/no dichotomy"IS9: At the one end lies the ideal of market efficiency as first formally described by Fama (1970). Somewhere towards to middle lies the Jensen-type interpretation of market efficiency and Basu's (2004) minimally rational markets. At the other end lies inefficiency. The truth, as in many cases, seems to lie somewhere in between. The discussion between proponents of efficient markets and researchers subscribing to explanations offered by behavioural finance can also be interpreted in the larger context of the development of finance theory as a whole. The late 1950s and 1960s produced the basic concepts of neoclassical finance theory based on seminal works by Merton Miller, 153 For example, by not being aware of weather-related mood swings affecting their trading decisions. 154 Jensen (1978), p. 95. 155 See Shiller (2002), p. 3. Feedback models, which predict rising share prices as a result of rising share prices, are one possibility of explaining excess volatility. See Shiller (2002), p. 15-17. 156 See Fama (1998), p.288. 157 See Cunningham (2000) for a discussion of the applicability of the chaos theory concept to financial markets. 158 See Mandelbrot/Hudson (2005), p. 285-298. 159 Basu (2004), p. 334.

33

Franco Modigliani, Harry Markowitz, William Sharpe, John Lintner and others. Neoclassical finance theory is essentially a theory about the equilibrium in capital markets. 16~ This proximity to economics has on the one hand helped to produce considerable achievements; on the other hand, it has caused criticism for not adequately considering, nor being able to explain, actual financial policy decisions by firms. 161 This criticism reflects similar arguments against neoclassical economic theory, which has been criticised for not properly reflecting reality and for its non-consideration of institutions. 162 From this criticism arose institutional economics and later neoinstitutional economics, which in turn have cross-fertilised finance theory with ideas about principal-agent relationships, information asymmetries, property rights and transaction costs. 163 The advantage of utilizing such concepts is that they better reflect the kaleidoscopic nature of reality. The price to pay is a departure from mathematically precise results, and the realization that finance theory may intrinsically not be able to fully explain reality, nor to precisely predict it. The desire with which financial theory was first developed by Bachelier and his contemporaries, namely to model reality with the mechanical precision of physics, is thus buried. Ironically, by discarding preciseness, finance theory again follows physics, where scientists have realised since the early 20 th century that time and space is not only relative, but also inherently uncertain and fuzzy. ~64

160 See 161 See 162 See 163 See 164 See

Williamson (1988), p. 587. Loistl (1990), p. 51. Williamson (2002), p. 595. Williamson (2002) for a detailed overview of this development. Greene (2004), p. 329-333, for a description of Heisenberg's uncertainty principle.

34

3

Short-term price performance of European equity carve-outs

3.1

Abstract

This chapter extends existing literature analysing a company's share price reaction to the announcement of an intended equity carve-out (ECO) in four ways: First, the unprecedented sample size (in German/European terms) of n=178, originating from 13 European countries in the 1984 to 2004 period, allows the examination of a series of variables used to explain the cross-section of abnormal returns for which results up to now have been either inconclusive or missing. Second, the share price reaction of a parent firm at various process-relevant points in time is analysed, reflecting the continuous information flow to the market. Third, an explicit distinction is made between "clean" and "contaminated" announcement dates. Fourth, the share price reaction of parent firms conducting an ECO at a later point in time to an ECO announcement by another parent firm is analysed. The main results add to the understanding of the sources of the documented announcement period returns for ECOs. First, announcement period abnormal returns are higher for ECOs occurring after 1998 and in countries with higher shareholder rights, supporting the notion of the significance of the relative value of internal vs. external capital markets for the ECO decision. Abnormal returns are also higher if a company states refocusing of the parent firm or development of the subsidiary firm as the primary motivation for the ECO. Additionally, sample firms show abnormally negative returns in the two days following the announcement, driven mainly by 'hot market' ECOs. Second, there is a pattern in share price returns across the three additional dates on which markets learn about the impending ECO: Abnormal returns for parent firms are highest on the first rumour date, close to zero on the first day of bookbuilding, and negative on the first day of trading, but increasing in the subsidiary firm's first day performance. Third, the separate analysis of "clean" vs. "contaminated" announcements reveals that abnormal retums for the latter are significantly higher than for the former, indicating that part of the abnormal returns noted in previous studies may be due to other factors. Fourth, non-announcing companies with future ECO candidates tend to show abnormal (both positive and negative) price reactions when another company announces an ECO, and the reaction is more pronounced for parent firms owning subsidiaries in the same industry as the announcing parent firm's subsidiary.

3.2

Introduction

Over the last two decades the analysis of the share price reaction of parent firms to the announcement of an intended ECO has proven a popular topic for academics. In general these studies find positive abnormal returns around the announcement date. 165 While 165

Appendix 10 and Appendix 11 summarizethe key results of the announcementperiod return analyses from the majority of the existing papers on this subject.

35

studies differ in their choice of explanatory variables, there is some consensus on a number of factors driving these positive abnormal returns. 166 One concern with studies in a German or European setting is the limited sample size, ranging from n=l 1 (Pellens (1993)) to n=58 (Wagner (2004)) in a German context, and n=66 (Bt~hner (2004)) in a European context. The original sample size in the present study (n-178) offers at least three advantages: First, it allows the re-examination of a series of variables used to explain the level of abnormal returns for which results in previous studies have not produced intra-paper conclusive and inter-paper consistent answers. Second, the multi-country nature of the sample allows the use of an additional explanatory variable (country/region) to analyse the link between abnormal returns and a country's institutional settings. Third, an additional variable hitherto unexplored in a European setting (the parent firm's primary motivation for the ECO) is examined. In addition to the official announcement date, there are at least three other points in time when new information regarding the ECO reaches the market (first unconfirmed rumour; beginning of the book-building phase; first day of trading). Investors are likely to react to this information, potentially causing (abnormal) share price movements. While the issue of various dates has been addressed 167, a systematic analysis is lacking. This study fills this void by analysing the share price reaction on multiple dates, as well as analysing the relationship between the reactions on these dates. Existing studies rarely address the issue of contaminated announcement dates. Announcements of ECOs do not always occur in isolation, but rather as part of a broader announcement, e.g., interim or annual results, or as part of a broader restructuring. Not controlling for this contaminates the results of the abnormal return calculation as it mixes different effects. The price for controlling is a decreased sample size, which may have been prohibitive for some of the existing studies, but is not so for this study. Finally, if a company's announcement of an intended ECO conveys news to the market, it seems plausible that this may also impact the share price of similar non-announcing companies. Slovin/Sushka/Ferraro (1995) analyse the share price reaction of rival firms in both the parent and the subsidiary firm's industry. They find that rivals in the subsidiary firm's industry show negative abnormal returns, and rivals in the parent firm's industry show no abnormal performance. In contrast, this study analyses the market reaction of firms with future ECO candidates. If markets view the announcement by one company as increasing the chances of a similar move by another company, the impact on the latter will be positive. If markets want to punish similar companies for not already having issued a similar announcement, the impact will be negative. The direction of the impact is thus undetermined by theory.

166

Appendix 12 groups hypotheses tested in the literature into eight variable classes, and indicates which of these have been found to be relevant in the explanation of abnormal returns. 167 See L6ffier (2001), p. 113-117.

36

The chapter is organised as follows: Section 3.3 reviews the relevant literature on announcement period abnormal returns in the context of ECOs. Section 3.4 discusses the methodology used, and section 3.5 presents the key results of the event study. Section 3.6 extends the analysis by considering various event dates, clean vs. contaminated announcement dates, and the reaction of non-announcing firms to the ECO announcement by another parent firm. Section 3.7 concludes.

3.3

Literature review

3.3.1 Evidence from the US Schipper/Smith (1986) are the first to analyse the effect of an ECO announcement. They contrast the negative share price reaction to a seasoned equity offering (SEO) announcement 168 with the positive share price reaction to an ECO announcement. They offer four explanations (without analysis) for this difference: First, by separating parent and subsidiary firm assets, the negative information about parent firm assets conveyed in an SEO is decreased, and the probability that NPV positive projects will not be foregone increases. 169 Second, both supply of and demand for information regarding the subsidiary firm increases. Third, the increased likelihood of a takeover of the subsidiary firm may increase its market value. Fourth, operating efficiency is likely to increase as a result of business restructuring and more incentive-orientated contracts for subsidiary firm managers. While the first explanation is the seed of the asymmetric information hypothesis formulated later by Nanda (1991), explanation two, three and in particular four form the basis of the divestiture gains hypothesis. The positive announcement period returns are confirmed in a series of studies, incl. Klein/Rosenfeld/Beranek (1991), Michaely/Shaw (1995), Byers/Lee/Opler (1996), Slovin/Sushka/Ferraro (1995), Hand/Skantz (1998), Mulherin/Boone (2000), Chemmanur/Paeglis (2001), Powers (2001), and Glatzel (2003). Nanda (1991) extends Myers/Majluf's (1984) asymmetric information model and develops an asymmetric information hypothesis for ECOs. His main conjecture is that by conducting an ECO, a firm not only reveals information about the value of the subsidiary firm's assets, but also about the value of the parent firm's assets. In his model, parent firms choosing an ECO tend to be undervalued by the market, whereas parent firms choosing an SEO tend to be overvalued by the market. Slovin/Sushka/Ferraro (1995) operationalise this conjecture by assuming that if parent firms choosing an ECO are undervalued, rival firms in the parent firm's industry should show positive abnormal returns when the parent firm announces an ECO, whereas rival firms in the subsidiary firm's industry should show negative abnormal returns. They find empirical evidence for the latter but not for the former. Thus, their results only partially confirm Nanda's (1991) asymmetric information theory. 168

Masulis/Korwar (1986) are the first to document the abnormal negative reaction to the announcement of SEOs. 169 This effect due to asymmetric information is predicted by the Myers/Majluf(1984) model of SEOs.

37

Slovin/Sushka/Ferraro's (1995) small sample size (n=36) may be a source for the inconclusive nature of the findings. In a similar analysis with a larger sample (n=183), Hulburt/Miles/Woolridge (2002) find a negative price reaction to an ECO announcement of companies in the parent firm's industry, which directly contradicts the asymmetric information hypothesis. The authors interpret this finding as support for the divestiture gains hypothesis: Benefits accruing to the announcing party may come partially to the detriment of its rivals. They also argue that both the asymmetric information and the divestiture gains hypothesis predict a negative share price reaction of subsidiary rival companies. Therefore Slovin/Sushka/Ferraro's (1995) finding can also be interpreted as supporting the divestiture gains hypothesis, rather than the asymmetric information hypothesis as originally intended by the authors. Allen/McConnel (1998) argue that managers value control over assets, are therefore reluctant to carve out subsidiaries, and will do this only when the firm is capitalconstrained. Consistent with this "managerial discretion" hypothesis, firms carving out subsidiaries have a lower operating performance and higher leverage ratios. They also find support for the divestiture gains hypothesis: Average abnormal stock returns are higher when the proceeds are used to pay down debt, compared to when the proceeds are retained. Fu (2002) finds that the level of pre-ECO period information asymmetry 17~ is positively associated with abnormal announcement returns. He also finds evidence supporting the managerial discretion hypothesis in the form of abnormal announcement returns being significantly greater when proceeds are paid out to creditors or shareholders, rather than retained for investment purposes. However, there is no support for corporate refocusing as a motive for ECOs, with cross-industry ECOs not exhibiting higher announcement period abnormal returns. Vijh (2002) tests a prediction of Nanda's (1991) model, and finds that returns increase as the ratio of subsidiary to non-subsidiary assets increases (excess returns average 4.9% when the pre-ECO subsidiary assets exceed non-subsidiary assets, and 1.2% when the non-subsidiary assets exceed subsidiary assets), which is inconsistent with the asymmetric information hypothesis and consistent with the divestiture gains hypothesis. Additional support for the divestiture gains hypothesis comes in the form of a more positive market reaction when parent firms divest subsidiary businesses unrelated to their business, when the proceeds are reported to be used to repay debt or to invest in new projects, and when the ECO is intended to create pure plays.

170

Fu (2002) uses a sequential trade microstructure model as developed by Easley/Kiefer/O'Hara (1996) to construct an empirical measure of information asymmetry. Specifically, the model is a mixed discrete-and-continuous time model with informed traders, uninformed traders and a market maker. Traders buy and sell the asset from and to the market maker on a series of discrete days, but within each day trades can occur continuously. Positive and negative news reach the market randomly. The model determines the overall probability of informed trading, which is assumed to be a proxy for the level of informational asymmetry. Empirically the model is estimated using high frequency transaction-level trading data.

38

Hulburt (2003) finds that those parent firms whose subsidiaries are taken over in the years following the ECO show a significantly higher abnormal share price reaction when the initial ECO is announced.

3.3.2 Evidencefrom Germany and Europe The first to analyse ECO announcements in Germany is Pellens (1993), who finds a statistically insignificant share price reaction. This result, which contradicts US evidence, may be due to two reasons: First, the sample size is fairly small (n=l 1), potentially causing spurious results. Second, the date of approval of admission to the stock exchange is used as the event date, which usually lags the actual announcement by some time period during which the market may have already incorporated the news into the share price. Pellens (1993) also finds that a higher level of underpricing leads to more negative abnormal returns for the parent firm; however, the small sample size does not allow assessing whether this relationship is significant. Hasselmann (1997) uses the first trading day as the event date, which seems inopportune given that the announcement date in most cases precedes the first listing day by weeks and often months. He finds an insignificant relationship between underpricing and abnormal retums. Kaserer/Ahlers (2000), using the same event date as Pellens (1993) but with a larger sample (n=23), confirm the US evidence and find a positive announcement period retum. They also find that abnormal retums are higher for primary than for secondary offerings, which they interpret in the spirit of Allen/McConnel's (1998) managerial discretion hypothesis as the market approving that intemal control is replaced by a more efficient extemal control. L6ffier (2001) confirms the positive announcement period retums. 171 She finds abnormal retums to be positively related to the degree of pre-event diversification (measured by self-constructed industry codes), the operating performance of the parent firm 172, and the level of inefficiency of the subsidiary firm under the current parent firm (as indicated in press releases). She finds no relationship between abnormal returns and blockholder ownership, the parent firm's leverage or absolute firm size. Langenbach (2001) also finds that abnormal returns are higher when parent and subsidiary firms are from different industries, and when disinvestments, rather than financing, is mentioned as a reason for the transaction. Contrary to L6ffier (2001), he also finds a positive relationship between pre-event free float and abnormal retums, which he interprets as implying that a high free float is the cause of a high demand for information and a i

171 172

L6ffier's (2001) sample comprises both ECOs and sell-offs. L6ffier (2001) argues that the relationship between abnormal retums and operating performance, measured as EBIT/equity-ratio, is U-shaped. However her charts seem to imply that the U-shaped regression line is driven by one outlier on the negative side. Removingthat outlier could easily lead to a linear positive relationship. See L6ffier (2001), p. 172.

39

173

necessary requirement for the market to exert a controlling influence on the company. Stienemann (2003), also using a sample of both ECOs and sell-offs, finds no relationship between an increased corporate focus and abnormal returns. Also, contrary to L6ffier (2001), he finds a positive relationship between parent firm leverage and abnormal returns, and also a positive relationship between abnormal returns and size (measured as subsidiary to parent firm sales ratio). Elsas/L6ffier (2005), developing theoretical arguments both for a negative and a positive relationship between pre-event ownership concentration and abnormal returns, find that higher levels of pre-event ownership concentration are associated with lower abnormal returns. They thus confirm Langenbach's (2001) finding but explain it differently: They link higher levels of pre-event ownership concentration to better control, thus leading to gains from an ECO to be of relatively lower value for firms which have already been well-managed prior to the event. They find no special monitoring role of banks in either of their proxies for bank control. They also confirm a positive relationship between abnormal returns and both increased corporate focus and the pre-event level of informational opaqueness 174, respectively. Wagner (2004) finds that parent firms in financial distress earn lower abnormal announcement returns, confirming Stienemann's (2003) and rejecting L6ffler's (2001) findings. In contrast to Kaserer/Ahlers (2000), he finds no relationship between abnormal returns and either primary or secondary offering. In his sample abnormal returns are also negatively associated with parent firm's Q, and positively associated with the percentage of the stake sold. This last finding offers an interesting parallel to similar findings from the US by Boone/Haushalter/Mikkelson (2003), who find that operating performance in the long run only increases when the complete stake in the subsidiary firm is sold. Both studies indicate that it is preferable to sell a larger rather than a smaller stake. The common finding of a positive relationship between abnormal returns and increased industry focus is not confirmed by Wagner (2004). Btihner (2004), for the first and hitherto only time using a European sample of ECOs, finds abnormal returns to be positively related to the delta in Tobin's Q between parent and subsidiary firm, and also positively related to the delta in the sales-based Herfindahl-index before and after the event. He interprets these findings as support for a positive market valuation of decreased corporate diversity and thus increased corporate focus. His evidence on the impact of the parent firm's leverage and market period on abnormal returns is inconclusive. 175 Underpricing does not seem to explain the level of abnormal returns, confirming Wagner (2004) and Hasselmann (1997), and rejecting Pellens (1993). The European nature of the ECO sample used in Btihner (2004) makes it the closest comparable study to the present analysis. The present analysis offers at least 173

This hypothesis is rejected by his findings that long-term performance is negatively related to the level of free-float. 174 This last result only holds for same-industryECOs. 175 The coefficients of the two variables are significant in the univariate regressions, while they are not so in the multivariate regressions (see Btihner (2004), p. 169 and p. 173).

40

four advantages: First, Btihner (2004) includes both ECOs and spin-offs in his sample. While some separate statistics are presented, the explanation of the cross-section of abnormal returns is only presented for the combined sample. Spin-offs differ from ECOs in some important aspects 176, and hence results based on a combined sample do not allow drawing valid inferences for other ECOs. In contrast, this study uses an ECOonly sample. Second, Btihner (2004) constructs his sample considering only eight European countries (without explaining the choice), whereas the sample in this study is constructed based on all European countries, and effectively finds ECOs in 13 countries. The sample is thus more exhaustive. Third, the size of the original sample used is considerably larger (n=178 vs. n=66). Sample size is a concern when splitting the sample along temporal dimensions, e.g., to analyse the impact of the market period on abnormal returns: Btihner's (2004) inconclusive finding when analysing the impact of certain 'hot market' periods may be directly linked to the small number of companies in his sample. Similarly, splitting the sample along geographical dimensions may produce lowly populated individual groups, limiting the ability to statistically assess their impact on abnormal returns. Consequently Btihner (2004) does not attempt this, whereas this study does. Fourth, Btihner (2004) does not control for a series of parameters previously shown to impact the level of abnormal announcement period returns, including the level of informational asymmetry and relative transaction size. The variables tested by Btihner (2004) may hence capture some of the impact of these omitted variables. Naturally, model specification is always contestable, and parsimony may be dictated by sample smallness. Using a larger sample (as this study does) comfortably allows the inclusion of relevant control variables without running the risk of model overfitting.

3.4

Data and methodology

Since the inception in its classical form in 1969 by Fama/Fisher/Jensen/Roll the event study has established itself as a prominent research tool in the repertoire of academics. 177 It has found widespread use in many areas of capital markets and corporate finance research, which seems attributable to its conceptual straightforwardness, methodological intuitiveness and economic powerfulness. 178 The basic idea of an event study is to compare the share price return conditional on an event occurring with the share price return conditional on the event not occurring. Since 176 See section 2.1.3. 177According to MacKinlay (1997), p. 13, the first event study is from 1933 by James Dolley. However, according to other papers surveying event study methodological development, time count seems to have started with FamafFisher/Jensen/Roll's(1969) paper, see Binder (1998), p. 111, Kothari/Warner (2005), p. 8 and Armitage (1995), p. 27. 178 Kothari/Warner (2005) count 565 papers published in five leading journals (Journal of Business, Journal of Finance, Journal of Financial Economics, Journal of Financial and Quantitative Analysis, Review of Financial Studies) from 1974 to 2000 using an event study as their primary research methodology.

41

these are mutually exclusive events, the share price return conditional on the event not occurring is simulated by a return-generating model. The results, cumulated across multiple securities and time periods, are then tested for statistical significance. Despite a multitude o f papers written on the subject, the fundamental design o f an event study has remained the same since 1969. While some modifications have rendered the event study methodology even more powerful, other variations have been shown to lead to only marginal improvements in accuracy, and are often outweighed by the additional time and resource requirements to implement them. 179 In the context o f this study, the 180 focus will be on those improvements that are appropriate in the present context. Datastream is the source for share price and accounting data for all available firms. All sample firms which are financial institutions are excluded.

3.4.1

Calculation of abnormal returns

Abnormal returns (AR) are calculated as the difference between realised and expected returns:

(1)

AR~, =R,t-E(R,~)

where Rit is the return o f security i on day t, and

E(Rit) is the expected return of security

i on day t. Returns are calculated as the natural logarithm o f security's i price on day t

(Pit), divided by its price on day t-1181:

(2)

Re, = ln[ P/' /

To conduct significance tests, the abnormal returns are cumulated both across securities and time. Cumulation across securities serves to calculate an average announcement effect across all firms in the sample. Cumulation across time accommodates uncertainty about the exact day o f the impact of the announcement on the share price, ls2 The average abnormal return

179

AARt across n securities for day t is calculated as

Comprehensive reviews of these developments can be found in Brown/Warner (1980), Brown/Warner (1985), Peterson (1989), Armitage (1995), MacKinlay (1997), Binder (1998) and Kothari/Wamer (2005). 180 This excludes, e.g., the use of a multivariate regression framework for the purpose of conducting the study, which has been developed by Thompson (1985), Binder (1985) and Malatesta (1986), and whose major advantage is the possibility to test combined hypotheses, and which is frequently used in the case of regulatory event studies. 181 For a discussion of logarithmic vs. simple returns see Dorfleitner (2003). This study's main results are robust to using simple returns. 182 The order of cumulation is arbitrary, see MacKinlay (1997), p. 24.

42

1 (3)

A A R t = - - * ~ ~A R i t F/

i=1

The cumulative average abnormal return CAAR ~83 across multiple periods T is calculated as

T (4)

CAAR = ~ AAR, t=l

In accordance with standard event study literature, AARt in (3) is calculated as the equalweighted average of the ARit. TM The reason is that "the focus on mean effects...makes sense if one wants to understand whether the event is, on average, associated with a change in security holder wealth ''~85. Alternatively, it is also possible to calculate valueweighted averages, where weights are based on the pre-event market value of the firm's equity. Comparing equal- and value-weighted averages may reveal if abnormal returns are related to firm size. Analysing value-weighted averages is predominantly sensible when the research question is whether a profitable trading strategy exists in which the investor weights the securities in her portfolio by value, rather than equally. However, such a trading strategy is not implementable when the weights are based on pre-event information, and the timing of the events is not known a priori. Consequently, while this study does report value-weighted CAR for comparison p u r p o s e s 186, all following analyses are based on equal-weighted CAR. The various measures of abnormal returns are tested for statistical significance using a variety of test statistics. The null hypothesis in these tests is that abnormal returns average zero. The expected return in (1) is based on a model describing returns conditional on the event not occurring, i.e., "normal" returns. Various return-generating models have been proposed in the literature. Cable/Holland (1999) summarise the most commonly used models - the capital asset pricing model (CAPM), the market model (MM), the mean adjusted returns model (MRM) and the market adjusted model (MAM) - into one general framework and show the value impositions on the parameters by each model187:

(5)

183

184

185 186 187

E(Rit )-- ~i -~-(yi - ~i )* Rj2 -1t-~iRmt

Cumulate average abnormal returns (CAAR) will simply be referred to as cumulative abnormal returns (CAR) in the remainder of this study. See Brown/Warner (1985), p. 7; MacKinlay (1997), p. 24; Binder (1998), p. 113; Kothari/Warner (2005), p. 11. Kothari/Warner (2005), p. 11. See last rows of table Table 2. See Cable/Holland (1999), p. 332-335.

43

where Rfi is the risk-free rate, Rmt is the return on the market portfolio, and a, fl and y are coefficients. Setting 7~- 1 yields the CAPM

(6)

E ( Ri, )--- a i -4-R fi -[--~[~i(Rmt - R ft )

Setting 7~- fli = 0 yields the M M

(7)

g(eit )-- 0Ci -[- ~iRm,

1 Further setting/3/= 0 and ai = R i = ~- * s/=1 Rit, where T is the number of observations in the estimation period, yields the MRM:

(8)

E(Rit)=Ri

Alternatively in (7) setting ai = 0 and fli = 1 yields the MAM:

(9)

E(Rit )

-- R m ,

The decision which return-generating model to use is guided by their empirically documented ability to explain cross-sectional return variability. The CAPM has lost some of its attractiveness as a model for normal returns since it has become clear that it is not a good predictor of actual returns, lss Alternative specifications, including conditional and consumption-based versions, have been developed, and in some cases seem to perform better than the original C A P M developed by Sharpe (1964) and Lintner (1965). 189 However, these models have not caught on in actual event studies. Arbitrage pricing theory (APT) and other multifactor models have also been shown to add little benefit, mainly because the marginal explanatory power of the additional variables is small. 19~ Similarly, using industry indices as explanatory factors in return-generating models does not improve the results in an event study. 191

188

189

190 191

See Fama/French (1993), p. 3: "The cross-section of average returns on U.S. common stocks shows little relation to [...] market [3s of the Sharpe (1964)-Lintner (1965) asset pricing model". See also section 2.3.4 of this thesis for an overview of capital markets anomalies invalidating the CAPM. For an example of both conditional and consumption-based CAPM versions, see Lettau/Ludvigson (2001). See MacKinlay (1997), p. 18-19. See Thompson (1988), p. 84.

44

The MM, on the other hand, is simple to implement, dominates the MRM and MAM in all situations and has been shown to produce robust results. ~92It therefore represents the model of choice for the majority of event studies. To implement the market model, the model coefficients in (7) are estimated in the estimation period via OLS for each security. The market index used for each security is a broad country index (MSCI). The expected return is thus

(1 O)

E(Rit ) -- 6~ie + t~ie * Rmt

where aie and fl~e are the OLS estimates of the coefficients in (7), and the abnormal return is calculated according to (1) as"

(11)

ARit -- Rit - ( a i e + i~ie ~ emt)

The coefficients are assumed to be constant throughout the estimation and event periods. This assumption is analysed and relaxed in section 3.6.1.

3.4.2

OLS assumptions

For the OLS estimates to have the desired BLUE-characteristics 193, certain assumptions are required: 9

The residuals are normally distributed with a mean of 0.

9

The residuals are not correlated with the explanatory variable (the market return).

9

The variance of the residuals is constant (i.e., no heteroscedasticity).

9

The serial covariance of residuals across time is 0 (i.e., no autocorrelation).

First, the distribution of residuals approximates a normal distribution more closely than do daily r e t u r n s . TM The remaining non-normality of residuals has not been found to be an issue in event studies. 195 Second, correlation of residuals with the market retum can be a problem when the event is more likely to occur in bull markets than in bear markets: Basing expected returns on data from market periods which are fundamentally different from the event period may lead to misspecified r e s u l t s . 196 The increased frequency of ECOs in the 'hot market' period between 1998 and 2000 may make the current sample subject to this issue. A 192 See Brown/Warner (1980), Brown/Warner (1985) and Chandra/Moriarity/Willinger (1990). 193 Best linear unbiased estimators. See von Auer (2003), p. 73. 194 See Brown/Warner (1985), p. 10 and Berry/Gallinger/Henderson (1990), p. 75. 195 See Henderson (1990), p. 293. 196 See Henderson (1990), p. 294.

45

potential remedy is to use prediction errors, rather than standard deviations, in the significance tests. As will be discussed in section 3.4.5, existing studies show that this modification leads to only marginally improvements in results. Third, to test for heteroscedasticity, a standard Goldfeld-Quandt test is applied. 197 Heteroscedasticity is found for 56 companies, and implies inefficient (yet consistent) estimators. An established way of dealing with heteroscedasticity is to use generalised least squares (GLS) instead of O L S . 198 GLS transforms the model into another model where the variance is constant, but which still allows the resulting parameter estimates to be interpreted analogously to the original (OLS) model. Fourth, to test for autocorrelation, a standard Durbin-Watson test is applied. Significant autocorrelation is found for 13 companies, and again implies inefficient (yet consistent) estimators. One way of dealing with autocorrelation is to use the estimated generalised least squares (EGLS) method by Cochrane/Orcutt, further developed by Prais/Winsten. 199 In this study the iterative version of the Cochrane/Orcutt procedure is implemented. The basic idea is to repeatedly estimate a transformed version of the model, including the correlation term responsible for the autocorrelation, until there is only marginal improvement in the accuracy of the correlation term. z~176 Both adjustments (for heteroscedasticity and autocorrelation) have been implemented, and the resulting abnormal returns and significance levels are found not to change materially. Therefore in the remainder of this study these adjustments are neglected. 2~ Another adjustment suggested in literature is for non-synchronous trading, i.e., the use of prices assumed to have been established within equal-spaced time periods when in reality they have not. This impacts the variances and covariances of shares and thus leads to biased market model parameters. Scholes/Williams (1977) develop estimators robust to non-synchronous trading. The basic idea for the [3-estimator is to use a weighted average of correlation coefficients based on the covariance of the day 0 share price return with the d a y - 1 , 0 and +1 market return, where the weighting is based on the market return autocorrelation. 2~ Various simulation studies have found the benefit

197

See von Auer (2003), p. 360-363. See von Auer (2003), p. 366-369. 199 See von Auer (2003), p. 396-397. 200 Another potential issue is that autocorrelation of residuals transfers into and therefore biases the calculation of cumulative abnormal returns; see Salinger (1992), p. 40-41. The bias decreases in the ratio of the days in the event period to the days in the estimation period, see Sweeney (1991), p. 380, and is unlikely to be material in this study, given the sufficient length of the estimation period. 201 Besides materially unchanged results, another reason why the heteroscedasticity adjustment has not been upheld is because it flaws the t-statistic in its basic version: The transformation of the model leads to the residuals having a standard deviation of very close to 1, which inflates the variance of the CARs calculated according to MacKinlay (1997), p. 21, and thus leads to very low t-statistic values. 202 See Scholes/Williams (1977), p. 315-319. 198

46

of implementing this adjustment to be negligible2~ and thus this adjustment is not implemented in this study.

3.4.3

Estimation and event window

Ideally an event study should determine the exact date of the impact of the event on the share price so that only one event window, [0], has to be analysed. In practice, longer event windows are chosen, for two reasons. First, determining the date of the impact of the announcement on the share price is not feasible with certainty. TM Second, an attempt is made to capture event-related share price movements not directly linked to the official announcement of the event, with the movements caused by either insider trading before the announcement or a slower than expected market reaction after the event. While the former cause seems plausible, the latter violates the assumption of an immediate market reaction to events, a tenet of market efficiency. This implies the applicability of asymmetric event windows, i.e., with the period preceding the event being longer than the period following the event. This choice partially alleviates the downsides of longer event windows: an increased probability of other events contaminating the results, and a lower power of test statistics. 2~ In general accordance with these considerations and other ECO event studies 2~ multiple event windows are used in this study: [-10;0], [-5;0], [-2,0], [-1;0], [0], [0;+1], [-1;+1], [-2;+2], [-5;+5] and [-10;+5]. While the choice of the specific event windows is arbitrary, a selection is required so that results can be presented in a readable way. Additionally, Appendix 13 lists the abnormal returns for each individual day in the [10;+10] event window period, allowing the inclined reader to calculate CARs in event windows of alternative durations. The estimation period is designated arbitrarily (but in line with other event studies which typically choose a period between 100 and 300 days) 2~ to be the period f r o m 230 days to -50 days relative to the event. The buffer between the end of the estimation period and the event itself is to ensure that the parameters are estimated independent of any event-related effects. The length of the estimation period, 180 trading days, represents a trade-off between a longer period, making the parameter estimates more reliable, and a shorter period, taking into account parameter instability across time.

203 See Brown/Warner (1985), p. 16 and Cowan/Sergeant (1996), p. 1744. 204 See also discussion in section 3.4.4 on the determination of the announcement dates. 205 See Brown/Warner (1985), p. 14-15. 2~ Appendix 10 and Appendix 11 for referenced periods. 207 See Peterson (1989), p. 38.

47

3.4.4

Announcement dates

The identification of the announcement dates is a crucial step in an event study, and Brown/Warner (1980) admonish researchers to spend a significant proportion of their time in determining these dates exactly. 2~ Announcement dates are identified by searching the LexisNexis and Factiva databases. LOffier (2001) discusses at length the appropriate way of determining the announcement dates in her study of sell-offs and E C O s . 2~ TO qualify as an announcement date, the announcement must be sufficiently concrete regarding the company's determination to carry out the ECO, must have been issued by the company or a company's representative (usually a board member), and the timing of the event must be in the foreseeable future (usually within the next 12 months). Clearly such specifications will always have a subjective element. Therefore in addition to the 'classical' announcement date, three other dates at which relevant news reach the market are analysed (see section 3.6.2.1). Two sources of uncertainty exist in determining the exact day of impact of the news on the share price. First, the official announcement may be preceded by periods of insider trading. Second, it is unclear whether the news reach the market on the same day the announcement is made. For newspaper articles, the impact day is likely to be the day before the article, if the announcement is issued during trading hours but before expiration of the newspaper's printing deadline. For news agency (e.g., AFX) announcements, the impact day is either the day of the announcement, if it is issued during trading hours, or the following day. In this study, the pinpointing of the announcement date is therefore based on the source mentioned in the database: For newspaper articles, the announcement is assumed to have occurred on the day preceding the date of the article. For news agency press releases, it is assumed to have occurred during trading hours, and thus the announcement date is set to the date of the release. 21~

3.4.5

Choice of significance test

Test statistics are one area where substantial progress has been made since the initial event study in 1969. A 'good' test statistic is one that will not falsely indicate an abnormal return when there is none (type 1 or a error) and not leave an abnormal return undetected when it exists (type 2 or [3 error). The basic idea behind all test statistics is to make a statement about the probability of finding the result of one's study assuming that a certain hypothesis is true. In event studies, the null hypothesis most often is that of no abnormal return. Consequently test 208 See Brown/Warner (1980), p. 249. 209 See L6ffler (2001), p. 110-116. 210 An alternative is to use Ball/Torous's (1988) maximum likelihood procedure explicitly dealing with event date uncertainty. Their analysis shows that the gains from using this methodology, compared to extending the event window by one or two days to accommodate the uncertainty, are marginal.

48

statistics take the form of a ratio of some measure of abnormal return, divided by some measure of standard deviation. The most basic test statistic uses the ratio of the average (cumulative) abnormal return to its own standard deviation, where the latter is based on the error variance from the market model. 21~ The quality of the test statistics is related to the characteristics of the data. Deviations of these characteristics from those required by statistical theory will worsen the quality of the test statistics. In an event study, there are three main causes for such deviations: 2~2 9

Heterogeneity of the error variances

9

Cross-correlation of errors

9

Shifts in variance around the event

Heterogeneous error variances can occur whenever the volatility of the securities considered differs, which usually is the case. Not controlling for this will lead to negatively biased test statistics because, as Boehmer/Musumeci/Poulsen (1991) point out, securities with large variances dominate the test. 213 Cross-correlation of errors can occur when the shares have the same estimation and forecast periods (event-clustering), the probability of which increase when companies are from the same industry (industryclustering). 214 Not controlling for this will lead to positively biased test statistics (due to under-estimation of the sample variance) and thus to higher rejection rates. Shifting variances can occur when the shares are more volatile in the event period, compared to the estimation period. 215 Not controlling for this will again lead to positively biased test statistics (due to under-estimation of the sample variance) and thus to higher rejection rates when the variance in the event-period is higher than in the estimation period. Various adjusted test statistics have been proposed to deal with one or more of these issues. 216 The following notation is used:

ARi,

abnormal return for security i on day t

211 See MacKinlay (1997), p. 24. 212 See Armitage (1995), p. 41. 213 See Boehmer/Musumeci/Poulsen (1991), p. 1991. Intuitively, the reason lies in the unequal impact of an increased variance on the calculation of the t-statistic: Assuming that the expected abnormal return of high variance securities has the same expected value as that of low variance securities, the numerator is not impacted by the increased variance, whereas the denominator increases because of the increased (strictly positive) variance. 214 Brown/Warner (1980) find that the impact of event-clustering is negligible when an equal-weighted index is used. However when a value-weighted index is used actual rejection rates are in excess of theoretical rejection rates, see Brown Warner (1980), p. 235. 215 Kothari/Warner (2005) describe an increase in event-period return volatility as "economically intuitive" (p. 14), with uncertainty-increasing factors either causing the event, or the event itself being the cause of an increase of uncertainty in the firm's economic environment. 216 The following exposition of test statistics is based on Armitage (1995), p. 35-42.

49

1 ,

ARt--- ~

N

Z ARit

average abnormal return across all securities on day t (portfolio

i=1

residuals) N

number of firms in sample

Rm

average market return in estimation period

Rmt

return on market index on day t

Rmr

return on market index on day r (in estimation period)

Si --

I/1 T

2 t~: ARi2t estimation period standard error for security i number of days in the estimation period

T

average portfolio residuals

/t = ~ AR t t=l

Share time series method: To address differing error variances, the abnormal returns of each share are standardised by their standard error. 217 This decreases the relative contribution of shares exhibiting high volatility levels and thus prevents an overestimation of the sample variance.

N

ARit

t shareti..... ies = Z Si "~ Portfolio time series method: To address cross-correlation of errors, shares are grouped into a portfolio, and the resulting time series of average (portfolio) errors is used to calculate a standard deviation. Using portfolio errors thus controls for potential crosscorrelation of errors.

tportfolioti..... ies

=If

AR t 1 ,~-,(--~,__~)2T I -~

217

t=l

Patell (1976) is the first to suggest this technique.

50

Cross-sectional method:

To address shifting variances, the standard deviation is

calculated from the observations on the event day, ignoring previous information. This prevents using biased variance estimates from the estimation period. 218

ARt t ......... tional = ] ( N I _ I , ~ ( A R i t _ A R , . j

,

i=1

Prediction error method:

This method is based on the share time series method but

calculates a more accurate measure of the error term used for standardization by considering that the event period error is a prediction error, rather than a regression disturbance term. 219 Simulation studies have shown this adjustment to lead to only marginal improvements in results. 22~

-- ~'~ t predictionerror '~.=

ARit

Si

1 1-~'--'1 t- (Rmt --"~m~ re ~'~(Rmr -'-emm~

t=l

Standardised cross-sectional method:

Each of the above methods manages to remedy

one of the issues but is subject to the other two, respectively. A test statistic addressing two of the issues - heterogeneous error variances and shifting variances - has been developed by Boehmer/Musumeci/Poulsen (1991).221 It combines the standardization of the share time series/prediction error method and the cross-sectional aspect of the crosssectional method.

t std cr ........

ARit Si

,~N

ti~ -I(N-11 * ~(Ai=I Rit - ARt )2 /

A final possibility to address all three issues is the use of a generalised least squares estimation, where the standardization of a share's abnormal return takes into account both its own variance and its covariance with the other shares' abnormal returns. The

218 An alternative for dealing with event-induced heteroscedasticity is the use of a generalized least squares technique as suggested by Collins/Dent (1984). 219 This is the case when estimation and event periods are separate from one another, as is usually the case in event studies. See Patell (1976), p. 256.

220 See Brown/Warner (1980) and Brown/Warner (1985), and Corrado/Zivney (1992). 221 See Boehmer/Musumeci/Poulsen (1991), p. 258.

51

caveat of this approach is that estimating the variance-covariance matrix limits the number of shares that can be used to the number of observations for each share. The appeal in practice of this method seems to be limited. 222 For this study, the decision which test statistic to use is based on the sample's characteristics. A potential heterogeneity of error variances is analysed using Bartlett's test for equality of variances, which has been found to be one of the most powerful tests of heterogeneity of variances. 223 Using this test, the null hypothesis of homogeneity of the error variances is rejected at the 1% significance level. Since it is known that Bartlett's test is sensitive to deviations from the normality-assumption, an alternative, which is less sensitive to non-normality- Levene's t e s t - is also used. 224 Again, the null hypothesis of homogeneity of the error variances is rejected at the 1% significance level. 225 The sample in this study is thus characterised by heterogeneous error variances. A potential cross-correlation of errors is unlikely in this study, given the dispersion of cases across both time and industry. An analysis of pairwise correlation coefficients for all abnormal return time series is performed to produce descriptive evidence on the (in)existence of cross-correlated errors. Out of all 100x99/2=4,950 correlation coefficients, 1.2% / 4.6% / 9.1% are significant at the 1% / 5% / 10% level. Given the random variable nature of the correlation coefficients, such closeness of actual to theoretical rejection levels is expected in the case of no correlation. A potential increase in error variances around the event is analysed in two ways. First, as suggested by Kothari/Warner (2005), the cross-sectional variability of returns during the event and non-event periods is calculated as the ratio of the average cross-sectional variances during the event and non-event periods. 226 The non-event period is defined as the estimation period, i.e., [-230;-51] days around the event. For descriptive purposes, the results from using multiple event periods are reported. For all combinations of event periods ranging from 50 to one day(s) prior to and one to ten day(s) following the event day, the mean (median) ratio of variances of event to non-event periods is 1.234 (1.232). The maximum (minimum) ratio is 1.718 (1.086). There is thus support for the hypothesis of a higher variance in the event period compared to the estimation period. Second, as suggested by Sanders/Robins (1991), the ratio of a variance estimator 2 and Sm3 2 ) to the standard market model residual incorporating a variance change (sin2

222 223 224

225

226

See Armitage (1995), p. 42. See DeShon/Alexander (1996), p. 271. See ILMES - Internet-Lexikon der Methoden der empirischen Sozialforschung in http://www.lrzmuenchen.de/-wlm/ilm 110.htm. The results are robust to all three alternatives of the Levene's test, which alternatively use the mean, median and trimmed-mean in the calculation of the test statistic. See Brown/Forsythe (1974), p.364365. See Kothari/Warner (2005), p. 15.

52

2 and O"m3 2 ) is used to evaluate a variance not incorporating a variance change (Om2 potential variance increase 227, where

$2m2= ~(-~t--m2~/Sir and sm3 2 =~(-~t-m3~/(sir) 2 , and U i='

1

""' ~ ' ( N _ I ) L i=1

0"m22

=

N

i=1

Sir

and

U

1

i=1

(Sir):

(N_I)Z

2

O"m3 - -

U 1 '

i=~lSi-2r with

SirS i l'lt'L =

+

(Rmt-emm~ -

Ti ~(Rmr_-~mm~

~j = l ARt Sir and ~l '

, m2 = ~

~j=l(Sir) ARt 2 Z ]

m3 = - - - y - - - - ~

Significance is assessed by a non-parametric ranking procedure comparing the ratio on the event day to the ratios on all non-event days. 228 Applying this test the null hypothesis of no variance increase is rejected at the 5% (m2) and 1% (m3) significance level, respectively. This again supports the hypothesis of a higher variance in the event period, compared to the estimation period To summarise, the sample in this study is affected by two out of the three issues identified above: Heterogeneous error variances and increases in the error variances around the event. Cross-correlation of errors is, as expected, not an issue. Consequently, the test statistic of choice for this study is the standardised cross-sectional method by Boehmer/Musumeci/Poulsen (1991). Other statistics are reported for reference. In addition to the parametric test statistics, non-parametric test statistics are reported. The advantage is that they are assumption-free regarding the abnormal returns distribution. Corrado/Zivney (1992) find that both a sign and a rank test dominate a ttest in detecting abnormal performance, with the rank test dominating the sign test. 229 For the rank test, the standardised rank of each abnormal return for sample firm i in its sample period is determined as

227 228 229

See Sanders/Robins (1991), p. 311-312. See Sanders/Robins (1991), p. 321. See Corrado/Zivney (1992), p. 477.

53

Ui t "-" rank(ARit)

l+Mi where A4i is the number of non-missing returns for sample firm i. The test statistic is

__ 1 s (Ui0 - 0.5 ) tranktest -- % i ~

Su

i=1

'

where the standard deviation s, is calculated as

Nt

s~ =

2

Z(Uit

1 , ~

i--1

~--~

-0.5)

~

,

t=-230

where

Nt is the number

of available returns on day t.

For the sign test, the sign of each abnormal return in the sample period is determined for each firm as 23~

Git = s i g n ( A R , t - M c d i a l / l ( A R i ) ) ,

where ARit is the abnormal return for sample firm i on day t. Git equals -1, 0 or 1 if the sign of the difference between the abnormal return on day t and the median abnormal return over the whole sample period is negative, 0 or positive, respectively. The test statistic is

_

1

,s

t signtest -- ~

i=1

Sg

where the standard deviation

sg is calculated as

Nt

2

i~l (Git ) Sg z t=-230

where

230

Nt is the number

of available returns on day t.

See Corrado/Zivney (1992), p. 468.

54

3.5

Empirical

3.5.1

results

Results of abnormal return calculation

Table 2 shows the results of the abnormal returns calculation, t-statistics and associated p-values (p0

57

65

63

61

67

58

60

53

58

53

t-stat rank test

NaN

NaN

NaN

NaN

4.275

NaN

NaN

NaN

NaN

NaN

p-value rank test

NaN

NaN

NaN

NaN

0.000

NaN

NaN

NaN

NaN

NaN

t-stat sign test

NaN

NaN

NaN

NaN

1.426

NaN

NaN

NaN

NaN

NaN

p-value sign test

NaN

NaN

NaN

NaN

0.154

NaN

NaN

NaN

NaN

NaN

CAR (vw)

1.2%

1.9%

1.8%

0.7%

1.0%

0.9%

0.6%

1.5%

0.4%

-0.2%

t-stat (vw)

1.320

2.732

3.789

1.869

3.547

2.258

1.323

2.398

0.476

-0.185

p-value (vw)

0.190

0.007

0.000

0.065

0.001

0.026

0.189

0.018

0.6355

0.8539

Table 2: C A R s across various event day windows

Abnormal returns are calculated for 100 out of the 178 sample firms. Excluded cases include ECOs carried out by financial parent firms TM, those whose announcement date has been contaminated by the release of other significant company news, and cases with insufficient data resulting either from the parent firm not being listed for the complete

231

These cases are excluded because certain accounting items used as independent variables in the analysis of the cross-sectional distribution of CARs have a different meaning for financial institutions.

55

estimation period, or from more than 50% zero returns during the estimation period. 232 Table 3 details the reduction from original to used sample size. CARs are significantly positive throughout all event windows, with the exception of those including more than one day following the event. Figure 1 visualises the CARs around the event date, where individual day returns have been summarised to both mean and median measures. Both measures show positive abnormal returns around the event date. A surprising finding is the negative abnormal return on the second and third day following the announcement. This issue is further analysed in section 3.5.2. R e a s o n for exclusion

Original firms in sample Event date not available Financial firms Contaminated announcement date More than 50% zero returns during estimation period Share price not available in DS during estimation period

Companies

2 24 38 7 7

C o m p a n i e s left after deduction

178 176 152 114 107 100

Table 3: Sample firms used in event study

Figure 1: CARs around event date - [-10;+10] day window

232

These cases represent days with no trading. The possibility of a closed stock exchange is eliminated by only considering those zero return days on which the respective country index experiences a nonzero return.

56

Table 2 shows various test statistics to assess statistical significance for various event windows. The relationships between these various test statistics confirm the considerations made above on the appropriate selection of the test-statistic, and are indicative of the sample characteristics. First, tsharetimeseries always exceed tbasio The reason is that if heterogeneous variances are not considered but occur in the sample, tstatistics will be negatively biased. Second, tportfotio is close to tbasic, exceeding it for all positive abnormal return event windows. If cross-correlation were a problem, the reverse relationship would be expected, because tbasic would be upward biased. Third, tcross-sectional a r e smaller than tbasic for seven out of the ten event windows, and the difference is close to zero in the remaining three event windows. The reason is that if increased variances are not considered but occur in the sample, t-stats are positively biased. Fourth, tpredictionerror are always smaller than but very close to tsharetimeseriesl. The adjustment for the prediction error nature thus seems of little practical relevance in this study. Fifth, tstandardised-cross-sectional are smaller than tsharetimeseriesfor all event windows, and slightly exceed tcross-sectionalfor seven out of ten event windows. Given the hybrid nature o f tstandardised-cross-sectional, this relationship is expected. To summarise, the results largely confirm existing studies on the level and significance of abnormal returns for ECO announcements. The empirical relationship between various t-statistics largely corresponds to theoretical considerations. The selected tstatistic has been shown to deal efficiently with the actual sample characteristics (heteroscedastic variances and an increased event period variance), and is thus appropriate for the specific sample under consideration.

3.5.2

Negative abnormal returns following the announcement

A somewhat startling finding in Figure 1 is the negative abnormal return in the two to three days following the announcement. The magnitude of the negative return cancels out most of the gains from the previous days. This result does not seem to have found widespread discussion in literature. The following section reviews empirical evidence from existing studies on this issue, and conducts a series of further analyses. Comparison with previous studies is difficult because many published papers do not graph CAR, nor do they report individual day abnormal returns. Those papers giving the required detail seem to confirm the finding in this study. Michaely/Shaw (1995) find that abnormal returns for their US sample of ECOs are significantly negative on the third day after the announcement. 233 The reaction over the [-2;2] day window is insignificantly different from 0, which they see as support for their hypothesis of ECOs being carried out when subsidiary firm assets are overvalued. Btihner (2004), for his European sample of ECOs, depicts CAR over a [-60;20] day window, and the chart shows a negative return of approx. -2% in the 2-5 days following the announcement. TM 233 See Michaely/Shaw (1995), p. 10. 234 See Btihner (2004), p. 151.

57

Wagner (2004), for his German sample of ECOs, finds that the gains on the announcement date are completely reversed in the 20 days following the event. He speculates that market participants overreact to the news of an impending E C O . 235 Although short-horizon return reversals have been previously analysed in literature, their "source [... ] remains an unresolved debate ''236. Lehman (1990) finds that portfolios of firms that had positive returns in one week tend to have negative returns in the next week, while portfolios with negative returns in one week tend to have positive returns in the next week. Return reversal has been explained as either being the result of changing required equilibrium returns (Ball/Kothari (1989)) or market microstructure phenomena such as the bid-ask spread (Jegadeesh/Titman (1995)), or resulting from investor overreaction on the basis of behavioural models (Barberis/Shleifer/Vishny (1998)). Mase (1999) believes that short-term trend reversals are the result of a market overreaction and its subsequent correction. Kaestner (2005) finds that firms with past earnings surprises experience opposite sign abnormal returns at the time of the subsequent earnings announcement. He offers the representativeness bias as an explanation: Investors extrapolate recent earnings surprises which on average are not sustained by future earning announcements, and stocks therefore experience a return reversal. Du (2002) develops a behavioural model explaining overreaction as the consequence of a growing investor confidence level: When average confidence is low, prices underreact to news about a certain event, leading to same sign post-announcement price drifts. Based on a self-attribution bias, average investor confidence increases because investors associate the price drift with their initial investment decision. When average confidence is high, prices overreact to news about similar events, leading to an opposite sign postannouncement price movement. Applied to this study, Du's (2000) model implies that investors who have repeatedly experienced positive announcement period returns may overreact to the news of an impending ECO, leading to the negative abnormal returns on the second day following the announcement as the overreaction is corrected. To empirically analyse the issue, daily abnormal returns are calculated for each day in the [-10;+ 10] day window around the announcement. Appendix 13 shows that both the mean and the median abnormal return on day two following the announcement are significantly negative. The median negative return on day three is only marginally significant based on a rank test. Abnormal returns on the remaining days following the announcement show no significance. Appendix 14 extends the event window to [50;20] days around the event. Three observations are noticeable: First, there is no abnormal performance in the [-50;-30] day window before the event. Second, abnormal returns seem to occur on certain days spread across the [-25;-5] day window. Third, abnormal returns are negative in the [+10;+20] day window, cancelling most of the gains from the previous 40 days. A potential explanation for this pattern is that either 235 See Wagner (2004), p. 16. 236 Subrahmanyam (2003), p. 1.

58

insiders or speculators begin buying the parent firm's stock about a month before the official announcement. This may be caused by or be the cause of rumours in the market about an impending announcement. This idea motivates the systematic analysis of abnormal share price reactions on days other than the official announcement date, as presented in section 3.6.2.1. Following the event date announcement, investors overreact to the positive news, and subsequently correct their overreaction, leading to negative returns. In the terminology of market traders, the parent firm share seem to lose some of the attractiveness it had prior to the official announcement, and investors may decide to 'sell on good news', leading to a partial or even complete reversal of prior gains. A simple way of testing this hypothesis is to analyse the correlation of [0] day and [+2] day abnormal returns. If investors indeed overreact to the initial announcement and thus 'sell on good news', those firms experiencing high abnormal returns on the announcement date should experience high negative abnormal returns on the following days. Figure 2 plots the abnormal returns on the two days. The correlation coefficient is significantly negative (p=0.0015). The idea of a trend reversal caused by investors correcting their overreaction (or, alternatively, by cashing in some of their gains) thus finds some support. The analysis also implies that the assessment of value creation occurring in an ECO is difficult when considering short time periods only. A proper analysis of potential value gains may necessitate a long-term analysis of operating and price performance, as conducted in a later part of this thesis.

Figure 2: Correlation of abnormal returns: Trend reversal

59

Additionally, the observed trend reversal on the second day following the event may be driven by various event factors. Geographical, temporal, industry and size factors can be conceived to play a role. Specifically, trend reversal in countries with higher shareholder rights should be lower, assuming that ECOs in those countries are carried out for value-enhancing motives, rather than selfish management motives. Trend reversal should be higher in the new economy period (1998-2000) than in earlier or later years, given the increased level of speculative prices in stock markets. Linked to this, the trend reversal should be higher in high technology-related industries which contributed to the stock market bubble in that time period. Finally, Zarowin (1990) finds that the overreaction hypothesis noticed by Jegadeesh (1990) and Lehman (1990), whereby portfolios with negative returns in one week tend to have positive returns in the following week, is largely due to small firms. If overreaction is mainly due to firm size, then small firms should show a higher level (in absolute terms) of second day negative abnormal returns compared to large firms.

Table 4 shows the results of a two-groups difference of means test for each of these four hypotheses. Second day abnormal returns in countries with high shareholder rights are less negative compared to countries with low shareholder rights, but the difference is insignificant (p=0.5398). Abnormal returns in the new economy period are significantly (p=0.0905) more negative than abnormal returns outside of these years, which are very close to 0. Abnormal returns in high-technology related industries (empirically defined as SIC code 4 and 7 industries) are not different from returns in other industries. Abnormal returns for small firms (defined as the bottom quartile of sample firms by market equity on the IPO date) are more negative than for the remaining firms (-1.1%, vs. -0.3%), but the difference is statistically not significant (p=0.2195). In sum, the analyses imply that abnormal returns on the second day following the announcement are partly due to companies carved out in the 1998-2000 new economy period. 237 Hypothesis

Mean Mean p(diff) Firms (Group 0) (Group 1) (Group 0)

Firms (Group 1)

High vs. low shareholder rights countries

-0.0044

-0.0084 0.5398

75

23

Non-new vs. new economyperiod New vs. Old Economy

-0.0008

-0.0101 0.0905

51

47

-0.0049

-0.0063 0.8167

70

28

Normal vs. Small firms

-0.0033

-0.0111 0.2195

73

25

Table 4: Tests for hypotheses regarding [+2] day window abnormal returns

237

The result is robust to the exclusion of 1998 from the hot market period.

60

3.5.3

Cross-sectional regression analys&

In an attempt to establish causality it is common practice to regress the abnormal returns on a series of event and firm characteristics. 238 OLS estimates are employed. 239 The general regression equation to be estimated is

Abnormal returns = f (event and firm characteristics)

It is well known that analysing cross-sectional firm data leads to heteroscedasticity problems, leading to inefficient (yet consistent) estimators. 24~ For this reason t-values are corrected using the White (1980) covariance matrix estimator. The variables included in the model are based on the hypotheses of interest, as well as established results from the existing literature (control variables). First, these independent variables are described, in conjunction with some descriptive statistics, and the motivation for their respective use is clarified. Second, the model is estimated and discussed.

3.5.3.1 Explanatory variables Date: The period in which an announcement occurs could impact abnormal returns in at least two ways. First, abnormal returns could be higher in 'hot market' periods. Second, there could be a trend in abnormal returns across time. Appendix 15 shows the mean abnormal returns per year, which are positive in all but four years, and significantly positive (at least at the 10% level) in five years. To test whether abnormal returns are driven by the 'hot market' periods resulting from the emergence of new stock markets for high-tech companies 241, the sample is divided into those companies with an event between (and including) 1998 and 2000, and those companies with an event outside o f this period. While the mean abnormal return is higher for the former group (1.4% vs. 1.2%), the difference is not statistically significant. Khanna/Palepu (2000b) conjecture that external capital markets become more efficient over time and internal capital markets therefore lose part of their value. 242 This is consistent with the idea of a growing popularity of refocusing companies on their core activities, and implies abnormal returns trending upward over time. To test this, the 238 239

240 241

242

See Binder (1998), p. 116. In addition to the OLS estimator, Karafiath (1994) tests three more refined estimators (consistent estimator least squares, weighted least squares, feasible generalized least squares), and finds that with sample sizes exceeding 75 these offer no benefit over OLS, even in the case of extreme calendar time clustering, i.e., shared event dates. See Drobetz/Kammermann/W~ilchli (2003), p. 10. See Ritter (2003), p. 422-426 for a discussion of the occurrence and timing of these new stock markets across Europe. See Khanna/Palepu (2000b), p. 281.

61

differences in mean abnormal returns between all combinations o f two consecutive periods are analysed. A p p e n d i x 16 shows that in all cases after 1987, later periods are characterised b y higher m e a n abnormal returns. This difference is significant at the 5% level for the 1984-1996/1997/1998 vs. 1997/1998/1999-2004 periods. The idea o f rising abnormal returns across time is thus supported, and a d u m m y variable is created, equalling 0 for all companies with an event between 1984 up to and including 1997, and 1 for all companies with an event including and after 1998.

Region:

The

geographical

dispersion

o f the

sample

allows

analysing

whether

institutional settings in certain geographies are associated with the level o f abnormal returns. Table 5 shows the abnormal returns by country for all event windows. Table 6 shows the p-values for all event w i n d o w s o f a variance and Kruskal-Wallis analysis testing whether country classification has explanatory p o w e r regarding m e a n and median abnormal returns. For most o f the shorter event windows, the difference between the country groups is significant. The fact that m e a n differences seem more significant than median differences implies n o n - n o r m a l i t y and a certain impact o f outliers, which is natural given the fairly small country group sizes. Event window

BE

[-10;0] -11.5% I-5;01 -3.2% I-2;0] -2.4% [-1;0] -2.5% [01 0.3% I0;11 -2.2% [-1;1] -4.9% [-2;2] -8.3% [-5;5] -17.8% [-10;5] -26.1% N 3

CH

10.6% 7.6% 5.7% 6.2% 6.6% 5.9% 5.4% 5.1% 4.7% 7.7% 5

D

2.7% -0.5% 1.0% 0.9% 0.7% 0.2% 0.4% 0.2% -1.2% 2.0% 41

FN

FR

2.2% -0.1% -2.5% 0.4% 0.6% 2.2% 2.0% -0.5% 4.6% 6.9% 5

0.1% 2.3% 3.2% 2.2% 0.6% 0.3% 1.8% 1.6% 2.1% -0.2% 8

IT

1.6% 1.4% 0.2% 0.0% 0.5% -0.4% -0.9% -0.5% 1.8% 2.0% 6

NL

NO

PT

1.3% -0.5% -0.3% 0.0% 1.1% 1.9% 0.8% 0.9% -0.2% 1.7% 3

-0.9% 2.1% 1.9% 3.3% 4.0% 2.6% 1.9% 0.5% -2.2% -5.3% 6

2.5% 0.1% 1.7% 2.4% 1.0% 1.9% 3.3% 4.1% -0.5% 2.0% 2

Table 5: CARs across countries

Event window

Mean

Median (Kruskal-Wallis)

[-10;0] [-5;0] [-2;0] [- 1;0] [0] [0;1]

0.448 0.565 0.215 0.069 0.106

0.130 0.369 0.241 O. 111 0.434 0.286

[-1;1]

0.022

0.019

[-2;2] [-5;5] [-10;5]

0.065 0.214 0.114

0.088

0.029

Table 6: Country classification as explanatory variable for CARs

0.140 0.136

SP

SW

UK

0.9% -6.2% 1.4% 1.8% -4.8% 4.6% 2.1% -4.3% 5.4% 2.9% -3.5% 6.1% 1.8% 0 . 4 % 3.7% 2.3% -0.6% 1.4% 3.4% -4.4% 3.8% 2.3% -6.1% 0.0% 1.0% -8.8% -7.5% 0.1% -10.3% -10.7% 7 8 6

62

To link country differences to economic theory, a grouping of countries into similar groups is performed. Similarity is assessed by a combination of factors regarding details of the legal and financial system of a country. There are at least three different ways of forming groups. First, countries can be diviffed into civil law and common law countries. 243 Civil law "uses statutes and comprehensive codes as a primary means of ordering legal material, and relies heavily on legal scholars to ascertain and formulate its rules ''244 In common law countries, "law is formed by judges who have to resolve specific disputes. Precedents from judicial decisions, as opposed to contributions by scholars, shape common law ''245. Second, the civil law countries can be subdivided into German/Scandinavian-civil law and French-civil law countries, originating from different historic backgrounds and forming clusters with similar characteristics. 246 Third, a measure of shareholder rights can be used to form two groups of countries with low and high levels of shareholder rights, respectively. 247 The hypothesis is that ECOs in countries where shareholders have a stronger position, compared to company's management, will produce ECOs with higher abnormal returns because shareholders will be interested in doing an ECO only if it increases shareholder value. If shareholders have a better position vis-a-vis the parent firm's management, the latter's potential for any type of agent behaviour disadvantageous for shareholders is limited. Table 7 shows the p-values for all event windows of a variance and KruskalWallis analysis testing whether these regional classifications have explanatory power regarding mean and median abnormal returns. For the first classification (common law, German/Scandinavian civil law and French civil law), significance is rare and limited to a single event window for the mean and median. A similar result is obtained for the second classification (common vs. civil law). The third classification (shareholder rights) produces significant mean and median differences for the majority of the shorter event windows, with countries in the high shareholder rights group producing significantly higher positive abnormal returns than countries in the low shareholder rights group.

243 See La Porta/Lopez-de-Silanes/Shleifer/Vishny (2000), p. 14. 244 La Porta/Lopez-de-Silanes/Shleifer/Vishny (1998), p. 1118. 245 La Porta/Lopez-de-Silanes/Shleifer/Vishny (1998), p. 1119. 246 See La Porta/Lopez-de-Silanes/Shleifer/Vishny (1998), p. 1117-1119 for more details. The authors form two groups for German and Scandinavian civil law countries, but find both to have similar levels of shareholder protection rights. Thus, for this analysis the two groups have been put together. Combining these two groups produces similar results to not combining them. 247 See La Porta/Lopez-de-Silanes/Shleifer (1999), p. 478. Specifically, the index is formed by adding 1 when the country allows shareholders to mail their proxy vote to the firm; when shareholders are not required to deposit their shares prior to a General Shareholders Meeting; when cumulative voting or proportional representation of minorities in the board of directors is allowed; when an oppressed minorities mechanism is in place; when the minimum percentage of share capital that entitles a shareholder to call an Extraordinary Shareholders Meeting is less than or equal to ten percent; and when shareholders have pre-emptive rights that can only be waived by a shareholders vote. The index thus ranges from 0 to 6. Countries are assigned to the high shareholder rights group when their index is above the median index value of all countries.

63

Event window

Law(3) Mean

Law(3) Median

Law(2) Mean

Law(2) Median

Antidirec Mean

[-10;0] [-5;0] [-2;0] [-1;0] [0] [0;1]

0.306 0.517 0.579 0.241 0.472 0.284

0.211 0.641 0.481 0.330 0.380 0.284

0.536 0.267 0.318 0.153 0.392 0.133

0.534 0.351 0.308 0.764 1.000 0.357

0.805 0.119 0.106 0.021 0.088

[-1;1]

0.084

0.062

0.034

0.086

0.070 0.009

[-2;2] [-5;5] [-10;5]

0.711 0.995 0.445

0.523 0.969 0.494

0.455 0.916 0.866

0.274 0.872 0.847

0.143 0.810 0.661

Antidirec Median

0.941 0.086 0.079 0.117 0.579 0.243

Mean Mean Low High Antidirec Antidirec

1.5% 0.1% 0.5% 0.5% 1.0% 0.8%

2.2% 2.9% 2.4% 3.0% 2.4% 2.5%

0.026

0.3%

3.1%

0.053 1.000 0.513

-0.2% - 1.0% 0.5%

1.6% -0.4% -1.1%

Notes: Law(3) refers to the regional classification scheme grouping countries into three groups (Anglo-Saxon common law, German/Scandinavian-civil law, or French-civil law). Law(2) refers to the regional classification scheme grouping counlries into two groups (Civil law group or AngloSaxon common law group). Antidirec refers to the regional classification scheme grouping countries into two groups (low vs. high shareholder rights). See text for details. Table 7: Regional classification as explanatory variable for CARs

Thus, the measure for shareholder rights is a significant explanatory variable in the cross-regional distribution of abnormal returns. The more powerful shareholders are relative to the parent firm's management, the higher the abnormal returns produced by an ECO announcement. There are two potential explanations. First, assuming that there exist scenarios in which a company's management may be interested in disposing of a subsidiary firm for opportunistic reasons, the result may indicate that only ECOs which are beneficial to shareholders, rather than to management only, are carried out. Second, regions where shareholder rights are relatively more developed may also be regions where external capital markets are well developed. Consequently, if external and internal capital markets are substitutes, and internal capital markets are seen as inefficient on average, the relative value of an internal capital market in regions with high shareholder rights (and well-developed external capital markets) is lower compared to regions with low shareholder rights (and less developed external capital markets). The opportunity costs associated with an internal capital market are therefore higher, and its partial closure through an ECO will be of more value in high shareholder rights regions. Consequently the ECO announcement will produce higher abnormal returns. Grouping countries into clusters with similar legal and financial systems may also help to assess the significance of any potential self-selection bias. Companies could choose to engage in an ECO as a function of their home country's legal and financial system. If this were the case, inferences drawn from this study could not be applicable to ECOs in general. The specific question therefore is: Is the frequency of ECOs related to the legal and financial systems of the home countries of the respective parent firms? If ECOs are associated with an increase in shareholder value (as indicated by the witnessed abnormally positive average announcement period returns), then the hypothesis is that

64

ECOs occur more frequently in countries with higher shareholder rights. To test the hypothesis, a two groups difference of means test is conducted, with the two groups consisting of high and low shareholder rights countries, respectively. The dependent variable is the relative frequency of ECOs in each country, measured as the total number of ECOs divided by the total number of all IPOs. While the average percentage is higher in countries with high shareholder rights compared to countries with low shareholder rights (4.6% vs. 3.9%), the difference is not statistically significant (p=0.6779). 248 The null hypothesis of no influence of the level of shareholder rights on the relative frequency of ECOs thus cannot be rejected. This result also implies that there is no indication of a sample self-selection bias by parent firms arising from the country group affiliation. Motivation: Companies announcing an ECO often state a reason for this decision. The nature of this reason may have an impact on the level of abnormal returns. Newslines in LexisNexis and Factiva are searched to determine this motivation. Any classification of motivations has an arbitrary element. A first search produces the following eleven motivational categories: Focus increase of parent firm, focus increase of subsidiary firm, repayment of debt/improvement of equity ratio, development of parent firm's business/acquisitions, development of subsidiary's business/acquisitions, increase of

operating flexibility of subsidiary firm and ability of subsidiary firm to partner, better access to capital markets/increase of identity, value increase for shareholders, employee participation, exit business, and other 249. Multiple indications are allowed. For the following analysis, only motivations of companies included in the sample (n=100) are considered. For 17 of these no motivation could be identified.

Motivation Focus increase parent Focus increase sub Repay debt/improve equity ratio Develop business/acquisitions parent Develop business/acquisitions sub Investor visibility

Regression coefficient

p-value

0.020 0.016 -0.005 -0.006 0.026 -0.009

0.065 0.374 0.658 0.703 0.007 0.440

Table 8: Stated motivation as explanatory variable for C A R s

Categories mentioned by less than 5% of companies are excluded (increase of operating flexibility of subsidiary firm, employee participation, exit business). The 'other' category is also excluded. To reduce multicollinearity, cross-correlations of the 248

Needless to say, insignificance may be driven by the small sample size (i.e., the 13 countries in the study). 249 The 'other' category includes the required disposal of a subsidiary for regulatory reasons (Oerlikon Btihrle/Pfeiffer Vacuum, ENI/Snam Rete Gas, United Pan-European/Priority Telecom, Gas Natural/Enagas), the listing of a subsidiary immediately following an acquisition (Fresenius/FMC), the use of proceeds for reacquiring shares used previously for an acquisition (France Telecom/Orange), and the listing of a subsidiary to shield the parent from impact of US oil pollution legislation (LeifHoegh/Bona Shipholding).

65

remaining variables are calculated. The aim is to identify positively cross-correlated variables and to recode them into a common variable. The only significant (at the 5% level) correlation is between 'better access to capital markets/increase of identity' and 'value increase'. These two motivations can be argued to have a common source, which can be labelled 'increase in investor visibility'. The two variables are hence combined into a single variable. 2s~ Following this recoding, there are no significant correlations among the six remaining variables. The [0] day abnormal returns are regressed on the six variables. Table 8 shows the coefficients and associated p-values. The coefficients on 'focus increase of parent firm' and 'development of subsidiary's business' are both positive and significant (at least at the 10% and 1% level, respectively). A natural question is whether the two sub-groups (companies which give either or both of the above reasons, and companies which give neither reason) differ in their abnormal returns. Table 9 shows the mean and median abnormal returns and associated p-values for the two subgroups, as well as p-values for the equality test of the means and medians of the two samples. Companies announcing either of the two motivations earn significantly positive abnormal returns (2.0%), while the remaining companies earn no abnormal returns (-0.1%). Consequently, for the purpose of the multivariate analysis, a dummy variable is created equalling 1 when either of the two motivations is stated in the announcement, and 0 otherwise. Vijh (2002) develops the "investment strategy hypothesis ''2s~, which is similar to the 'development of subsidiary's business' motivation: Markets view the retention of proceeds within the subsidiary firm to finance new projects as positive. Consistently, Mikkelson/Partch (1986) find that excess returns of SEOs are insignificantly different from zero when the company announces that proceeds will be used to finance capital expenditures, whereas otherwise they are significantly negative. McConnell/Muscarella (1985) show that announcements of increases in capital expenditures earn positive abnormal returns.

Statistical values

# of companies Mean p-value (mean=0) Median p-value (median=0) p-value (mean difference=0) p-value (median difference=0)

Without announcement

With announcement

27 -0.1% 0.621 -0.4% 0.701 0.015 0.008

55 2.0% 0.001 1.1% 0.000 NaN NaN

Table 9: Difference between companies with and without announcement 250

251

I.e., all cases where at least one of these variables has been mentioned are combined into a single variable. Vijh (2002), p.155.

66

Post-event stake: Boone/Haushalter/Mikkelson (2003) find that operating performance of the parent firm is negatively related to the stake retained in the subsidiary firm. 252 This implies that abnormal retums should also decrease in the stake retained by the parent firm. Alternatively, a positive relationship between abnormal retums and postevent ownership could also be construed, and interpreted as markets preferring the parent firm to retain a larger stake to exercise a monitoring function, or to retain synergies. Combining these two ideas implies the possibility of a quadratic, rather than a linear, relationship between post-event stake and abnormal retums. Figure 3 plots abnormal retums in the [0,+1] day window vs. post-event stakes and yields graphic support for this notion. A test of the mean difference in abnormal retums between companies in the two extreme quartiles and companies in the two middle quartiles shows a significantly positive difference (2.4% vs. 0.0%) in abnormal returns. Consequently, a dummy variable equalling 1 when the parent firm's post-event stake is in either of the extreme quartiles, and 0 otherwise, is constructed.

Figure 3: Abnormal returns as a function of the post-event stake

Profitability: Profitability of the parent firm, if indicative of management efficiency, could impact abnormal returns in two different ways: More efficient management can be assumed to carry out more efficient ECOs, and the market reaction should thus

252

See Boone/Haushalter/Mikkelson (2003), p. 74.

67

increase in the parent firm's profitability. 253 Alternatively, a less efficient parent firm engaging in an ECO can be considered to create relatively more value, as the subsidiary's assets will be brought under a more efficient management after the E G O . 254 The market reaction should thus decrease in the parent firm's profitability. Equivalently, the value creation potential of high profitability companies may be doubted by markets ('Why change a winning team' hypothesis). Profitability is measured as return on assets (EBIT/total assets) and as return on sales (EBIT/sales).

3.5.3.2

Control variables

Industry: Industry affiliation could be relevant in at least two ways. First, companies in different industries are subject to different competitive forces and could therefore systematically differ in abnormal returns. 255 Second, abnormal returns could be driven by whether parent and subsidiary firms belong to the same or different industries. Table 10 shows the mean abnormal returns per first digit SIC code group, which is significantly positive for two of out of the eight groups. 256 A variance analysis is used to check whether the first digit of the SIC code helps explain the difference in abnormal returns. The null hypothesis of no explanatory power cannot be rejected (p=0.8312)

First digit SIC code

Abnormal return

No. of companies

p-value

2.31% 2.15% 2.30% 1.57% 0.56% 1.01% 0.58% 2.83%

1 6 12 32 24 10 11 3

NaN 0.137 0.027 0.018 0.158 0.127 0.281 0.290

0 1 2 3 4 5 7 8

Table 10: Industry classification as explanatory variable for CARs

Abnormal returns have also been linked to an increase in corporate focus, which assumedly is stronger if parent and subsidiary firms come from different industries. 257 lndustry is defined as all companies sharing the same first two digits SIC code. 2s8 An industry dummy variable is created, equalling 1 when parent and subsidiary firms are in the same industry, and 0 otherwise. 259 The dummy is expected to have a negative 253 254 255 256 257 258 259

See L6ffler (2001), p. 169-170. This is based on Kaserer/Ahlers' (2000) conjecture that ECOs are optimal if external capital markets exercise a better control over the subsidiary's assets than internal capital markets. See Stienemann (2003), p. 174-178. There is no group '6' because these are financial firms excluded from the analysis. See Vijh (2002), p. 160-161. See Vijh (2002), p. 187. See Vijh (2002), p. 187.

68

coefficient, since negative synergies are less likely in same-industry conglomerates, whose partial break-up through the ECO therefore is expected to produce less value relative to the partial break-up of a cross-industry conglomerate.

Informational asymmetry /opaqueness: Based on Krishnaswami/Subramaniam (1999), informational asymmetry is proxied by the standard deviation of the market model residuals. 26~ If the resolution of informational asymmetry is a source of value gains in ECO, the level of pre-event informational asymmetry should be positively related to abnormal returns. Similar to Krishnaswami/Subramaniam (1999) and Elsas/L6ffler (2005), the interaction term between the industry dummy and the measure of informational asymmetry is also included as a control variable. This allows differentiating the impact of opaqueness on same- and cross-industry ECOs.

Number of business segments: Another way of capturing informational asymmetry is through the number of business segments. 261 A higher number of business segments reported before the ECO may render an adequate valuation of the company more difficult for investors. A partial break-up of such a company is expected to create value through resolution of this complexity. The number of business segments should therefore be positively associated with the level of abnormal returns.

Relative size: If the market reaction to an ECO announcement is positive on average, abnormal returns are likely to increase with the relative size of the transaction. It seems sensible to assume that the importance of a subsidiary to the parent firm and its visibility to the market increases with the relative size of the subsidiary to the parent firm, and that the significance of a transaction is related to the percentage stake sold. Relative transaction size is thus proxied by the ratio of subsidiary to parent firm market capitalization at the time of the IPO, multiplied by the percentage stake of the subsidiary sold by the parent firm.

Financial distress: Companies selling assets or carving out a subsidiary firm have been found to have higher debt ratios than their industry peers. 262 Assuming managers value size, they will only engage in perimeter-decreasing financing activities if other less costly sources of funding are unavailable. This implies that financially constrained parent firms will have lower abnormal returns than financially flexible parent firms: While the former are considered by market participants to be forced to carry out the ECO for financing purposes, the latter are considered to choose the ECO voluntarily, potentially for value-enhancing purposes. The level of financial constraint is proxied by a coverage ratio, defined as EBIT/Interest. 263

260

See Krishnaswami/Subramaniam (1999), p. 86. Using analyst forecast dispersion is not possible due to limited data availability. 261 Vijh (1999), p. 298, uses this variable in his explanation of the long-termprice performance of ECOs. 262 See Langa/Poulsen/Stulz(1995) for asset sales and Allen/McConnell (1998) for ECOs. 263 This ratio is also used by Asquith/Gertner/Scharfstein(1994), p. 628.

69

Primary vs. secondary offering: Kaserer/Ahlers (2000) find for their sample that parent firms performing an ECO as a primary offering show higher abnormal returns compared to companies designing the ECO as a secondary offering. They interpret their finding as evidence for markets preferring the parent firm to not retain direct control of the divestiture proceeds. Since many ECOs are carried out as a mixture of a primary and a secondary offering, a variable indicating the percentage of shares sold by the parent firm is constructed. For the sample in the present study, 48 (27.0%) firms perform the ECO as a primary offering, 50 (28.1%) firms as a secondary offering, 54 (30.3%) perform a mixed offering, and for the remaining 26 (14.6%) firms the variable could not be determined, mostly because the IPO prospectus was not available.

3.5.4

Models

The model is estimated using OLS regression and the dependent variables described above. 264 Various models are estimated, based on the inclusion of certain variables only, as shown in Table 11. 265 All models produce control variable coefficients of the expected signs (with the exception of the sign for the % of secondary offering). The control variable for industry is negative, indicating that ECOs by cross-industry parent firms produce higher positive abnormal returns. The control variable for opaqueness is negative. Since the interaction term is also included, the opaqueness variable measures the effect of opaqueness for cross-industry ECOs. The negative sign implies that opaqueness decreases the positive impact of increased industry focus. The coefficient of the interaction term is positive, implying that when parent and subsidiary firms are from the same industry, an increased level of pre-event informational asymmetry (and therefore a higher potential for its partial resolution) produces positive value effects. These results are consistent with findings in previous literature. 266

264

265

266

In addition to the OLS estimator, Karafiath (1994) tests three more refined estimators (consistent estimator least squares, weighted least squares, feasible generalized least squares), and finds that with sample sizes exceeding 75 these offer no benefit to OLS, even in the case of extreme calendar time clustering, i.e., shared event dates. Since the sample size in the present study exceeds 75, OLS (rather than one of the alternatives) is employed. Model I includes all variables discussed above. Model III has removed all variables whose significance is below 10%. Model II is the model producing the highest adjusted R2 value in this process of stepwise removal of variables. Model IV to VI exclude the financial distress and the period dummy, which are significantly correlated with the opaqueness measure. Model V and VI remove certain variables with less than 10% significance. Stepwise removal leads to a model with no variables, hence removal is stopped before. See Krishnaswami/Subramaniam (1999), p. 102-109, and Elsas/L6ffler (2005), p. 15.

70

Regressor

Expected sign

Constant Opaqueness

1-]

Industry

[-]

Industry*Opaqueness

[+]

Number of segments

[+|

Relative size

[+]

Financial constraint

[-]

% secondary offering

[-]

Period dummy

[+]

Region dummy

[+]

Motivation dummy

[+]

Post event dummy

[+]

ROA

[+/-]

N Adj. R2 p-value F-test

Model I

Model II

Model III

Model IV

Model V

Model VI

-0.013 0.400 -0.887

0.002 0.834 -0.914

-0.003 0.692

-0.011 0.501 -0.562 0.130 -0.035 0.183 1.719 0.104 0.002 0.119 0.015 0.102

-0.018 0.154

-0.014 0.255

0.717 0.257 0.002

0.741 0.233 0.002

0.055

0.072

0.076

0.038

-0.042 0.110 1.915 0.078

-0.036 0.178 1.757 0.110

0.002 0.250 0.019

0.016

0.011

0.037

0.076

0.060

0.009 0.687 0.002 0.909 0.019

0.022

0.021

0.101

0.049

0.029

0.010 0.274

0.003 0.822

0.026

0.026

0.031

0.027

0.031

0.058

0.058

0.087

0.054

0.045

0.031 0.048

0.022

0.020

0.022

0.017

0.022

0.051

0.023

0.080

0.015 0.115

0.012 0.318 -0.068

-0.082

-0.065

0.012 0.291 -0.068

-0.065

-0.066

0.074

0.013

0.012

0.060

0.041

0.042

67 10.4% 0.109

67 13.7% 0.032

67 7.3% 0.068

67 11.2% 0.077

67 12.9% 0.025

67 13.6% 0.016

Table 11: C r o s s - s e c t i o n a l regression of C A R s

The coefficient for relative size is significantly positive, indicating that larger E C O s produce more positive abnormal results. The coefficient for n u m b e r o f segments is positive, implying that the larger the conglomerate, the more it stands to profit from an ECO because o f the removal o f negative synergies. The coefficient for the percentage o f secondary offering is marginally positive, with a very high p-value, indicating that this variable has almost zero explanatory power. This result does not support Kaserer/Ahlers (2000), w h o find that primary offerings earn higher abnormal returns than secondary offerings 267, but is consistent with W a g n e r (2004), who also finds no relationship between abnormal returns and either primary or secondary offering. The variables o f prime interest produce interesting results. First, the period d u m m y has a statistically significant positive coefficient. The results from the univariate analysis thus carry over to the multivariate analysis, and indicate that E C O s carried out after 1998 consistently earn higher abnormal returns than E C O s from before that year. This supports the notion o f external capital markets b e c o m i n g relatively more efficient, compared to internal capital markets; the opportunity costs associated with an internal

267

A potential explanation for this discrepancy may be their relatively small sample size (n=23). Their paper also uses a fairly long [-500;0] days event window.

71

capital market thus increase over time, and its partial closure through an ECO is rewarded more highly by investors. Second, the region dummy has a statistically significant positive coefficient. The results from the univariate analysis again carry over to the multivariate analysis, and indicate that ECOs carried out in countries with high shareholder rights on average earn more positive abnormal returns than ECOs in countries with low shareholder rights. As pointed out above, if higher shareholder rights are associated with better-developed external capital markets, the interpretation may be similar to the one offered for the period dummy: Opportunity costs of an internal capital market are higher when external capital markets are well-developed, and the partial closure of the former will be more highly rewarded by investors than in the case of less well-developed external capital markets. The only comparable result in the literature is from Bt~hner (2004), who develops a similar argument for an increasing abnormal return trend across time but fails to find empirical evidence for it in his sample, potentially due to his limited sample size. Third, the motivation dummy has a statistically significant positive coefficient. Companies stating that they carry out the ECO to either increase the focus of the parent firm or to develop the subsidiary's business, earn consistently higher abnormal returns than companies stating a different motivation. This result synthesises results by Langenbach (2001), who finds that if companies state disinvestments as a motivation for the ECO, abnormal returns tend to higher; and by Vijh (2002), who finds that if a company announces that proceeds are invested into new or existing subsidiary projects, abnormal returns tend to be higher ("Investment strategy hypothesis"Z6S). Another result by Vijh (2002), namely that giving financing reasons as a motivation for ECOs lead to higher abnormal returns ("Financing strategy hypothesis"269), cannot be confirmed: Announcing that the parent firm will use proceeds to repay debt does not seem to be a valuable proposition to investors. Fourth, the post-event dummy has the expected positive sign, but is not statistically significant. The results from the univariate analysis thus cannot be confirmed. While intuitively appealing, the idea of a parent firm having to either sell the complete stake, or retain a considerable majority in order to produce a positive market reaction, cannot be upheld. Fifth, the coefficient on parent firm profitability is significantly negative. 27~ This supports the 'Why change a winning team hypothesis': Investors seem wary of wellmanaged companies carving out a subsidiary firm, and doubt the potential for considerable improvements following the ECO. Equivalently, companies which are badly managed (as evidenced by a low profitability) are likely to profit more from a 268 Vijh (2002), p.155. 269 Vijh (2002), p.155. 270 The analysis uses sales margin (EBIT/sales) as a measure of profitability. Using asset margin (EBIT/total assets) produces similar results in sign, but of lower or no statistical significance.

72

separation of ownership in the subsidiary firm, and thus produce higher abnormal returns when announcing an ECO. This result is consistent with Elsas/L6ffler's (2005) explanation of their finding that abnormal returns are lower when pre-event ownership concentration is high: They argue that firms which are better monitored prior to the ECO stand to benefit less from a separation of parent and subsidiary firm. Analogously, the finding that abnormal returns are lower when pre-event profitability is high can be interpreted as showing that firms which are more pl'ofitable prior to the ECO also benefit relatively less from an ECO.

3.6

3.6.1

Various extensions

Assumption on marketparameter stability

Applying market parameters estimated via OLS from estimation period data to event period data is based on the assumption that the parameters remain constant across the two periods. The validity of this assumption for this study is assessed in two ways: First, betas calculated from the pre-event estimation period are compared to betas calculated from a post-event period, which is defined as beginning ten days after the identified announcement date.

Figure 4: Difference between pre- and post-event beta

73

Figure 4 shows the percentage point change in beta as a function of the cumulated number of companies showing an equal or lower percentage point change than a given level. Roughly a quarter of the sample companies show changes of 40% or more in their beta estimates. Second, to assess statistical significance, Brown/Lockwood/Lummer's (1985) test for structural change in the market parameters is implemented. TM The test statistic F is distributed with (k, 7"1+ T2-2k) degrees of freedom:

F=(1/k)*

I2

(Pes,

S est

nt_

*(t~es, --~event)' where

2 1

S event

X;wn,X v o,

number of parameters, here k = 2 ~est

coefficient from OLS, based on estimation period data

fl~v~n, coefficient from OLS, based on event period data Are,,

column vector with independent variable in estimation period

Xeven, column vector with independent variable in event period According to this test, 10.8% (4.9%) of the companies show a significant (at the 5% level) change in their beta (alpha) parameter. Brown/Lockwood/Lummer (1985) suggest using a switching regression technique determining firm-specific event and estimation periods. As an alternative, and in accordance with the recommendations by Peterson (1989), market parameters are also estimated using both pre- and post-event estimation periods. Similar to Dann/Mikkelson (1984), the estimation period is defined as [230;+190] days around the event, excluding the 50 days prior to and ten days following the event date. Appendix 17 shows all major results using this alternative estimation period. Two points are noteworthy: First, the differences in abnormal returns in shorter event windows are not material, and the abnormal returns remain significant at previous levels. Second, the difference between pre- and post-event estimation period and preevent estimation period only abnormal returns is positive, and increases with the length of the event window. This seems counterintuitive at first: One could have assumed that when using pre-and post-event period estimation period data, an increased volatility of the sample companies following the event leads to higher beta estimates, higher expected returns and thus to lower abnormal returns. However, this assumes that the alphas remain constant. To analyse this, regressions are re-run using only pre- and postevent period data, respectively. The average difference between post-event and pre271 According to Brown/Lockwood/Lummer(1985), the most common procedure to analyse a structural

change in the coefficients of a linear model is the test by Chow (1960). However, they point out that in the presence of heteroscedasticitythis test may be biased, see Brown/Lockwood/Lummer(1985), p. 318-320. Their test takes potential heteroscedasticityinto account.

74

event period beta is +0.1030, significant at the 5% level, whereas the average difference between post-event and pre-event alpha is -0.0016, significant at the 1% level. The fact that abnormal returns based on pre- and post-event estimation period data are higher than abnormal returns based only on pre-event estimation period data therefore implies that the negative impact of the lower alphas outweighs the positive impact of the higher betas; this leads to lower expected returns and thus higher abnormal returns. Using only pre-event estimation period data may therefore bias abnormal return calculations. The magnitude of the bias seems economically insignificant for shorter event periods but increases in event period length. This result implies that event studies employing longer event windows need to test the robustness of their results to alternative specifications (i.e., only pre- and combined pre- and post-event period related) of the market model parameters.

3.6.2

Multiple dates

In addition to the difficulty of identifying the announcement date as discussed in section 3.4.4, there are two other issues regarding the date: First, the announcement date is not the only date at which the market learns about the ECO, and thus other dates could be of interest. Second, ECO announcements do not always occur in isolation and can thus be accompanied ('contaminated') by other announcements. These two issues are discussed in the following sections.

3.6. 2.1

Three additional dates of interest

The choice of announcement dates for the event study has an arbitrary element to it. As L6ffier (2001) points out, there are at least four dates of interest in the context of an ECO, which are: First rumour date: Date where either the company mentions for the first time that a (partial) disposal of a particular subsidiary firm is being contemplated, without a clear specification of either disposal method, date, or level of commitment; or where sources other than the company speculate about the potential disposal of a subsidiary firm, without the company confirming these rumours. Announcement date: Date when the company announces that it plans to carve out a particular subsidiary firm, including the specification of an approximate date in the foreseeable future (usually within the next 12 months). This is the 'classical' announcement date also used in the event study in section 3.5.1. Bookbuilding date: Date where company provides market with additional details on the IPO, including at least the price range for the bookbuilding, and often additional information on the use of the proceeds and the stake to be sold. In many cases this day immediately precedes the first day of the bookbuilding phase.

75

IPO date: First day of trading of the subsidiary firm, when market participants learn about the market valuation views regarding the subsidiary firm. These dates are determined for each of the sample firm, based on newslines searches in LexisNexis and Factiva. Each date provides the market with additional information regarding the subsidiary firm, possibly leading to abnormal price reactions. Consequently a separate event study using each of the three dates (in addition to the original event study using the announcement date) is performed. The first rumour date could be identified for 25, the announcement date for 176, the bookbuilding date for 136 and the IPO date for 176 companies. The economic hypothesis is that abnormal returns will be highest the first time the market learns about a possibly impending ECO, as the market has not had a chance to react to this possibility before. 272 The news on the bookbuilding date can be either positive or negative: For example, the bookbuilding price range usually indicated on this date can be seen by the market as either too low or too high. On average, the share price impact is likely to be distributed with a mean of zero. Abnormal returns on this date should hence be close to zero. The abnormal return on the IPO date is undetermined and could be either positive or negative. The obvious question is how to estimate market parameters for all four events. There are at least two alternatives. First, each of the four classes of dates can be treated separately, so market parameters are calculated from estimation periods with relative start and end dates as described above. This can be called the 'variable method' because the value of the market parameters of each company will differ for each date. Second, one common estimation period preceding all four dates for each company can be designated, and market parameters calculated from this estimation period. This can be called the 'constant method' because the same value for the market parameters is used for all classes of dates for a specific company. The advantage of the 'variable' method is that it takes into account changing volatilities as the process unfolds. As described above, volatilities tend to increase from the period preceding to the period following the announcement date. The disadvantage of the 'variable' method is that the estimation period of subsequent dates is impacted by preceding dates. For example, a company whose share price rises following rumours regarding an impending ECO, but which only announces this ECO at a later stage, will have upward-biased market parameters because the impact of the rumour date announcement lies within the estimation period of the announcement date. Upward biased market parameters (in rising markets) decrease the power of tests to find positive

272

An exception to this, of course, is insider trading. However if this trading leads to unexpected share price movements, this is often mentioned in the press and, if linked to an ECO, mentioned as being the result of speculation regarding such a step. Therefore only that amount of insider trading not causing perceptible share price movements is not considered by the analysis, the effect of which by definition must be limited.

76

abnormal returns on the announcement date. The advantages and disadvantages of the 'constant' method mirror those of the 'variable' method. Consequently, both alternatives are implemented. From the above, it follows that the 'constant' method should produce higher abnormal returns compared to the 'variable' method, because market parameters and thus (assuming rising markets) expected returns of the 'constant' method will be downward biased, and market parameters and expected returns of the 'variable' method will be upward biased. Appendix 18 shows average abnormal returns and associated p-values for all event windows. As hypothesised, the abnormal return is highest for the rumour date. The abnormal returns on the announcement date under the variable method are, of course, those reported previously, since the methodology used in this section is the 'standard' event study methodology used in section 3.5.1. The abnormal returns on the bookbuilding date are close to zero (as hypothesised) or slightly negative. This implies that positive and negative news regarding the ECO are close to normally distributed across all transactions. The abnormal return on the IPO date is significantly negative for almost all event windows. This result is consistent with Mathesius (2003), who finds a significantly negative abnormal return on the IPO date for a sample of 12 German ECOs in the 1998 to 2000 period. 273 The finding suggests that markets on average are negatively surprised by the valuation the market places on the subsidiary firm. All results are similar for both the 'variable' and the 'constant' method. The abnormal returns under the 'constant' method are mostly higher than under the 'variable' method, which confirms the considerations on the impact of rising volatilities and subsequent opposite sign biases in market parameters in the two methods. Two natural extensions concern the positive abnormal returns on the rumour date, and the negative abnormal returns on the IPO date. First, one could assume that rumour date abnormal returns partially anticipate and dilute subsequent announcement date returns. To analyse this, announcement date abnormal returns for the subgroup of companies for which rumour dates could be identified are compared against announcement date abnormal returns for the subgroup of companies for which rumour dates could not be identified. The analysis produces no significant difference in means and medians. The reason for this result, however, may be the small number of companies (n=17) for which a rumour date is available and could be identified. Second, the negative return for the parent firm around the IPO date could be linked to the level of underpricing for the subsidiary IPO. In analogy to existing literature, underpricing is measured as the percentage return from the offer price to the closing price on the first day of trading. TM There are two effects: First, underpricing represents a loss of value to the parent firm because an asset has been sold for less than its intrinsic value (assuming first day closing prices are indicative of the true asset value). Hence,

273 See Mathesius (2003), p. 140. 274 See Cliff/Denis (2004), p. 2878.

77

higher underpricing is likely to lead to more negative returns for the parent firm. Second, higher underpricing could also lead to a positive first day performance, because a stronger first day performance raises the value of the stake retained by the parent firm, hence leading to less negative returns. To analyse which effect dominates in the tradeoff, the abnormal return of the parent firm on the IPO date is regressed on the first day performance of the subsidiary firm (i.e., the measure of underpricing). The coefficient is positive and significant at the 1% level. The null hypothesis of no relationship between IPO date abnormal return for the parent firm and first day performance of the subsidiary firm is thus rejected. The result implies that the value increase of the parent firm's remaining stake resulting from a positive first day performance outweighs the value loss to the parent firm resulting from the underpricing. Two caveats regarding the analysis of abnormal returns on the four different dates are in order. First, the relatively small number of companies for which rumour dates could be identified could be either the result of other companies not being subject to pre-event rumours, or of these rumours not showing up in the databases used for the analysis. A plausible reason for the latter alternative is the uncertainty regarding rumours, lying in their nature, which may prevent them from being reported by the press. Second, while the main results reported are robust to using medians rather than means, significance is lower for many event windows, implying the impact of outliers. To summarise, the analysis of four separate dates on which the market learns new information regarding the ECO produces results largely in line with intuition and economic theory. Markets react positively when they first learn about a possible ECO, and the reaction is stronger for earlier announcements of a given company. Later announcements on average are seen neutrally to negatively. A positive first day performance of the subsidiary firm is associated with less negative returns to the parent firm on the IPO date. The analysis also implies that when conducting an event study it is not only important to identify a single 'right' date, as repeatedly pointed out in the literature, but also to consider the existence of more than one relevant date; the nature of information conveyed on those dates may differ systematically.

3.6.2.2

Contamination of announcement dates

ECO announcements often occur not in isolation, but as part of a wider anl~ouncement e.g., interim or annual results. Event studies not controlling for this implicitly assume that the average impact of such contaminating announcements is zero. This is a strong assumption, and it seems plausible that parent firms may attempt to time the announcement to coincide with other favourable news to positively reinforce the effect of the combined announcement. To test this hypothesis, the abnormal returns analysis is re-run for all cases previously excluded from the analysis due to contaminated announcement dates (n--38). Contamination is defined as the contemporaneous announcement of other significant

78

news likely to move share prices, including announcements regarding interim and annual results of the parent firm, major revisions of the parent firm's forecasts, additional major restructurings, a forced sale of the subsidiary firm, acquisitions, alliances, bond issuances, reacquisitions of the outstanding minority of another subsidiary firm, price increases for a company's products and the delayed flotation of another subsidiary firm. Appendix 19 shows the resulting CARs for the event windows. The difference compared to the non-contaminated sample is always positive, and significant at least at the 10% level for six of the ten event windows. The implication is twofold: First, there is some evidence that companies time the announcement of ECOs to coincide with other positive news. Second, event studies must control for announcement date contamination, as this may significantly change the level of calculated abnormal returns.

3.6.3

Impact on non-announcingfirms 275

The announcement of an ECO by one parent firm could have an impact not only on the announcing parent firm, but also on other companies, either in the same or in a different industry, owning potential ECO subsidiaries. Two potential analyses are thinkable: First, benchmark companies with characteristics similar to the sample companies could be identified, and their share price reaction on the days of the announcement by the sample companies could be analysed. Second, the share price reaction of companies within the sample to prior announcements by other companies within the sample could be analysed. The advantage of the latter approach, besides easier implementability, is that only companies actually having carried out an ECO are included in the analysis, making it unnecessary to find and apply contestable matching criteria. The direction of the abnormal reaction is undetermined by theory: Companies could react positively to another company's announcement because market participants anticipate a similar announcement from those companies. Companies could react negatively because market participants punish a company for not (yet) having issued a similar announcement. For each company in the sample, the share price reaction of a subgroup of other companies on the day of the ECO announcement by this company is analysed based on the methodology described in the preceding sections. Subgroups are formed in five different ways, based on subsidiary SIC codes, and comprising all companies with at least one, two, three or four different last digits, respectively, as well as without consideration of the SIC code. The required steps within each subgroup are repeated for each company in the subgroup so that each company is compared to all companies in the subgroup with announcement dates later than this company. This yields a maximum number of average abnormal returns of R (=number of separate event studies), with

275

I thank Prof. Dr. Schiereck for suggesting this analysis.

79

S

R:Z(,-,,) s=l

where n is the available sample size, S is the number of subgroups (=5), and N~. is the number of companies in subgroup s. The actual number of average abnormal returns is lower because not all companies within the subgroup are listed at the time of the announcement of a given company in that subgroup. Significance is assessed in three ways. First, subgroup grand averages are calculated and their significance is assessed. Table 12 shows for each subgroup the grand average of the average abnormal returns. Difference in last .... SIC code digits

Mean

Median

0 digit (=same SIC code)

0.67%

-0.15%

p-value

No. of observations

0.3171

23

1 digit

0.51%

-0.03%

0.2666

36

2 digits

0.17%

-0.03%

0.5001

61

3 digits

-0.11%

-0.05%

0.4820

90

0.16%

0.03%

0.0517

99

4 digits (=different SIC code)

Table 12: Average reaction of non-announcing companies: Grand subgroup averages

None of the industry-based subgroups show significant abnormal returns. The subgroup comprising all companies irrespective of industry has a slightly positive abnormal return, which is significant at the 10% level, implying a positive assessment by market participants of a possible future similar announcement. Second, since grand averages may be close to 0 as a result of both positive and negative average abnormal returns within the subgroups, each average abnormal return in each subgroup is tested for significance. Then the distribution of p-values in each subgroup is compared to the expected distribution if on average there was no average abnormal return. Appendix 20 shows the percentage of p-values below 1%, 5% and 10%, respectively, for each subgroup, where p-values are based on three different tests (2sided, 1-sided positive, 1-sided negative) to accommodate the lack of theoretical grounding for a directional hypothesis. The assumption is that if there was no abnormal performance, the distribution of pvalues in each subgroup would be close to its expected theoretical distribution, which implies that the percentage of p-values below 1%, 5% and 10% should be close to 1%, 5% and 10%, respectively. The results show that these percentages are higher, which implies that companies tend to show an abnormal share price reaction to the ECO announcement by another company. The direction of this reaction is assessed via the pvalues based on the two 1-sided tests (for positive and negative abnormal returns, respectively). For each of these, empirical rejection levels again considerably exceed theoretical rejection levels. This implies that both more average positive and more

80

average negative reactions tend to occur than in the absence of an announcement. Also, the percentage of p-values tend to be more in excess of their theoretical levels for the case of same-industry firms (as measured by the 2-digit SIC code) than for the case including all firms. This indicates that parent firms with subsidiaries in the same industry as the subsidiary of the announcing parent firm tend to react more strongly than parent firms with subsidiaries in other industries. Third, considering the multiple test nature of the analysis, a Bonferoni multiple test procedure is also applied. 276 The basic idea is to adjust the required level of significance to reflect the fact that multiple tests are conducted to test the same hypothesis. Specifically, the alpha level is divided by the number of tests used. At least one of the individual tests must produce a p-value lower than this adjusted alpha level for the null hypothesis to be rejected. Table 13 shows the number of individual tests that produce pvalues lower than the adjusted alpha for a 1%/5%/10%-probability to falsely reject the null hypothesis. Tests

1%

5%

10%

2-sided

6

8

9

1-sidedplus

3

4

5

1-sidedminus

3

5

5

Table 13: Number of p-values in excess of Bonferoni-adjusted alpha level

The null hypothesis of no reaction of similar companies is rejected at the (Bonferoniadjusted) 1% level in 6 out of 255 tests. Positive and negative abnormal returns seem to occur at about the same rate. To summarise, there is some evidence that the announcement of an ECO by one company impacts the share price of other companies owning potential ECO subsidiaries. The reaction is more pronounced for parent firms owning subsidiaries in the same industry as the announcing parent firm's subsidiary. The direction of the impact is less clear, however, and, while slightly positive on average, the direction of abnormal returns seems to depend on the specific situation.

3.7

Conclusion

This study documents a series of extensions to the body of knowledge regarding ECOs. First, abnormal announcement period retums are higher when they occur after 1998 and in countries with higher shareholder rights. These findings support the notion of the

276

See Bortz (2005), p. 129-130. Apparently there are no direct or indirect sources as to the nature of Mr./Mrs. Bonferoni, but the procedure has come to be thus called.

81

changing relative value of internal vs. external capital markets277: The better developed the latter, the higher the premium awarded to the announcing company for partially closing the former. 278 Abnormal returns are also higher if a company states either refocusing of the parent's or development of the subsidiary's business as a motivation for the ECO. Abnormal returns are negative in the two-day window following the initial announcement, reverting some of the gains from the announcement period. A crosssectional analysis reveals that this is mostly due to negative returns for ECOs from the 'hot market' period of 1998 to 2000, where a high level of initial enthusiasm about an ECO was often followed by a phase of disenchantment. Second, there is a pattern in share price returns across the three additional dates on which markets learn about the impending ECO. Abnormal returns are highest when the market first learns about (unconfirmed) rumours regarding a potential ECO. Abnormal returns at the beginning of the book-building phase are close to 0 or slightly negative, implying a close to normal distribution of positive and negative surprises regarding the details of the transactions. Abnormal returns on the first day of trading are negative on average, but increase in the subsidiary firm's first day performance, presumably because the remaining stake of the parent in the subsidiary firm increases in value. Third, the separate analysis of 'clean' vs. 'contaminated' announcements reveals that abnormal returns for the latter are significantly higher than for the former. This implies that companies tend to announce ECOs in conjunction with other news regarded as positive by investors. It also cautions future studies to control for potential announcement date contamination, which may be prevalent in past studies. Fourth, non-announcing companies with potential ECO candidates tend to show abnormal price reactions when another company announces an ECO. The reaction is more pronounced for parent firms owning subsidiaries in the same industry as the announcing parent firm's subsidiary. The direction of the impact is less clear, however, and seems to depend on the specific situation.

277 See Khanna/Palepu (2000b), p. 281. 278 Time period-dependency of results will be a recurring theme in later parts of this study, e.g., in the analysis of long-term operating and long-term price performance.

82

4 4.1

Long-term operating performance of European equity carve-outs Abstract

This chapter analyses the long-term operating performance of European parent and subsidiary firms involved in an equity carve-out (ECO) in a multi-year window around the event, both in terms of growth (sales, EBIT, assets and capital expenditure) and profitability (return on assets, return on sales). Both parent and subsidiary firms grow stronger and are more profitable than benchmark companies in the year of the ECO, and are less profitable in the year following the ECO. The findings are similar to previous results on IPOs and SEOs, and a simple test of the level of abnormal accruals supports the hypothesis that firms manage their earnings. For parent firms, there is little evidence of positive abnormal operating performance in the second and third year following the ECO, casting doubt on the divestiture gains hypothesis, according to which the parent firm should experience an improvement in operating performance following the ECO. For subsidiary firms, there is some evidence that growth continues to be abnormally positive in the two years following the ECO, indicating that part of the stand-alone gains may be permanent. Two distinct types of parent firms conduct successful ECOs: The first are financially non-distressed firms with a low number of business segments aiming to develop the business of their subsidiary firms, in which they continue to hold a considerable stake after the ECO. The second are financially distressed firms aiming to restructure by increasing the focus of their operations through the sale of a considerable stake in one of their subsidiary firms. Subsidiary firms grow stronger when parent firms intend to develop the subsidiary's business and when carved out of conglomerates with fewer business segments, implying that subsidiary firms require time to adapt to their new status as stand-alone entities. Subsidiary firms become more profitable when parent firms retain a larger stake, either because of better monitoring or because of informational asymmetry: The parent firm will not sell a large stake if it knows that operating performance is likely to improve in the coming years. Subsidiary firms also become more profitable when their parent firms have not been financially distressed prior to the ECO (implying that they inherit some of the characteristics of their previous existence), when carved out from parent firms in a different industry (due to the resolution of negative synergies), and when not carved out in the Internet bubble period (because of a lower quality of ECOs in that period).

4.2

Introduction

Analysing the operating performance in a multi-year window around the ECO is interesting for at least four reasons. First, analysing operating performance (in addition to share price performance as in chapter 5) focuses on fundamental data less subject to market frenzy. According to the divestiture gains hypothesis, ECOs increase the competitiveness of parent and subsidiary firms, for example because managerial skills

83

are better suited for the core operations than for the subsidiary operations (John/Ofek (1995)), or because incentives between managers and shareholders are more aligned (Daley/Mehrotra/Sivakumar (1997)). Vijh (2002) subdivides the divestiture gains hypothesis into multiple components279: The refocusing strategy hypothesis posits that positive value effects arise from the dissolution of negative synergies between parent and subsidiary firms coming from different industries. The financing strategy and investment hypotheses posit that ECO proceeds are used to repay debt and to finance new projects. The complexity hypothesis links value gains to the increased visibility of the new stand-alone subsidiary firm, which is easier to evaluate for investors. Finally, the managerial incentives hypothesis states that the better alignment of shareholder and manager interests through the use of stock-based compensation will increase firm value. The impact of these hypotheses should show in the operating performance of the companies involved in an ECO. 28~As pointed out by Bowman/Singh/Useem/Bhadbury (1999), this effect is likely to materialise only over a multi-year period. TM Consequently an analysis of operating performance in a multi-year window around the ECO is required to test the divestiture gains hypothesis. Second, analogously to the point made in the literature on the conglomerate discount 282, companies conducting an ECO choose to do so, and are hence not necessarily a random selection from the universe of all companies. For example, companies could self-select on some measure of pre-event operating underperformance, and the ECO would then represent a means of addressing a critical company situation. More generally, any pattern in pre-event operating performance may contribute to the understanding why firms choose to conduct an ECO. Third, existing evidence supports the idea of earnings management occurring in the context of IPOs (Teoh/Welch/Wong (1998)) and SEOs (Loughran/Ritter (1997)). In an ECO, both parent and subsidiary firms may be tempted to use accounting procedures with a view to reporting above-average results around the time of their visit to the capital market, hoping to become more attractive to potential investors. This could then be followed by a period of seemingly declining operating performance as the effects of the accounting procedures are reverted. Fourth, long-term operating performance may have explanatory significance for the long-term share price performance. In the long term, the market value of an asset should equal its fundamental value. If operating performance is a proxy for fundamental value, positive long-term operating performance should lead to positive long-term price performance.

279

280 281 282

See Vijh (2002), p. 155-156. Apart from the hypotheses detailed in this study, Vijh (2002) finds that in some cases, ECOs are motivatedby other reasons, including takeover defence, tax reduction, and regulatory compliance. See Hulburt/Miles/Woolridge(2002), p. 91. See Bowman/Singh/Useem/Bhadury(1999), p. 35. See Campa/Kedia(2002), p. 1731, and Martin/Saryak (2003), p. 53.

84

This study adds to the current body of knowledge in at least three ways: First, both parent and subsidiary firm operating performance is analysed in a [-3;+3] year window around the ECO. The longer time horizon of the analysis (compared to Hulburt/Miles/Woolridge (2002)) allows testing whether ECOs lead to tangible improvements in operating performance over a multi-year period. Second, the longer time horizon allows testing the hypothesis that earnings management occurs, similar to the findings for IPOs and SEOs. A simple test of the level of abnormal accruals is implemented to assess this previously untested hypothesis. Third, the cross-sectional distribution of both level (i.e., growth rates) and profitability (i.e., margins) measures of operating performance is explained as a function of various ECO characteristics, including the stake retained by the parent in the subsidiary firm, dummies for financial distress, industry association, the region in which the ECO occurred, the motivation stated for the ECO, and controlling for size and time period. Finally, the results on longterm operating performance are used in a later study as an explanatory variable for the long-term price performance. The hypothesis is that there is a positive relationship between operating and share price performance for companies involved in an ECO. The chapter is organised as follows: Section 4.3 reviews relevant aspects of previous studies on the operating performance changes following ECOs. Section 4.4 describes the general methodology applied, and section 4.5 describes the data and the specific analyses. Section 4.6 discusses and interprets the results, and section 4.7 concludes.

4.3

Literature review

Previous studies have analysed the long-term development of a firm's operating performance in the context of various corporate finance events. John/Ofek (1995), using a US sample of 321 divestitures from the 1986 to 1988 period, report that focusincreasing asset sales lead to improvements in the selling firm's return on sales and retum on assets in each of the three years following the sale. They argue that a potentially better fit between buyer and divested asset explains some of the value gains for the seller. The change in operating performance is also positively associated with the abnormal announcement retum, indicating that markets are able to at least partly anticipate later operating improvements. Daley/Mehrotra/Sivakumar (1997), using a US sample of 85 spin-offs from the 1975 to 1991 period, document an increase in retum on assets for spin-off parent firms from the year prior to the year following the event. Retum on assets only increases when parent and subsidiary firms are active in different industries (based on their respective two digits SIC code). The findings are interpreted as supporting the hypothesis that spin-offs serve to remove negative synergies between unrelated business segments and allow managers to use their abilities in those operations which they understand best. 283

283

Daley/Mehrotra/Sivakumar (1997) also hypothesize that the value increase in cross-industry spin-offs can be explained by bonding benefits, where bonding refers to the fact that managers willingly forego

85

For IPOs, Mikkelson/Partch/Shah (1997) find that operating performance decreases in the first year after the event, and stays constant in the following nine years. They find no support for their hypothesis that changes in the ownership by insiders and blockholders explain operating performance. Small companies perform worse than large companies. They also speculate that declines in operating performance are linked to a company assessing its operating performance as favourable and potentially not sustainable, and hence going public at an opportune moment in time. Teoh/Welch/Wong (1998) find that IPO companies reporting a high level of accruals relative to their cash flow, which is interpreted as a sign of earnings management, have a worse share price performance than companies not applying this measure. Loughran/Ritter (1997) find that the operating performance of firms conducting an SEO peaks around the time of the event and deteriorates thereafter. They interpret this as evidence of earnings management by issuing firms. Valuation multiples at the time of the event do not seem to incorporate an expectation of deteriorating operating performance, indicating that the market wrongly expects the (earnings management induced) operating improvements around the time of the event to be permanent. Another series of studies analyses the operating performance of ECOs. Hulburt/Miles/Woolridge (2002) find positive sales, assets and capital expenditure growth rates for subsidiary firms, and improved profitability (return on assets, return on sales) for parent firms in the first year following the event. Boone/Haushalter/Mikkelson (2003) analyse the impact of the stake retained by the parent in the subsidiary firm, and find that the operating performance of the parent firm improves only when the latter disposes of its entire stake. Subsidiary firm operating performance is unaffected by the stake retained. In contrast, Powers (2003) finds that there is a positive relation between the subsidiary firm's long-term operating performance and the stake retained. He also finds that operating performance of the subsidiary firm peaks in the year of the ECO and declines in the three following years. Parent firms are also more highly leveraged and underperform their benchmarks both before and following the ECO. Powers (2003) interprets his findings as evidence for the financing rationale of ECOs, whereby parent firms sell equity in their subsidiary firms at times when their operating performance peaks. To summarise, existing studies imply that while spin-offs (in particular when leading to an increase in industrial focus) improve operating performance in the years following the event, capital-raising activities such as IPOs and SEOs are associated with operating performance deterioration in the year following the event, possibly due to earnings management. Since ECOs combine elements of both a spin-off and an IPO, the direction of the change in operating performance is undetermined a priori. Existing evidence is scarce and limited to the analysis of operating performance in a two-year window around the event (Hulburt/Miles/Woolridge (2002)). The determinants of the cross-sectional distribution of LTOP results are also unclear. For example, evidence on the possibility to cross-subsidize between business segments by separating the subsidiary from the parent firm. Howeverthey find little empirical support for this bonding hypothesis (p. 272-277).

86

the impact of the stake retained by the parent firm on the subsidiary firm's operating performance is mixed, which may be the result of two conflicting forces: On the one hand, a higher stake retained could lead to more intense monitoring of the subsidiary by the parent firm, resulting in a positive development of operating performance. On the other hand, a lower stake retained could lead to more independence for the subsidiary firm, with the benefits of independence (e.g., better alignment of management and shareholder interests) leading to positive operating performance.

4.4

General methodology

Compared to long-term abnormal share price performance, the analysis of long-term abnormal operating performance is methodologically less contentious. Three choices need to be made. First, what measure of operating performance should be employed? Second, what test statistic should be used for statistical inference? Third (and most important), how is abnormality assessed, i.e., how is a firm's expected performance defined, and what benchmark is it compared against? These three questions are addressed next. First, in a comprehensive simulation study, Barber/Lyon (1996) find that using either return on assets (ROA), retum on cash-adjusted assets, return on sales or retum on the market value of assets produces well-specified test statistics. A cash flow based return measure produces less powerful test statistics. Consequently the first four measures are implemented in this study. Second, Barber/Lyon (1996) find that the Wilcoxon signed-rank test statistic is more powerful than the conventional t-statistic in all sampling situations. They attribute this to the cross-sectional distribution of operating performance being fat-tailed. Consequently statistical significance is assessed via this non-parametric test statistic. Third, to assess abnormality of operating performance, a benchmark of 'normal' performance (i.e., excluding the impact of the ECO) needs to be established. This benchmark is based on the performance of firms matched to each sample firm on industry, size and/or prior performance. Level models equate a sample firm's expected performance with the performance of these matched benchmark firms. The disadvantage of this methodology is that it ignores the past relationship between the sample firm's performance and the benchmark. This may lead researchers to wrongly attribute differences in operating performance to an event which is not the cause for these differences. TM Consequently Barber/Lyon (1996) recommend using change models, which incorporate a sample firm's actual past performance. 285 A firm's 284

285

Barber/Lyon (1996) provide an illustrative example: A firm which has had a high operating performance relative to its industry peers as a result of unusually profitable projects is likely to continue to produce above-averageprofits after the event. However these above-averageprofits are obviously not caused by the event. Therefore not controlling for this difference in prior performance will bias results. See Barber/Lyon (1996) p. 365-367.

87

expected performance is therefore estimated as the sum of its past performance, plus the change in the performance of its industry-, size- and/or performance matched benchmark firms. Specifically, sample firm i's expected performance in year t, E(Pi), is

E(Piit )= Pi,t-1 + ([:)lit -- PIi,t-, ), where Pi, t-1 is firm i's performance in year t-l, and (Plit- PIi, t-l) is the change in the operating performance of sample firm's i benchmark. Barber/Lyon (1996) conduct a simulation study by creating random samples of firms drawn from the entire universe of firms available in the Compustat database. For these random samples (i.e., samples whose average firm characteristics closely match the average firm characteristics of all companies in the benchmark universe) they find that using benchmark firms consisting of either industry-, size- or performance-matched companies produce well-specified test statistics: Empirical rejection rates are close to theoretical rejection levels, which would be expected if there was no abnormal operating performance on average. 286 However, if the sample firm characteristics systematically deviate from those of the other companies in the benchmark universe, misspecifications can occur. Specifically, Barber/Lyon (1996) find that when sample companies have ROA outside of the two central deciles, only the performance-matched benchmark produces well-specified and powerful test statistics. Size-related biases do not seem to produce misspecification, except in conjunction with unusual ROA: For samples consisting of small firms with high ROA even the performance-matched benchmarks produce misspecified test statistics. 287 In this case Barber/Lyon (1996) suggest using a benchmark matching firms on size and performance, without regard to industry. This benchmark also produces well-specified test statistics across all other combinations of performance and size quantiles. Given these findings, it is important to assess the composition of the sample used in the present study relative to the total universe of potential benchmark firms. If the sample consists only of firms whose ROA is in the fifth or sixth decile of all firms in the benchmark universe, then either industry, size or performance-matched benchmark firms will produce well-specified test statistics. If, on the other hand, the sample firm contains a considerable number of firms whose ROA is either above or below the two central deciles, then only the performance-matched benchmarks will produce wellspecified results. Consequently, for each sample firm, the rank of its ROA in its matching year relative to all other companies is calculated. Appendix 21 shows the rank for each sample firm, as well as mean and median ranks for the whole sample for each year in the study. Only a small percentage of the sample firms have ROA within the fifth or sixth decile, for which Barber/Lyon (1996) find no misspecification of either of the above-mentioned benchmarks. The remaining companies have ROA in either higher or lower deciles. The mean and median rank are outside of the two central deciles for 286 See Barber/Lyon(1996), p. 396. 287 See Barber/Lyon(1996), p. 386.

88

almost half of all years in the sample period. Therefore only performance-matched benchmarks (as opposed to industry- and/or size-matched benchmarks) are likely to produce well-specified results. Similarly, if the sample only contains small firms with high ROA, using performancematched benchmarks will lead to misspecified test statistics. To test this, Appendix 22 shows the size rank (where size is measured as book value of assets in accordance with Barber/Lyon (1996)) for each sample firm in its matching year relative to all other companies in the benchmark universe. On average parent firms are in the upper size quantiles. Appendix 23 visualises the ROA rank/size rank combinations for each sample firm. 3 out of 133 companies are in the top performance/small size portfolio. 288 Hence, while there are only few companies in the top performance/small size portfolio (for which Barber/Lyon (1996) notice misspecifications in most matching methods), the sample is characterised by some large firms. Therefore a test methodology using performance- and size-matched benchmarks without regard to industry is also implemented, because it produces well-specified test results across all combinations of performance and size quantiles. 289 Lie (2001) performs similar simulations and finds that the most robust matching procedure matches on past levels of and changes in operating performance, and on market-to-book ratios. 29~ The disadvantage of matching on all three criteria is that matching firms may not be available for all sample firms, necessitating the establishment of alternative selection rules to prevent a sample selection bias arising from the non-consideration of firms for which no benchmark can be found. In that case, Lie (2001) suggests that the matching of control firms must take into account the characteristics of the sample: If sample firms are characterised by high market-to-book rations (MTB) relative to the benchmark universe, then MTB should be used as a matching criterion. Hence, the ranking analysis described in the above paragraph is repeated for MTB. Appendix 24 shows that the sample firms' MTB are dispersed across the quantiles. The sample thus is not biased towards high- or low MTB firms, and Lie's (2001) recommendations on which of his five benchmarks are best-specified are valid for this study without adaptation. TM

4.5

Data and specific analyses

The sample of ECOs is constructed as described in section 2.1.5. All financial companies are excluded from the analysis. Further companies are dropped because of

288

289 290

291

In accordance with Barber/Lyon (1996), 'top performance' refers to the top 33%-quantile of ROA and 'small size' refers to the bottom 33%-quantile of size. See Barber/Lyon (1996), p. 387. The matching procedure also involves industry matching, but if no control firm is found with the required operating and market-to-bookcharacteristics, industry matching criteria are relaxed. See section 4.6.1.2 for more details on how the Lie (2001) benchmarks are constructed.

89

limited data availability. Because different time horizons are considered, the number of companies varies across the analyses. All data required for the construction of the performance measures (sales, EBIT, total assets, cash, common equity, market value of equity, capital expenditure, industry code 292) is from Datastream. Local currencies are converted to Euro using fiscal-year end exchange rates. In accordance with Barber/Lyon (1996), the following performance measures are constructed293: Return on assets (ROA), defined as operating income (earnings before taxes and interest, EBIT) divided by the average book value of total assets, where the average is calculated as the sum of year-beginning and year-end book value of total assets divided by two; return on cash-adjusted assets (ROCAA), defined as EBIT divided by the average book-value of cash-adjusted assets, where cash-adjusted assets equal total assets less cash, and the average again is calculated as the sum of yearbeginning and year-end book value of assets divided by two; return on sales (ROS), defined as EBIT divided by sales; and return on the market-value of assets (ROMVA), defined as EBIT divided by the average market value of total assets, calculated as book value of total assets less the book value of common equity, plus the market value of common equity, and where the average again is calculated as the sum of year-beginning and year-end value of assets divided by two. ROA, ROCAA, ROS and ROMVA are thus all based on a measure of operating income (EBIT), rather than net income, because the former does not include special items (in addition to interest and taxes) and is therefore considered to be a 'cleaner' measure of operating performance. TM ROA measures the return the firm generates on all of its assets. However, since the cash balance is also likely to be influenced by financing (rather than operating) activities, it may be more appropriate to exclude those assets which are cash and cash equivalents (ROCAA). This is particularly true if the event under consideration is a capital-raising activity, such as an ECO. The disadvantage of using ROCAA (as opposed to ROA) is that it assumes that all cash is non-operating, while in reality a firm does need a certain amount of cash to run its operations. An issue with both ROA and ROCAA is that the book value of assets (calculated on the basis of historic costs) used in the calculation may not reflect the true market value of the assets, which will over- or underestimate the operating performance measure. One way of overcoming the historic cost problem is to use the market value of assets (ROMVA). The disadvantage of using market values is that market values may fluctuate considerably, rendering the estimate of operating performance subject to the arbitrary choice of a specific date. Another issue with both ROA and ROCAA is that a firm may possess non-operating assets which increases the denominator and will bias ROA and 292

293 294

Industry codes are based on the NACE (Nomenclature G6n6rale des Activit6s Economiques dans l'Union Europ6ene) system, rather than the SIC (Standard Industrial Classification) system, which is popular in US papers. First, NACE seems a more appropriate measure, given the European nature of the sample. Second, for European companies, availability of NACE codes in Datastream is far superior to availability of SIC codes. See Barber/Lyon (1996), p. 387-393. See Barber/Lyon (1996), p.363-364.

90

ROCAA downward. Using sales, rather than assets, may overcome this problem (ROS). The disadvantage of using ROS (as opposed to ROA, ROCAA and ROMVA) is that a firm which increases its sales (and its profits pro rata) while maintaining its level of assets does not show an improvement in ROS-based profitability, although it has increased its productivity in relation to the capital which was employed to produce the profits. Therefore since no measure of operating performance dominates, all four are implemented. Analogously to Hulburt/Miles/Woolridge (2002), operating performance is assessed both by changes in the profitability of operations (as measured by the four variables just described), and by changes in the level of operations (as measured by annual growth rates of sales, total assets, EBIT and capital expenditure, respectively). The identification of the universe of relevant benchmark securities is based on Datastream. First, all securities whose home market is listed as one of the 13 countries in this study are identified (28,683 firms). Second, all securities whose currency does not correspond to one of the 13 countries' local currencies or the Euro, or which are not listed in their home country, are excluded (these are usually secondary listings at another stock exchange). Third, all securities for which no or limited share price data is available are excluded. Fourth, securities with a non-equity nature (e.g., German "Genussscheine") are excluded. Finally, those securities listed by Datastream as having been replaced with another security are excluded. This procedure results in a total relevant universe of 11,288 companies. 295 Appendix 25 shows how these companies are distributed across countries and industries. All financial companies (NACE code 8) are excluded. This leaves 9,384 firms. For these firms, all required accounting data is retrieved on an annual basis, resulting in 51,053 firm-year observations. The following analyses are implemented in Matlab. The specific benchmarks are constructed using algorithms based on Barber/Lyon (1996). 296 These consist of a selection criterion, and a series of alternative rules if no company matches the first criterion. These alternative rules are necessary to avoid a selection bias resulting from the non-consideration of companies for which no benchmark can be found. Given that the sample in the current study consists of ECOs from different European countries, the algorithms presented by Barber/Lyon (1996) need to be modified. Since the alternative rules are not theoretically based, their design

295

296

In a previous version of this study, securities with extreme price (>200% in any month) and accounting item (>1,000% in any year) movements were excluded, because these were assumed to be tiny firms and/or penny stocks potentially distorting the analysis. As was pointed out to me, this created the impression of arbitrariness. In the present version of the study, these companies are therefore not excluded from the benchmark universe. Instead, the issue of outliers is addressed in each respective analysis. Usually this takes the form of winsorizing variables at +/- three standard deviations as suggested by Ang/Zhang (2004), p. 763. The key results in this analysis are robust to either method. See Barber/Lyon (1996), p. 369-371.

91

and respective order will be based on some indicative evidence regarding their relative influence on firm performance. 297 Assuming that the operating performance of companies is at least partially influenced by their home countries' economies, it seems reasonable to assume that matching on geography, in addition to all factors mentioned above, will produce higher-quality benchmarks. However, matching on an additional factor also exacerbates the curse of dimensionality, and thus potentially reduces the number of benchmark companies for each sample firm. Also, it is unclear which criterion to relax first, i.e., how to rank the alternative rules. To clarify, consider the first benchmark, two digits matched. According to Barber/Lyon (1996), this benchmark includes all firms with the same first two digits of the industry code as sample firm i, excluding the sample firm. Adding the country factor imposes the additional criterion of the benchmark companies having to originate from the same country as sample firm i. If there is no other firm with the same two digits industry code from the same country, at least two alternatives exists: Either the industry criterion is relaxed, and the alternative rule is to use all companies with the same one digit industry code from the same country; or the geographical criterion is removed, and the alternative rule is to use all companies with the same two digits industry code without regard to the country (i.e., Barber/Lyon's original rule). The underlying question is whether a company's operating performance is more in line with its one digit industry code peers in the same country, or with its two digits industry code peers across countries. A straightforward way to assess this is to analyse the mean standard deviation of the operating performance measure of these two groups. The idea is simple: If the mean standard deviation of one group significantly exceeds the mean standard deviation of the other group, the latter is more homogeneous, hence likely to produce closer benchmarks, and should be preferred in the order of alternative rules. For each two digits industry, the standard deviation of the operating performance measure of all available firm years is calculated for each year from 1984 to 2004, then averaged across all years, and then averaged across all industries, both simply and weighted by the number of firm-year observations in each industry. This procedure is repeated for all 117 combinations of 13 countries and nine one digit industries. The simple (weighted) average standard deviation for the two digits industry group is 12.1% (12.4%), and 8.5% (11.4%) for the 'one digit and same country' group. The difference is statistically significant (p-value of 0.0028 for a two groups difference of means test). Operating performance is thus more in line among firms with the same one digit industry code coming from the same country, than it is within the group of firms with the same two digits industry code coming from different countries. Therefore matching on geography seems more important than matching on industry. The ranking of the alternative sorting rules is adjusted to reflect this, as described below.

297

An alternative would be to conduct a simulation study similar to Barber/Lyon (1996), explicitly taking the geographical factor into account, and testing the specification of various benchmarks incorporating this factor. However, this by far exceeds the scope of this study.

92

The first benchmark, two digits matched (BL1), comprises all firms sharing the same first two digits of the industry code and originating from the same country as sample firm i, excluding this sample firm. If none of the companies meet this rule 298, all companies with the same one digit industry code from the same country are used. If this rule is not met, all companies with the same two digits industry code without regard to the country are used. The second benchmark, four digits matched (BL2), comprises all firms sharing the same four digits of the industry code and the same country as sample firm i, excluding the sample firm. If there is no other firm with the same four digits (three digits) 299 industry code from the same country, all companies with the same three digits (two digits) industry code from the same country are used. If neither of these rules is met, all companies with the same four digits (three digits; two digits) industry code without regard to the country are used. The third benchmark, size-matched (BL3), comprises all firms sharing the same two digits (one digit) industry code and having book value of assets within 70%-130% of the sample firm, and coming from the same country as the sample firm. If neither of these rules is met, all companies with the same two digits (one digit) industry code and having book value of assets within 70%-130% of the sample firm, without regard to the country, are used. If neither of these rules is met, the firm with the same two digits code from the same country with the closest book value of assets is used. The fourth benchmark, performance-matched (BL4), comprises all firms sharing the same two digits (one digit, no digit) industry code and having ROA either within 90%110% of sample firm i's ROA, or within +/-0.01 of this value 3~176 and in the same country. If this rule is not met, all companies with the same characteristics as above are used, except without regard to country. The fifth benchmark, size and performance-matched (BL5), comprises all firms having book value of assets within 70%-130% of the sample firm, ROA either within 90%110% of the sample firm's ROA, or within +/-0.01 of this value, and coming from the same country. If this rule is not met, all companies with the same characteristics as above are used, except without regard to country. If this rule is not met, all companies having book value of assets within 70%-130% of the sample firm are determined, coming from the same country, and then the company with the closest ROA is used as

298

299

300

Here and in the following sections, 'not meeting the rule' refers to either no company matching the data requirements (i.e., sharing the same industry code, originating from the same country etc.) or, while matching the data requirements, not having sufficient further data for each year in which data is available for the sample firm. For the sake of brevity and readability, the alternative rules are described in a compact fashion: The terms in the brackets replace the immediately proceeding term as the object of the immediately following rule. The latter addition is due to Lie (2001), to accommodate sample firms whose ROA is close to zero, see Lie (2001), p. 82.

93

the benchmark. If this rule is not met, the same criteria are applied without regard to country. For all five benchmarks, the median value of the operating performance measure of the resulting group of firms is used as the specific benchmark for the respective sample firm. Matching of parent firms is performed based on annual data from three years prior to the ECO. In the analysis of subsidiary firms, matching is also based on annual data from three years prior to the ECO, but all performance measures are based on year-end assets, not weighted assets as in the case of parent firms. 3~ The selected benchmark companies are kept constant throughout the time horizon of the analysis. If a sample firm gets delisted subsequent to the ECO, no abnormal returns are calculated for the remaining years. If benchmark companies get delisted, performance is calculated based on the remaining benchmark companies.

4.6

Empirical results

The following section discusses the results of the empirical analyses of the long-term operating performance of parent (section 4.6.1) and subsidiary (section 4.6.2) firms. A potential explanation for the results, earnings management, is analysed (section 4.6.3). Subsequently the cross-sectional distribution of performance measures is explained and interpreted in a multivariate regression framework for parent (section 4.6.4) and subsidiary (section 4.6.5) firms.

4.6.1 Empiricalresultsforparent firms 4.6.1.1 Barber~Lyon(1996) methodology Appendix 26 and Appendix 27 show the abnormal profitability and growth, respectively, for parent firms, using the four different profitability and the four different level performance measures, in combination with the five different benchmarks. Appendix 28 shows the algorithms used with each benchmark and the number of companies included in each respective analysis. Given the sample characteristics, benchmarks BL 1, BL2 and BL3 could be misspecified because they do not consider the sample firms' past performances, so care should be taken when interpreting results based on these benchmarks. BL4 and BL5, on the other hand, have been shown by Barber/Lyon (1996) to produce well-specified results. Interpretation of results will therefore focus on these two benchmarks.

301

Since IPO prospectuses in most cases give accounting data for three years preceding the IPO year, average assets are only available for two years prior to the ECO. The alternative to using year-end assets is thus to match in the second year before the ECO. Robustness tests show that the main results of the analysis are similar, but using year-end assets allows the analysis of abnormal performance in the second year prior to the ECO, and therefore this alternative is chosen.

94

The most striking result is a statistically significant outperformance by parent firms in all four operating profitability measures in the year of the ECO, followed by a statistically significant underperformance in ROCAA, ROS and ROMVA in the year following the ECO. There is no significant over- or underperformance in either the two years preceding, or the second and third year following the ECO (with the exception of a significant underperformance in ROA and ROMVA in the second year preceding the ECO). The results are consistent with Loughran/Ritter (1997), who find that the operating performance of companies issuing seasoned equity peaks in the year of issuance and declines thereafter. They suggest that earnings management is one potential explanation for their finding. Section 4.6.3 analyses this hypothesis in more detail. The results are inconsistent with Hulburt/Miles/Woolridge (2002), who find a significantly positive abnormal performance of parent firms in ROA, ROCAA, ROS and ROMVA in the year following the ECO. 3~ They interpret their findings along the lines of the divestiture gains hypothesis, arguing that parent firms have become more efficient following the ECO. However, as pointed out by Teoh/Welch/Wong (1998), earnings management is likely to continue in the first year following the event, as firms aim to prevent lawsuits from investors disappointed about a sudden reversal in earnings following the IPO. 3~ Hence, Hulburt/Miles/Woolridge's (2002) result can also be consistent with the assumption of earnings management. Also, the fact that there is no abnormal performance in the second and third year following the ECO is an indication for the transitory nature of the abnormal positive operating performance. The findings in the present study therefore do not support the divestiture gains hypothesis, according to which there should be tangible improvements in operating performance in the years following the ECO. An alternative explanation for not finding positive abnormal performance in the years following the ECO is that firms would have suffered an even worse operating performance, had they not engaged in the ECO. The value gain from the ECO would then come in the form of avoiding a potentially negative performance and realizing a 'normal' operating performance, rather than improving from a 'normal' performance to an abnormally positive performance. However, the fact that there is no significantly negative abnormal performance in the years preceding the ECO casts some doubt on this explanation. Looking at growth rates, parent firms grow their sales and assets significantly stronger than matched companies in the two years preceding the ECO, and show a significantly higher EBIT growth rate in the year preceding and in the year of the ECO, when they also have significantly higher asset growth rates. Again this finding is consistent with the notion of parent firms managing earnings to peak in the year of the ECO. Growth rates of sales are significantly below those of matched companies in the third year following the ECO, and growth rates of assets are significantly lower in the second and 302

303

A closer comparison with Hulburt/Miles/Woolridge (2002) is difficult because they report neither results for the years preceding, nor for the years following the first year after the ECO. See Teoh/Welch/Wong(1998), p. 1936.

95

third year following the ECO. Most of the remaining growth rates in the second and third year (including EBIT and capex) are also negative, albeit not significantly so. As discussed in section 4.6.3, a potential explanation for this is the reversal of the accounting mechanisms used to manage earnings around the time of the ECO. The finding of negative growth rates is also consistent with the idea of carved-out subsidiary firms being the high-growth segments of companies: Once they are carved out, the remaining parent firm is likely to grow slower. TM

4.6.1.2 Lie (2001) methodology To check the robustness of the results, five alternative benchmarks as suggested by Lie (2001) are constructed. He determines single control firms rather than taking the median value of control groups as in the Barber/Lyon (1996) methodology. While the algorithms for the alternative rules are based on Lie (2001), they have been adapted to incorporate the additional geographical dimension of the sample. The first benchmark, L1, identifies all firms with the same two digits (one digit; no digit) industry code, and ROA within 90%-110% or within +/-0.01 of the sample firm, from the same country. The firm with the closest ROA to that of the sample firm is chosen as the control firm. If no firm is found, the process is repeated without regard to country. The second benchmark, L2, identifies all firms with the same two digits (one digit; no digit) industry code, and a change in ROA within 90%-110% or within +/-0.01 of the sample firm, from the same country. The firm with the closest change in ROA to that of the sample firm is chosen as the control firm. If no firm is found, the process is repeated without regard to country. The third benchmark, L3, combines L 1 and L2. It identifies all firms with the same two digits (one digit; no digit) industry code, ROA within 80%-120% or within +/-0.01 of the sample firm, and a change in ROA within 80%-120% or within +/-0.01 of the sample firm, from the same country. The firm with the lowest sum of absolute differences of ROA and ROA change to the sample firm is chosen as the control firm. If no firm is found, the process is repeated without regard to country. The fourth benchmark, L4, extends L1. It identifies all firms with the same two digits (one digit; no digit) industry code, ROA within 80%-120% or within +/-0.01 of the sample firm, and a MTB within 80%-120% or within +/-0.1, from the same country. The firm with the closest ROA to that of the sample firm is chosen as the control firm. If no firm is found, the process is repeated without regard to country.

304

The growth rate in specific cases obviously depends on whether (and if, how: fully, at equity, as an associate investment) the remaining stake in the subsidiary is consolidated into the parent's group accounts.

96

The fifth benchmark, L5, extends L3. It identifies all firms with the same two digits (one digit; no digit) industry code, ROA within 80%-120% or within +/-0.01 of the sample firm, a change in ROA within 80%-120% or within +/-0.01, a MTB within 80%120% or within +/-0.1, and from the same country. The firm with the lowest sum of absolute differences of ROA and ROA change to the sample firm is chosen as the control firm. If no firm is found, the process is repeated without regard to country. As above, matching is performed based on annual data from three years prior to the ECO for parent firms. For subsidiary firms, matching is performed using annual data from two years prior to the ECO because change figures are used, and the data in the IPO prospectus in most cases dates back only three years. The selected benchmark company is kept constant throughout the time horizon of the analysis. If either the sample firm or the control firm gets delisted subsequent to the ECO, no abnormal returns are calculated for the remaining years. Appendix 29 shows the results of the abnormal performance analysis using these five alternative benchmarks, and ROA as the operating performance measure. Given Lie (2001)'s simulation findings, interpretation is limited to the best-specified benchmarks, L3 and L5. The main results are analogous to the previous analysis: There is clear indication of a significant outperformance in the year of the ECO, and significant underperformance in the year following the ECO. In the remaining years in the analysis there is no evidence of significant over- or underperformance. The results using this alternative methodology are thus consistent with the result using the standard Barber/Lyon (1996) methodology, increasing the confidence in the robustness of the results.

4.6. 2 Empirical results for subsidiary firms Appendix 30 and Appendix 31 show the abnormal profitability and growth, respectively, for subsidiary firms, using three different profitability measures 3~ four different level performance measures and five different benchmarks based on Barber/Lyon (1996). Again as in the analysis of parent firms, interpretation of results focuses on benchmarks BL4 (performance- and industry-matched) and BL5 (performance- and size-matched), which have been shown to produce well-specified results. 3~ All three profitability measures indicate a positive performance in the two years prior to the ECO, as well as the year of the ECO, though significance is generally not found. 3~ In the year immediately following the ECO, performance is negative across all three profitability measures, and significantly so for ROCAA, while marginally not so for the

305

306 307

The fourth profitability measure, ROMVA, is not available for subsidiaries because obviouslymarket values are not available prior to the ECO. See section 4.4 of this paper on methodology,see also Barber/Lyon (1996) p. 386-387. For ROCAA, results are marginally not significant for the year of the ECO (p=0.1000 and p=0.1001 for BL4 and BL5, respectively).

97

two remaining operating performance measures. In the two following years there is no trend and significance in results. Looking at growth rates, subsidiary firms show significantly higher growth in sales, EBIT, total assets and capital expenditure up to the year of the ECO, compared to benchmark firms. For sales and assets, this higher growth pertains in the two years following the ECO. The EBIT growth rate in the year following the ECO is negative (albeit not significantly so), mirroring the findings on the below-average profitability measures. In the third year, growth rate of sales and assets is similar to, and the EBIT growth rate continues to be below, that of benchmark firms. Capex growth rates are below benchmark in the year following the ECO, and seem to revert to normal in the second and third year. Combining the results from the analysis of profitability and growth measures, the following picture emerges: Subsidiary firms perform well both in profitability and growth terms in the years preceding the ECO. They experience a drop in profitability in the year following the ECO while continuing to grow their sales and assets base for one to two years after the ECO, suggesting that at least part of the stand-alone gains for subsidiary firms may be of a permanent nature. Potential explanations for the drop in profitability measures include timing considerations and earnings management. The next section analyses these questions in detail.

4.6.3

Explanations for patterns in operating performance

4.6.3.1 Market timing or earnings management Both parent and subsidiary firms on average seem to experience an outperformance of various operating performance measures around the time of the ECO, and a drop in these measures in the one and two years following the ECO. There are at least two potential explanations. First, Loughran/Ritter (1997), who find that the operating performance of companies issuing seasoned equity peaks in the year of issuance and declines thereafter, explain their finding as being the result of earnings management. A similar interpretation could be made in this study both for parent and for subsidiary firms: Parent firms planning an ECO aim to present themselves to capital markets in a favourable light, hoping to thereby increase proceeds from the ECO. Specifically, they attempt to paint a rosy picture of their subsidiary firms as businesses of above-average profitability and growth through the use of creative accounting techniques. While hence earnings management seems more likely on the subsidiary level, it may also occur on the parent firm level, if the parent firm believes that positive group results will be interpreted by potential investors as a positive signal on how the group (and by deduction the subsidiary firm) is managed. Second, Powers (2003), who finds that subsidiary firm absolute operating performance tends to peak in the year of the ECO, suggests that parent firms time the ECO to

98

coincide with an abnormally high level of subsidiary firm performance. 3~ Parent firms thus take advantage of a period of unusual and potentially unsustainable subsidiary firm operating performance to sell a stake, hoping that markets will interpret positive performance as lasting rather than temporary, and therefore pay more money to buy the stake. Again, the argument also seems applicable for the operating performance of parent firms themselves: If the visit to the capital markets coincides with a period of positive operating performance development, investors may take the parent firm's positive operating performance as a positive signal for the future operating performance of the subsidiary firm. As pointed out by Rangan (1998), the two explanations are not mutually exclusive. 3~ Indeed, it seems difficult to distinguish between the two, and finding evidence for one does not rule out the existence of the other. Section 4.6.3.2 analyses whether at least part of the pattern in parent and subsidiary firm operating performance can be explained by earnings management. Chapter 5 on LTPP will analyse whether there is evidence on market timing, too.

4.6.3.2 Analysis of potential earnings management Earnings management can be defined as occurring "when managers use judgement in financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting numbers ''31~ Essentially earnings management works by increasing current earnings through borrowing against future earnings. Means of actual implementation include the choice of accounting methods, revision of estimates (e.g., for the average usable life of certain assets), acceleration or deferral of sales revenues and expenses, creation or liquidation of reserves, off-balance sheet financing (e.g., through sales & lease back transactions), LIFO layer liquidations and debt-equity swaps. TM Since it cannot be directly observed, empiricists have developed a series of variables considered to be indicators of earnings management. Jones (1991) focuses on total accruals and develops a firm-specific time series model to estimate abnormal accruals. Teoh/Welch/Wong (1998) regress total accruals on the change in sales (less receivables) and on property, plant & equipment (both scaled by total assets) in the cross-section of all non-IPO firms in the same two digits industry. They attempt to decompose the total change in accruals into a nondiscretionary part (presumably caused by industry effects) and a discretionary part, which is assumed to be the result of earnings management.

308 309 310

See Powers (2003), p. 37. See Rangan (1998), p. 102.

Healy/Wahlen (1999), p. 368. 311 See Ducharme/Malatesta/Sefcik (2001), p. 370, for a good overview of some of these measures.

99

The present study uses a related but somewhat cruder way of assessing earnings management. Based on Kim/Park (2005), total accruals (TAC) are calculated as net income before extraordinary items less operating cash flow. Total accruals are scaled by total assets (TAC/TA). Operating cash flow is defined as EBITDA less interest, taxes and capital expenditure. Essentially TAC/TA therefore captures only accruals related to depreciation. 3~2 For each sample firm, TAC/TA is calculated for each year in the [-2;+3] year window around the ECO. To accommodate changes in accruals arising from industry and country specific factors, an expectations model is used: Expected TAC/TA in year t is calculated as TAC/TA in year t-1, plus the median change in TAC/TA for all firms in the same two digits industry and in the same country. If this benchmark group contains less than five firms with available data, one digit industries are used. If this benchmark group again contains less than five firms with available data, all firms in the same two digits industry irrespective of the country are used. If this benchmark group again contains less than five firms with available data, all firms in the same one digit industry irrespective of the country are used. Abnormal TAC/TA is calculated as the actual TAC/TA less the expected TAC/TA. If earnings management occurs, abnormal TAC/TA is likely to be positive in the year before and the year of the IPO, and likely to be negative in the following years as the effects of the previous earnings management are reverted. Table 14 shows mean and median abnormal TAC/TA for parent and subsidiary firms for each year in the [-2;+3] year window around the ECO. For parent firms, abnormal TAC/TA increases f r o m 1.6% to 0.4% from Y-2 to Y-l, indicating a relatively higher level of accruals in the year prior to the ECO (although the positive value in Y-1 is not significantly different from 0). The measure is not different from 0 in the year of the ECO, contradicting the hypothesis of earnings management by parent firms. However, this could also imply that the potential earnings management is reverted faster than expected. For subsidiary firms, the mean abnormal TAC/TA is significantly positive in the year of the ECO, and significantly negative for the subsequent year. The negative abnormal TAC/TA is marginally not significant in Y+2 (p=0.1078); and no significance is found for Y+3. This pattern corresponds exactly to the earnings management hypothesis: Positive abnormal accruals in Y0 (initiation of earnings management), negative abnormal accruals in Y+ 1 and Y+2 (gradual reversion of earnings management), and no abnormal accruals in Y+3 (no earnings management effect). The fact that abnormal accruals are positive in the year of the IPO, i.e., using year-end data for the year in which the IPO occurs, confirms Teoh/Welch/Wong's (1998) assumption that earnings management may continue in the time after the IPO, potentially for two reasons: First, firms reverting their accounting methods immediately following the IPO may have to face lawsuits by

312

It would have been desirable to use a better proxy for operating cash flow, as EBITDA-interest-taxcapex does not take into account certain accounting items, e.g., an increase in current assets or a decrease in current liabilities, which decrease operating cash flow but not the proxy used in this study. Unfortunately operating cash flow was not available for a large part of the sample firms so that an extended analysis was not implementable.

100

disgruntled investors. 313 Ducharme/Malatesta/Sefcik (2004) show that the probability of a lawsuit increases in the level of accruals at the time of the IPO. Gradually (rather than abruptly) reverting the accounting mechanisms may help to decrease this likelihood. Second, lock-up arrangements often prevent owners from selling further shares, thereby maintaining the incentive to present the company in a positive light. Parent firms Mean TAC/TA p-value N Median TAC/TA p-value N

Y-2 - 1.6% 0.0049 98 -0.5% 0.0334 98

Y-1 0.4% 0.3963 104 -0.1% 0.7686 104

Y0 -0.3% 0.6465 107 -0.2% 0.1759 107

Y+I -0.5% 0.2261 107 -0.4% 0.0818 107

Y+2 0.0% 0.9832 108 0.1% 0.6304 108

Y+3 0.6% 0.1304 107 0.1% 1.0000 107

Subsidiary firms Mean TAC/TA p-value N Median TAC/TA p-value N

Y-2 0.3% 0.6898 22 0.3% 0.5235 22

Y-1 1.1% 0.3377 31 0.1% 1.0000 31

Y0 2.3% 0.0757 42 0.1% 0.8776 42

Y+I -2.5% 0.0153 60 -1.1% 0.0273 60

Y+2 - 1.9% 0.1078 64 -0.5% 0.1686 64

Y+3 1.0% 0.1424 68 0.1% 1.0000 68

Table 14: Total accruals/total assets by relative event year

There is thus some support for the idea that subsidiary firms intentionally manage eamings to appeal to potential investors. A further analysis of the issue, using more detailed data and more sophisticated measures of total accruals, would be desirable to check the robustness of this finding. Rangan (1998) uses quarterly data to more accurately time the occurrence of eamings management. Marquardt/Wiedman (2004) find that firms issuing equity are more likely to manage earnings upward by accelerating revenue recognition. 314 Collecting the data required to test whether these findings also hold in the case of ECOs may be cumbersome, but may yield some further insights into how earnings are managed in this concrete setting. The present study does not attempt to explicitly distinguish between the two potential causes for abnormal operating performance (earnings management and market timing). Rangan (1998), in his analysis of SEOs, uses a simultaneous equation model to differentiate between the two explanations, and finds evidence for both. Given the crude measure of earnings management used in the present study and the lack of quarterly data, a similar analysis in the context of ECOs is left for future research. While this section presents some evidence for the existence of earnings management, chapter 5 on the long-term price performance will show that there is some evidence of managers 313 See Teoh/Welch/Wong (1998), p. 1936. 314 See Marquart/Wiedman (2004), p. 461.

101

being able to time the ECO to occur after a positive price development of the parent firm. Assuming this timing capability refers not only to assessing market valuation levels, but also operating performance levels, the combined results of the present study find support for both explanations.

4.6.4

Explanation of cross-sectional results for parent firms

Focussing on mean abnormal performance measures may obscure a considerable variation in performance across sample firms. The following section therefore attempts to explain part of the cross-sectional distribution by linking firm performance to explanatory variables in a multiple regression framework. Explanatory variables are based on relevant findings in existing literature, as well as results from the STPP analysis in chapter 3. The dependent variables are winsorised at three standard deviations, analogously to the recommendations by Ang/Zhang (2001), to prevent the impact of outliers. 315 Bollinger/Chandra (2004) find that while winsorizing is a common procedure in econometric studies and usually justified as a means of reducing the impact of measurement error, it is not uncontroversial and may introduce additional biases into the data. They also find that winsorizing is almost always better than truncating 316, and winsorizing at a 1% level (corresponding to approx. +/- three standard deviations in a normal distribution) is preferable to winsorizing at a 5% level (corresponding to approx. +/-2 standard deviations). 317

4.6. 4.1 Independent variables Stake retained in subsidiary firm: Boone/Haushalter/Mikkelson (2003) find that parent firm profitability improves only when the entire stake in the subsidiary firm is carved out. The impact of the remaining stake on operating performance levels (i.e., growth rates) does not yet seem to have been analysed in the literature. Also, Boone/Haushalter/Mikkelson (2003) only take into consideration the remaining stake immediately following the ECO. In contrast, this analysis identifies the remaining stake held by the parent firm in each of the three years following the ECO. 318 The advantage of this measurement, as opposed to using a single data point from the time of the ECO, is that it takes into account changes in ownership in the years following the event. Such changes may have been planned and announced at the time of the event, but not carried out until a later point in time, maybe due to market capacity reasons. Theoretical arguments can be found for both a positive and a negative relationship between operating performance and the remaining stake. It may be positive because the parent 315 See Ang/Zhang (2004), p. 763. 316 Winsorizing recodes values exceeding a certain threshold with the threshold value. Truncating (or trimming) removes these values. 317 See Bollinger/Chandra (2004), p. 13. 318 The required data is manually retrieved from parent and subsidiary firm annual reports.

102

firm's interest in the subsidiary firm is likely to increase with stake retained. Consequently the parent firm is more likely to exercise an efficient monitoring function from which the subsidiary firm may profit, as well as the parent firm itself through its remaining stake. For example, Monsen/Chiu/Cooley (1968) find that owner-controlled firms have a significantly higher return on equity than manager-controlled firms 3~9, and explain this result by differing incentive systems for managers and for owners (i.e., an incentive alignment hypothesis when owners are also managers). 32~ On the other hand, the relationship between operating performance and stake retained can be negative: A lower stake leads to a higher level of independence for the subsidiary firm, thereby potentially increasing the latter's flexibility and helping to establish a closer alignment between shareholders' and managers' interests by avoiding entrenchment. For example, Hirschey (1999) finds that profitability of commercial banks is negatively related to managerial stock ownership. TM The relationship between operating performance and remaining stake may also be non-linear: Morck/Shleifer/Vishny (1988) find that profitability increases significantly when a firm's board holds between 0% and 5%, and decreases significantly when the ownership is between 5% and 25%, and explain their results by a combination of the incentive alignment and entrenchment hypotheses. The present study, by using the remaining stake as an independent metric variable in a multivariate regression, abstracts from this possibility and assumes a linear relationship.

Financial distress: Madura/Nixon (2002) find that distressed parent firms on average experience a negative abnormal long-term price performance. Distress is measured by a coverage ratio, defined as EBIT divided by total interest expense. A dummy variable equalling 1 is assigned to all firms whose coverage ratio is below one. The economic intuition behind the variable is that if a company is not able to service its interest obligations out of its operating profit, then the firm is likely to be in some state of financial distress. There are at least two conceptual alternatives to defining distressed companies. First, distress could be assessed based on poor stock price performance. However, Asquith/Gertner/Scharfstein (1994) argue that using interest coverage ratios as a measure of financial distress is more suitable than using filters based on share price performance: Since market prices may contain the market's assessment of the likelihood of bankruptcy as well as of the costs of financial distress, "a sample based on poor stock market returns could be biased to firms with relatively costly financial distress ''322. Second, a measure of the probability of future bankruptcy could be used. The prototype 319

Monsen/Chiu/Cooley (1968) define an owner-controlled firm as a firm in which a blockholder owns at least 10% and shows signs of active management, or 20% without such signs. In a managercontrolled firm there exists no block ownership of larger than 5%. See Monsen/Chiu/Cooley (1968), p. 437-438. 320 See Monsen/Chiu/Cooley(1968), p. 441-442. 321 However, Hirschey (1999) points out that high managerial stock ownership is typically found in smaller banks. Underperformance by closely-held banks could therefore be due to size and potential diseconomies of small scale operation (see Hirschey (1999) p. 213). 322 Asquith/Gertner/Scharfstein (1994), p. 628, footnote 2.

103

is the Z-score model pioneered by Altman (1968), based on a linear discriminant analysis. In its wake a series of other models have been developed, including more refined generalised linear and neural network models. 323 While these models are likely to produce more accurate forecasts of future financial bankruptcy, they are also more costly to implement. For the purpose of this study, it therefore seems appropriate to use the dummy approach suggested by Asquith/Gertner/Scharfstein (1994), which has found widespread application in empirical literature. 324 The hypothesis is that, similar to long-term price performance, long-term operating performance will be worse for parent firms which are in a state of financial distress at the time of the ECO. The expected sign of the coefficient is thus negative.

Motivation: In the analysis on STPP, it was found that when companies state either focussing of the parent firm or development of the subsidiary's business as motivations for the ECO, the market reacts more positively to the ECO announcement. Conceivably, these two motivations may have different effects on the operating level measures and operating profitability measures, respectively: The hypothesis is that the motivation to focus on the parent business is more likely to lead to a positive development in the parent firm's operating profitability measures, whereas the motivation to develop the subsidiary's business is more likely to lead to a positive development in the subsidiary firm's operating level measures. The motivation dummy from the STPP analysis is therefore split into two separate dummies, each equalling 1 if the company at the time of the event announces that its intention is to increase the focus of the parent's business (motivation dummy 1), or to develop the subsidiary's business (motivation dummy 2), respectively. Size: The impact of the separation between parent and subsidiary firms on the performance of the parent firm could potentially increase in the size of the ECO. Therefore a control variable measuring the transaction's relative size, defined as the ratio of subsidiary to parent firm market value of equity at the time of the IPO, is used. 325 The coefficient is expected to have a positive sign: The more significant the transaction, the more should operating performance be positively impacted (assuming that ECOs have positive value consequences as indicated by the results in the STPP analysis). Industry: Cross-industry ECOs have been found to produce higher announcement period returns than same-industry E G O s . 326 Analogously to Vijh (2002), a dummy variable is created, equalling 1 if parent and subsidiary firms are from the same industry, and 0 otherwise. The expected relationship between the dummy variable and

323 See Altman/Narayanan (1997) for a survey of bankruptcy prediction models. 324 Other studies employing a coverage ratio-based dummy approach include Wagner (2004), p. 18; Hovakimian/Titman (2003), p. 4; and Molina (2005), p. 1439. 325 Madura/Nixon (2002) use the same size variable. 326 See Vijh (2002), Elsas/L6ffler (2005), and also the results in chapter 3 in the present study.

104

LTOP is negative: The break-up of a cross-industry parent/subsidiary firm combination is expected to lead to the disappearance of a larger amount of negative synergies.

Number of business segments: In the STPP analysis, it was found that firms with a higher number of business segments tend to experience more positive announcement period returns. The number of business segments hence could also be linked to operating performance. The relationship, however, is ambiguous: On the one hand, more diversified firms (i.e., firms with a higher number of business segments) stand to benefit more from a separation of business segments than less diversified companies. On the other hand, the higher the number of business segments, the lower on average the relative impact of the separation of one of these segments on the remaining parent firm. A higher number of business segments could also indicate that each segment is more closely tied into the conglomerate, and has more links towards other business segments. The sign of the regression coefficient is therefore undetermined.

Abnormal accruals: Firms engaging in earnings management could have negative growth rates in the years following the ECO as the accounting measures used to produce positively biased results in the time prior to the ECO are reverted. Roosenboom/Goot/Mertens (2003), using a sample of Dutch IPOs, find that companies with a high level of discretionary accruals at the time of the IPO have a worse share price performance than companies with lower levels of discretionary accruals. Hence, if abnormal accruals proxy for earnings management, the regression coefficient should have a negative sign: The more a firm has previously engaged in earnings management, the worse its operating level measures will develop.

'Hot market' period." As pointed out by Ritter/Welch (2002), long-term price performance may be driven by data from the Internet bubble period. Conceivably this argument may also hold for long-term operating performance: Firms carved out in this period may have had a lower quality because investor scrutiny may have been lower as a result of the general stock market hype. Therefore a hot market dummy (equalling 1 when the event occurred between 1998 and 2000, and 0 otherwise) is included in the regression. Region: In the analysis of STPP, it was found that ECOs carried out in countries with higher shareholder rights on average earn higher announcement period returns than ECOs in countries with lower shareholder rights. The same variable is included here to test whether the institutional setting of the country in which the ECO occurs is related to the operating performance development. The choice of which dependent variable to use is arbitrary to a certain extent, given five different benchmarks, four different profitability and four different level measures. It seems sensible to use at least one profitability measure and one level measure as dependent variables, to be able to explain both changes in the profitability of operations,

105

as well as growth of operations. The profitability measure chosen is ROS, and the level measure chosen is sales growth. 327

4.6. 4.2

Growth as dependent variable

Appendix 32 shows the results of a series of models aiming to explain parent firm sales growth across one, two and three years following the event. The most striking result is the statistical significance of the coefficient on the remaining stake variable: It is significantly positive across all models. This implies that parent firms retaining a higher stake in the subsidiary firm are profiting from this in the long term in the form of stronger growth rates. This makes intuitive sense because if carved-out subsidiary firms are the high-growth segments of a conglomerate 32s, then retaining a higher stake allows the parent firm to grow more than retaining a lower stake. 329 Also, the higher stake may allow the parent firm to combine the positive consequences of continued monitoring with the increased flexibility the subsidiary firm enjoys as a result of the ECO. Determining the optimal stake to be carved out hence involves a trade-off between these two effects. The coefficient of the distress variable has the expected negative sign in all models and is significant in some, confirming the finding by Madura/Nixon (2002) regarding longterm price performance for the case of long-term operating performance: Parent firms which are financially distressed before the ECO face limited growth prospects after the ECO. Again the result is intuitive: The descriptive sample statistics show that the mean and median of relative gross proceeds and relative asset size, which are measures of the relative importance of the ECO for the parent firm, amount to 33.0% (8.9%) and 44.8% (8.5%), respectively. 33~ Hence, the significance of the ECO event for the parent firm may be small, and the potentially positive value effects may not suffice to alleviate the parent firm's financial problems. Consequently these financial problems manifest themselves in the form of lower growth rates in the years following the ECO. The coefficient of the industry dummy has a negative sign in all models, and significance is found for the one-year models. This mirrors findings from the STPP analysis: The announcement period return is higher when parent and subsidiary firms are from different industries. Similarly, growth in operating performance is stronger for such cross-industry ECOs. The likely reason for this higher growth is that potential gains from an ECO are higher for cross-industry than for same-industry ECOs: When parent and subsidiary firms have activities in unrelated businesses, synergies between them may be negative, e.g., because management must split its attention between two 327

Key results are robust to the use of the alternative operating performance measures. See Powers (2003), p. 40. 329 Also, parent firms retaining a majority stake in the subsidiary usually are able to fully consolidate the subsidiary, leading to a higher group growth rate compared to situations when a minority stake cannot be fully consolidated. 330 See Table 1 in section 2.1.6 of the present study.

328

106

very different business segments. Also, management may not have an equal amount of knowledge regarding the different business segments. Separating such a combination therefore eliminates a larger amount of negative synergies, compared to when parent and subsidiary firms are active in the same industry. The coefficient on motivation dummy 2 (develop subsidiary business) is positive in all models, and significantly so in all two- and three-year models: Parent firms announcing that they will develop the subsidiary's business grow significantly stronger than parent firms not announcing this motivation. Conversely, a parent firm's announcement that it aims to refocus on its operations does not seem to affect its subsequent growth (as evidenced by the non-significant coefficient on motivation dummy 1). Finally, the coefficient on the number of segments is significantly negative across all models, and significantly so in all one-year models. This implies that, at least in the shorter term, the positive effect of separating out one subsidiary through an ECO is dominated by the negative effect of having multiple business segments in the first place. All remaining variables do not produce significant results: A parent firm's sales growth does not seem to be related to the relative transaction size, to whether the ECO occurred in the 'hot market' period and in a country with higher shareholder rights, and to the level of accruals prior to the ECO.

4.6. 4.3 Profitability as dependent variable Appendix 33 shows the results of a series of models aiming to explain ROS across one, two and three years following the event. The coefficient on the remaining stake held by the parent firm is negative in all models, and significantly so in some of them: Profitability decreases when the stake retained by the parent firm increases. This finding is consistent with the results found for the US by Boone/Haushalter/Mikkelson (2003), and supports the divestiture gains hypothesis: The higher the concentration on the remaining parent firm's business, the higher the gains associated with an ECO. It also supports the popular notion of the advantage of focussing on one's core business: Presumably management capacity is limited, and concentrating one fewer activities leads to more positive results. The coefficient of the distress dummy is significantly positive in all models. This implies that companies in financial distress have fared better as a result of the ECO than companies not in financial distress prior to the event. This result offers an interesting contrast to the analysis using sales growth as the dependent variable, where a negative relation is found between financial distress and operating performance. Firms in financial distress stand to profit relatively more (compared to non-distressed firms) from a change in the corporate structure in profitability terms, potentially because profitability has been low in the first place.

107

The coefficient on the motivation dummy 1 (increase in parent firm focus) is positive in all models, and significantly so in the two-year models. Hence, as hypothesised, parent firms who state that they want to increase the focus of their operations through an ECO manage to improve the profitability of their operations more than firms which do not state a similar motivation. Also, as expected, motivation dummy 2 (development of subsidiary's business), does not have explanatory power for parent firm ROS. None of the remaining variables shows significant coefficients.

4.6.4.4

Interpretation of results

Drawing the results from the two analyses together, an interesting picture emerges as to what constitutes a successful ECO. There are two distinct scenarios in which companies carrying out an ECO develop positively. These scenarios are based on the findings of significant regression coefficients in the explanation of the cross-sectional distribution of measures of operating performance levels (scenario 1) and profitability (scenario 2). The first scenario is a growth story: A parent firm announces that it aims to develop the business of one of its subsidiary firms, by carving out the business and partially listing it on the stock exchange. In order to profit from the subsidiary firm's growth, only a small stake is listed, and the parent firm continues to hold a considerable stake. However, growth of the parent firm is only achieved when the latter has been financially solid in the time before the ECO: Benefits from the ECO are generally not significant enough to outweigh serious financial problems of the parent firm, and growth can only be based on a sound financial basis. The parent firm will also grow stronger if it has a limited total number of business segments: Growing a subsidiary firm's (and thus group) results requires considerable management attention. A high number of business segments distracts this attention, whereas a low number allows the parent firm to concentrate on what it originally set out to achieve. The second scenario is a restructuring story. A parent firm announces that it aims to increase the focus of its operations. This step has been motivated, among others, by the fact that prior to the ECO, the parent firm has been in a state of some financial distress: The parent firm wants to use the ECO to address this critical situation. The profitability of its operations improves more relative to a firm which has not been in financial distress, because the former has more 'room for improvement', i.e., begins from a worse starting position. To achieve an improvement in profitability, a large stake in one of its subsidiary firms is sold to allow the parent firm's management to concentrate on its core business. Thus, there are two scenarios when ECOs are successful, and each scenario has a set of different characteristics. This finding also has implications for the line of research analysing the reasons why companies carry out an ECO: Any such analysis needs to bear the two distinct scenarios in mind.

108

4. 6. 5

Explanation of cross-sectional results for subsidiary firms

As in the analysis of parent finn LTOP, both sales growth and ROS is used as a dependent variable in two sets of models. The independent variables are the same as in the analysis of parent firm LTOP, with the exception the level of abnormal accruals which is not used as a variable: Since the required data comes from the IPO prospectus (not all of which are available), the number of subsidiary finns for which the measure can be calculated is limited, decreasing the available sample size in the multivariate regression considerably if included.

4. 6. 5.1

Growth as dependent variable

Appendix 34 shows various models with sales growth as the dependent variable. The strongest result across all models in all years is the significantly positive coefficient on the motivation dummy 'development of subsidiary finn's business'. This result is intuitive, since it implies that subsidiary firms whose parent firms announce that ECO proceeds are used to develop the subsidiary firm's business will grow stronger than subsidiary firms whose parent finns do not make such an announcement. The coefficient on the number of business segments is significantly negative across all models. This result implies that subsidiary firms carved out of parent firms with more business segments grow slower. The conglomerate nature of the parent firm seems to hamper the subsidiary firm's growth potential, despite its newly-acquired stand alone status. The subsidiary firm may have had ties with other business segments (e.g., shared operations and administrative functions), which may be stronger when there are more segments. Once the subsidiary is separated from the parent firm, it has to 'leam' to be independent, resulting in slower growth relative to benchmark firms. This result confirms the qualitative considerations by Nick (1994), who speculates that successful ECO candidates are subsidiary firms able to conduct their business independently from the parent firm and whose links to other business segments are limited. TM

4.6.5.2 Profitability as dependent variable Appendix 35 shows various models with ROS as the dependent variable. The regression coefficient of the remaining stake variable is positive in all models, and significantly so in some of them. There are two potential explanations for this finding. First, as suggested by Powers (2003) who in his US sample also finds a positive relation between the subsidiary firm's long-term operating performance and the parent firm's remaining stake, this result could be due to informational asymmetries: Management will only sell a large stake in the subsidiary firm if it expects that operating performance will decrease; it will retain a large stake if operating performance is expected to develop

331

See Nick (1994), p. 244.

109

positively. Second, the higher stake retained may increase the motivation for monitoring the subsidiary firm, leading to more positive operating results for the subsidiary firm. The distress dummy is negative in all models, and significantly so in most of them. This result harmonises with the finding by Madura/Nixon (2002) that subsidiary firms carved out of distressed parent firms have a worse share price performance compared to companies carved out of non-distressed parent firms. The result in the present study suggests that this finding also holds for operating performance. Apparently, the distress symptoms of the parent firm also concern the subsidiary firm, and are carved out along with it. If distress is, for example, the ultimate result of bad management practices or other organizational issues, it seems plausible that the same effects will be observed in the subsidiary firm, leading to a lower profitability. The industry dummy is significantly negative in almost all models. This result mirrors the finding from the parent firm LTOP analysis, where it was found that parent firms grow stronger after cross-industry ECOs: Subsidiary firms from cross-industry ECOs also perform better than subsidiary firms from same-industry ECOs. The finding is also consistent with the results from the STPP analysis, which showed that announcement period returns in cross-industry ECOs are higher than in same-industry ECOs. The implication is that negative synergies are higher in cross-industry parent/subsidiary firm combinations, and consequently value gains from their dissolution are higher. The regression coefficient on the size dummy has a positive sign, and is significant in many models. Larger subsidiary firms thus seem to enjoy a better profitability development. This may be an indication that the small-firm effect, which is usually found in studies of long-term price performance, is also relevant in the explanation of long-term operating performance. The sign of the coefficient on the hot market dummy is negative, and significantly so in the one-year models. This result confirms considerations by Ritter/Welch (2002), who argue that long-term price performance may be driven by data from the Internet bubble period, for the case of long-term operating performance. An alternative interpretation is that as external capital markets have become more efficient, parent firms carve out subsidiary firms to reduce opportunity costs associated with an inefficient internal capital market, irrespective of the subsidiary firm's suitability for being a stand-alone company. This explanation is further described in the following section, where results are interpreted holistically.

4.6.5.3

Interpretationof results

Again, as in the case of parent firm LTOP, determinants of the development of subsidiary firm operating level measures and profitability measures are different. The most important driver of subsidiary firm growth is the parent firm's intention to develop the business of the subsidiary firm. The dedication of the parent firm may take various forms, e.g., capital investment, long-term contracts between parent and subsidiary firm,

110

or the transfer of talented management resources to the subsidiary firm. This dedication towards the subsidiary firm may be hampered if the parent firm has too many business segments: Since management capacity is limited, owning a large number of segments is likely to distract firm management from its focus on developing the subsidiary firm. Consequently subsidiary firms carved out from conglomerates with a high number of segments grow slower than subsidiary firms carved-out from conglomerates with fewer segments. Improvements in profitability seem mainly related to the dissolution of negative synergies between cross-industry parent/subsidiary firm combinations: When business segments are active in unrelated industries, the positive effects of a conglomerate (e.g., a shared internal capital market) may be outweighed by its negative effects (e.g., firm management's lack of knowledge regarding the success requirements in different industries). Despite the positive effects of a separation, the subsidiary firm seems to inherit some of the characteristics of the parent firm. In particular, some of the reasons responsible for the financial distress of the parent firm seem to be transferred to the subsidiary firm, as evidenced by a worse development of the profitability of subsidiary firms carved out of financially distressed parent firms. For example, if inadequate management practices or inefficient operational processes cause financial distress, floating a subsidiary firm will not resolve these issues; additional restructuring is required. As pointed out by Heugens/Schenk (2004), organizational restructuring indeed often follows portfolio restructuring. 332 However, the impact of this restructuring may take time to materialise. Consequently operating performance is likely to be worse for companies having to deal with such legacy issues. If the parent firm continues to own a considerable stake in the subsidiary firm after the initial ECO, this stake may motivate the parent firm to efficiently monitor the subsidiary firm's operations. Intuitively, it could be argued that this motivation should be highest when the subsidiary firm is under complete control of the parent firm (i.e., pre-ECO); however, the fact that there is additional public scrutiny of the subsidiary firm resulting from a floated minority stake also decreases the potential for opportunistic behaviour by firm management. For example, a parent firm may have been withdrawing capital from a profitable subsidiary firm to cross-subsidise other business segments for political reasons, while the subsidiary firm could have used that capital for profitable investments. If a minority stake in the subsidiary firm is floated, the parent firm is required to act more in line with value considerations. Hence, the overall impact of the parent firm's influence on the subsidiary firm's operations is more positive even though the parent firm's stake has decreased. It was mentioned that there is an alternative explanation for the negative sign of the hot market dummy variable, in addition to ECOs from the Internet period being less successful than ECOs from other periods: Subsidiary firms from ECOs carried out before 1998 may have fared consistently better than subsidiaries carved out after 1998, 332

See Heugens/Schenk (2004), p. 88.

lll

i.e., including ECOs from 2001-2004. Recall that in the STPP analysis it was found that parent firms announcing an ECO after 1999 have earned higher announcement period returns than parent firms announcing an ECO before 1999. This finding was interpreted as support for the hypothesis forwarded by Khanna/Palepu (2000b) that the relative value of internal capital markets has decreased as external capital markets have become more developed over time 333, resulting in higher opportunity costs related to internal capital markets, and rendering their partial closure through an ECO more attractive for parent firm shareholders. The finding that profitability is higher for ECOs carved out before 1999 is consistent with this hypothesis: In earlier periods, only those subsidiary firms with a clear potential for profitability improvement were carved out, because the value of internal capital markets was higher relative to less developed external capital markets, and subsidiary firms were a part of these valuable internal capital markets. In later periods, as the relative value of internal capital markets decreased and their opportunity costs increased, the value of subsidiary firms as components of internal capital markets also decreased. Hence, subsidiary firms were not only carved out when there was a clear potential for profitability improvement, but also for the sake of partially closing costly internal capital markets. This may have decreased the average profitability improvement. 334 This view suggests that parent firms face a trade-off when deciding whether to carve out a subsidiary firm: On the one hand, subsidiary firms are components of internal capital markets and thus valuable. On the other hand, internal capital markets can be associated with opportunity costs. The result of the trade-off is determined by two factors: The development state of the external capital market, and the potential for profitability improvement of the subsidiary firm as a stand-alone firm. The more the external capital market is developed, and the higher the potential for profitability improvement, the higher the incentive for the parent firm to close down the internal capital market by carving out the subsidiary firm.

4.7

Conclusion

This chapter analyses the long-term operating performance of parent and subsidiary firms in a multi-year window around an ECO. Both parent and subsidiary firms grow stronger and are more profitable than appropriate benchmarks in the year of the ECO, and are less profitable in the year following the ECO. In the second and third year following the ECO profitability and level measures return to normal for parent firms, whereas subsidiary firms continue to grow, indicating that at least part of the positive effects from the ECO may be permanent. The findings are robust to the use of alternative LTOP methodologies. The findings regarding positive abnormal performance with a subsequent reversal trend are consistent with previous analyses on 333 334

See Khanna/Palepu (2000b), p. 281. Replacing the hot market dummy with a period dummy equalling 1 when the ECO occurred after 1998, and 0 otherwise, the multivariate regression also produces a significantly negative coefficient for the dummy. Given that only 17 out of the 178 ECOs occurred in the 2001 to 2004 period, it is thus difficult to distinguish between the two proposed explanations.

112

SEOs. Two explanations are offered. First, both parent and subsidiary firms may be inclined to manage earnings to render themselves more attractive to financial markets. The analysis of total abnormal accruals indicates that this actually occurs. Second, parent firms could time the ECO to coincide with peaks in performance. For parent firms, the finding of no abnormal operating performance in the second and third year following the ECO casts doubt on some of the elements of the divestiture gains hypothesis, according to which the parent firm should experience an improvement in operating performance in the years following the ECO. Also, there is no evidence for a self-selection bias whereby companies engaging in an ECO self-select on a measure of pre-event operating underperformance: None of the benchmarks and methodologies supports the idea of an operating underperformance prior to the ECO. Changes in both level and profitability measures of operating performance are explained through cross-sectional multivariate regressions. The analysis reveals two distinct scenarios when parent firms conduct successful ECOs: In the first scenario (a growth story), a financially non-distressed firm with a low number of business segments aims to develop the business of one of its subsidiary firms, in which it continues to hold a considerable stake after the ECO. In the second scenario (a restructuring story), a financially distressed firm aims to address the critical situation by increasing the focus of its business operations. It does so by selling a considerable stake in one of its subsidiary firms, allowing management to channel its attention on the parent firm's core business. Any future analysis on ECO motivation should bear these two markedly different scenarios in mind. On a subsidiary firm level, the determinants of growth and profitability are also different. Those carve-outs whose parent firms intend to develop the subsidiary firm's business grow stronger than carve-outs whose parent firms do not state this motivation. They also grow stronger when carved out of conglomerates with fewer business segments. This indicates that the subsidiary firm inherits some of the characteristics of its previous existence as part of a conglomerate. Consistent with this notion, subsidiary firms whose parent firms are financially distressed are less profitable than benchmark firms, supporting the idea that part of the distress symptoms are carved out along with the subsidiary firm. Subsidiary firms also become more profitable when the parent firm retains a larger stake in the subsidiary firms (either because of better monitoring, or because of informational asymmetries between parent firm and capital markets), when carved out from parent firms in a different industry (because of the elimination of negative synergies between parent and subsidiary firm), and when carved out before 1999. The latter finding may either be driven by ECOs from the Internet bubble period, when the average quality of ECOs may have been worse than in other periods, or by changing opportunity costs of internal capital markets: If the opportunity costs of internal capital markets have increased as their relative value decreased resulting from more developed external capital markets, parent firms may have decreased their requirements on the suitability of subsidiary firms to be carved out. This in turn will have lowered the average quality of ECOs over the years.

113

5 5.1

Long-term price performance of European equity carve-outs Abstract

This study analyses the abnormal long-term price performance (LTPP) of European parent and subsidiary firms involved in an equity carve-out (ECO) over the 36 months following the event. Abnormal LTPP tends to be negative on average for both parent and subsidiary firms, but assessment of significance depends on the specific methodology applied. Significance is more frequent in equal-weighted schemes, suggesting that underperformance is more pervasive in smaller firms. This result is found using a variety of methodologies and test statistics, including numerous variations of the BHAR methodology and calendar time methods. Subsidiary firms tend to outperform in the first months after the ECO, potentially due to a self-attribution bias by investors, continued earnings management, and supporting share purchases by the ECO firms and the underwriting investment bank. Parent firms outperform benchmarks in the 12 months prior to the event. Also, there is a negative relationship between pre-event and post-event performance, and valuation levels in the subsidiary firm's industry develop more positively over the 36 months preceding the ECO than valuation levels in other industries. All three findings suggest that parent firms time the ECO to occur in periods of high prices and relative valuation levels. Multivariate analyses of the cross-section of LTPP results reveal that there is a strong positive relationship between both a parent firm's LTOP profitability and LTOP level measures on the one hand, and its LTPP on the other hand, implying that companies which have managed to grow profitably also enjoy a positive market valuation. Postevent performance is also partially a correction of the announcement period return, indicating an overreaction by investors to the announcement of an ECO. Subsidiary firms perform better when they are carved out of parent firms which are not financially distressed (indicating the subsidiary firm may inherit some of the causes of financial distress), when arising out of cross-industry ECOs (indicating that the removal of negative synergies is an important driver of share price performance), when they have been profitable prior to the ECO (suggesting that market value is driven by fundamental values), and when carved out outside of the Internet bubble period (because of a lower average subsidiary firm quality).

5.2

Introduction

The characteristics of an ECO raise a question concerning the long-term price performance (LTPP) of parent and subsidiary firms involved in an ECO: While companies going public and companies engaging in an SEO have been repeatedly shown to underperform various benchmarks in the years following the event, companies

114

spun off from their parent firms have been found to outperform various benchmarks. Carved-out subsidiary firms have a dual nature: On the one hand, they represent a means of raising capital (either for the subsidiary or for the parent firm) by selling part of the company. On the other hand, there are a means of separating ownership between parent and subsidiary firm. They thus have aspects both of a 'classical' IPO (associated with a negative abnormal performance) and of a corporate restructuring tool (associated with a positive abnormal performance). Similarly, a parentfirm involved in an ECO can be argued to be similar to a company involved in an SEO and thus be expected to underperform appropriate benchmarks. Altematively, based on the literature discussing divestiture gains as a source of value creation in ECOs, parent firms could be expected to outperform appropriate benchmarks as a result of improvements in the firm's operations following the ECO. Therefore it is a priori unclear what the average direction of any potential abnormal long-term price performance is for both parent and subsidiary firms involved in an ECO, and the question is thus subject to empirical analysis. Previous studies on the LTPP of ECO firms have produced mixed results. The inconsistency is likely to have two causes: First, the two-fold nature of ECOs as both a capital-raising and a corporate restructuring activity may increase the probability of finding contradictory results. Second, the lack of a generally accepted methodology for measuring long-term price performance prevents an adequate cross-study comparison of results. This study adds to the current body of knowledge in at least three ways. First, a series of established testing procedures commonly used for detecting abnormal LTPP is applied to a European sample of ECOs. As specifically pointed out by Lyon/Barber/Tsai (1999), the contested nature of LTPP analysis necessitates the confirmation of previous results on the basis of a different sample. 335 Since most existing studies focus on US ECOs, this study's use of a European ECOs contributes by performing an out-of-sample test. Test procedures include numerous variations of the buy-and-hold-returns (BHAR) methodology, as well as two calendar time portfolio method alternatives. In addition, this study employs an estimator recently developed by Jegadeesh/Karceski (2004) in a simulation setting. The associated test statistic overcomes some of the most severe methodological issues surrounding the use of BHAR in LTPP analyses. Specifically, it incorporates an estimator of the variance-covariance matrix, thus taking into account potential cross-correlation of sample firm returns. To the author's knowledge, this study represents the most complete attempt to analyse LTPP of ECOs. Second, the market timing hypothesis first suggested by Ritter (1991) for IPOs is tested for ECOs using three different tests. First, pre-event performance is analysed to test whether ECOs tend to occur after periods of a positive share price development. Second, the relationship between pre-event and post-event performance is analysed. If positive pre-event performance is followed by negative post-event performance, this would also suggest that parent firms issue equity in their subsidiary firms at times when 335

Lyon/Barber/Tsai (1999), p. 198.

115

share prices are high. Finally, changes in the relative industry valuation levels are analysed to assess whether parent firms time the ECO to occur during periods when valuation levels in the subsidiary firm's industry have developed more positively compared to valuation levels in other industries. Third, long-term post-event price performance of both parent and subsidiary firms is explained as a function of various event- and firm-related characteristics, similar to the preceding analyses of parent and subsidiary firm LTOP. The link between LTPP and LTOP is also analysed: The hypothesis is that in the long run, the market value of a firm should be driven by its fundamental value as proxied by its operating performance. Hence, a positive LTPP should be more likely in cases with a positive LTOP. The market's short-term reaction to an ECO announcement is also used as a potential explanatory variable. A significant regression coefficient implies that markets are not fully able to anticipate the long-term consequences of an ECO: If they overestimate the consequences, this results in a post-event price drift with a sign opposite to the initial reaction; if they underestimate the consequences, this results in a post-event price drift with the same sign as the initial reaction. Behavioural models can accommodate both predictions on the basis of psychological biases. The chapter is organised as follows: Section 5.3 reviews relevant aspects of previous studies. Section 5.4 discusses methodology, and section 5.5 describes the data and the specific analyses. Section 5.6 contains the results of the analysis and their economic interpretation, and section 5.7 concludes.

5.3

Literature review

Literature on LTPP is bountiful, despite the fact that the "analysis of long-run abnormal returns is treacherous ''336. The focus of this review therefore is on empirical studies on those events relevant in the context of this study: IPOs, SEOs, spin-offs, tracking stock, and E C O s . 337 In a seminal paper, Ritter (1991) finds that companies going public underperform sizeand industry-matched benchmark firms in the three years following the IPO by 29.1%.338 Underperformance is more negative for companies going public in years of

Lyon/Barber/Tsai (1999), p. 198. Additional long-term price performance analyses have been performed, among others, in connection with mergers (Asquith (1983), Agrawal/Jaffe/Mandelker (1992)), debt offerings (Spiess/AffleckGraves (1999)), analyst recommendations (Desai/Jain (2004)), bond rating changes (Dichev/Piotroski (2001)), dividend initiations and omissions (Michaely/Thaler/Womack (1995), earnings announcements (Bernard/Thomas (1990)), new exchange listings (Dharan/Ikenberry (1995)), share repurchases in the open market (Ikenberry/Lakonishok/Vermaelen (1995), Mitchell/Stafford (2000)) and as tender offers (Lakonishok/Vermaelen (1990), Mitchell/Stafford (2000)), proxy fights (Ikenberry/Lakonishok (1993)), reverse LBOs (Mian/Rosenfeld (1993)), share issue privatizations (Megginson/Nash/Netter/Schwartz (2000)) and stock splits (Dharan/Ikenberry (1995), Ikenberry/Rankine/Stice (1996)). 338 See Ritter (1991), p. 10.

336

337

116

high IPO volume. He suggests a "windows of opportunities ''339 hypothesis whereby managers possess market timing capabilities and issue shares when they are valued highly relative to other firms' shares by the market. Similar results are confirmed in a series of studies for the US (Loughran/Ritter/Rydqvist (1994), Loughran/Ritter (1995)) and are also confirmed in out-of-sample studies, among others for the UK (Harris (2003)) and for Germany (Ljungqvist (1997), Stehle/Erhardt (2000)). The similarity of SEOs and IPOs leads Spiess/Affieck-Graves (1995) to analyse the LTPP of companies engaging in an SEO, and they document an abnormal negative price performance over the three years following the SEO. They interpret this as support for Ritter's "windows of opportunities ''34~ hypothesis. These results are challenged by studies of SEOs showing lower and insignificant levels of underperformance (Eckbo/Masulis/Norli (2000)) and underperformance being restricted to small firms (Brav/Geczy/Gompers (2000)). However, Jegadeesh (2000) shows that the latter findings are based on misspecified benchmarks. 341 Using factor models and various firm-matching methods he reconfirms the original findings by Spiess/Affieck-Graves (1995) of long-term price underperformance of SEOs, irrespective of size and growth characteristics. These results for SEOs are confirmed in an out-of-sample test by Stehle/Erhardt (2000) for a sample of German SEOs. Clarke/Dunbar/Kahle (2004) find that for secondary offerings 342 there is on average no statistically significant abnormal performance, but for a subsample where sellers are insiders (defined as including CEOs, officers, directors, presidents, founders, and the chairman of the board) abnormal performance is significantly negative. They also find that the operating performance of this subsample decreases, again supporting the "windows of opportunity ''343 hypothesis: Companies seem to time the secondary offering to coincide with a peak in operating performance. Cusatis/Miles/Woolridge (1993) analyse the LTPP of spin-offs and find a positive abnormal LTPP for both parent and subsidiary firms. However, positive performance is limited to cases where the parent and subsidiary firms are taken over subsequent to the event. In contrast, Desai/Jain (1999) find the positive abnormal performance of spinoffs not to be dependent upon the acquisition of parent or subsidiary firm, but rather on whether the parent firm's focus increases following the event (measured either by SIC codes or by a sales-based Herfindahl index) or not. They also document no abnormal returns in the three years prior to the event. In contrast, Michaely/Shaw (1995) find an underperformance in the two years following the event of their sample of spin-offs and ECOs. The contradictory finding regarding spin-offs may be due to the nature of their 339

Ritter (1991), p. 3. Ritter (1991), p. 3. 341 In the case of Eckbo/Masulis/Norli (2000), IPOs were included in the control sample. In the case of Brav/Geczy/Gompers (2000), the factor-model seems misspecified and produces higher retums for large than for small firms, even when not issuing seasonedequity. 342 In a secondary offerings already existing shares are sold. In a primary (seasoned equity) offering newly created shares are sold. 343 Ritter (1991), p. 3. 340

117

sample (consisting of master limited partnerships), as well as their choice of benchmark (value-weighted index, no control firms), which also puts their findings regarding ECOs into question. Veld/Veld-Merkoulova (2004) analyse a European sample of spin-offs and find no abnormal LTPP. However, they indicate that their sample contains 18 spinoffs in Germany in the 1987-2000 period, which may imply that they also include ECOs in their sample. TM For tracking stock, Billett/Vijh (2004) find an underperformance relative to various benchmarks in the three years following the event, and also an underperformance in the year prior to the event. Combined parent firm and tracking stocks, however, show no abnormal performance following the event. Anslinger/Carey/Fink/Gagnon (1997) find that one, two and three year returns of ECO subsidiary firms on average outperform the Russell 2000 index. 345 Parent firms repeatedly carving out subsidiary firms also outperform the index. 346 The median performance, however, is worse than the index performance. Qualitatively they list value creation from corporate competence centres, HR functions (motivation, compensation, talent retention, succession planning), finance functions (funding for new ventures, new investors) and an increased stock market scrutiny as potential reasons for the positive average performance. Prezas/Tarimcilar/Vasudevan (2000) find that while ECO subsidiary firms and common IPOs have similar price performances in the first year following the respective event, ECO subsidiary firms significantly underperform IPOs on a three-year basis. However, the robustness of these results has been called into question because only a single methodology (control firms) is applied. 347 Vijh (1999), using a US sample of ECOs from the 1981 to 1995 period, finds that carved-out subsidiary firms perform in line with various benchmarks over three years following the ECO. In the 1981 to 1988 period subsidiary firms even outperform some of the benchmarks. Parent firms show no underperformance over the complete 1981 to 1995 sample period. Vijh lists three potential explanations for the fact that there is no systematic underperformance: First, parent firms are less focussed in their respective businesses before the ECO, and the relative increase in focus following the event may have positive value consequences. Second, parent firms continue to exercise a monitoring function over the subsidiary firm by not selling the entire stake. Third, the reputation of the parent firm may prevent it from overpricing its subsidiary firm when selling it to the market. However, all three explanations do not seem to have consistent 344

345

346

347

Bt~hner (2004) lists only two spin-offs in Germany (Hoechst/Celanese and Gehe/Takkt). More recently, Bayer/Lanxess is another example. The Russell 2000 index measures the performance of the 2,000 smallest companies in the Russell 3000 Index, which is made up of 3,000 of the biggest U.S. stocks. The implicit assumption is that ECO parent and subsidiary firms are smaller companies. Anslinger/Carey/Fink/Gagnon (1997) do not mention whether this assumption is explicitly tested, and do not justify their use of this particular index. Anslinger/Carey/Fink/Gagnon (1997) do not report results for parent companies which do not repeatedly carve out. See Schikowsky/Schiereck/VNkle/Voigt (2005), p. 8 and p. 19.

118

statistically significant explanatory power. Vijh (1999) finds some evidence that the number of business segments before the ECO is positively related to the long-term price performance, supporting the divestiture gains hypothesis: Subsidiary firms carved out of many-segment conglomerates profit relatively more from their new independence, compared to subsidiary firms carved out of conglomerates with fewer segments. Madura/Nixon (2002) find that both parent and subsidiary firms on average show a negative LTPP. However, analysing the cross-section of their results they find that negative performance is exclusive to distressed parent firms, and to subsidiary firms carved out of distressed parent firms. Madura/Nixon (2002) argue that controlling for financial distress is necessary because a parent firm may be motivated to engage in an ECO to improve its liquidity. To assess whether a parent firm has been in a state of financial distress prior to the ECO, they employ a dummy variable equalling one if the interest coverage ratio is less than one. 348 Their finding suggests that the parent firm transfers some of its distress symptoms onto the subsidiary firm. Powers (2003) finds a positive LTPP for subsidiary firms in his sample of 181 US ECOs from the 1981 to 1996 period in the three years after the event, but positive overall performance is mainly due to positive first year performance. In the second and third year after the ECO, subsidiary firms underperform the benchmark (consisting of control portfolios of size and BTM-matched firms). Powers (2003) also finds that LTPP is positively related to the stake retained: The higher the stake retained by the parent firm after the initial ECO, the higher is the share price performance of the subsidiary firm in the following years. He interprets this finding as support for the hypothesis that a parent firm will only sell a large stake if it views the market's assessment of the subsidiary firm as very positive, and anticipates that this assessment may mean-revert in the future. Conversely, if the parent firm views the market's assessment of the subsidiary firm as negative, only a small stake is sold. In contrast, Annema~allon/Goedhard (2002) find that in their sample of 204 ECOs from the 1986 to 1997 period subsidiary firms offering more than 75% of their total shares for public trading outperform the S&P500 index, while subsidiary firms offering less than 75% of their total shares for public trading underperform the S&P500 index in the two years following the event. This result contradicts Powers (2003), because it implies that LTPP is negatively related to the stake retained. Annema/Fallon/Goedhard (2002) argue that subsidiary firms need to gain independence in order to profit from the benefits associated with a stand-alone company (e.g., removal of negative synergies between unrelated business segments, better alignment of managers and shareholder interests, increased competitiveness due to more managerial flexibility). 349

348

349

See Madura/Nixon (2002), p. 174. This definition seems to be based on Asquith/Gertner/Scharfstein (1994). See also discussion in section4.6.4.1 on the adequacy of this definition. There are two potential explanations for these contradictory findings: First, the sample used by Annema/Fallon/Goedhard (2002) contains an undisclosed number of European ECOs, making a direct comparison of the two studies difficult. Second, the two findings could potentially be harmonized by assuming a non-linear relationship between the subsidiary firm's long-termprice performance and the

119

In a German context, Kaserer/Ahlers (2000) find a negative but statistically not significant abnormal performance of parent firms in the 180 days following the event, using a standard event study methodology with a sample of 23 German companies. Langenbach (2001) finds a negative abnormal performance of subsidiary firms relative to a market index in the three years following the event for a sample of 28 German companies. Schikowsky/Schiereck/V61kle/Voigt (2005) compare the performance of ECO subsidiary firms to common IPOs, using a variety of benchmarks and test statistics. They find some indication of an underperformance of ECO subsidiary firms over a two-year period following the event when using equal-weighted observations. However, significance of results disappears when using value-weighted observations, a result confirmed when they apply the calendar time method based on the Fama-French (1993) three-factor model. They caution researchers to use a variety of methods to ensure robustness of results. If underperformance of capital-raising activities such as IPOs and SEOs is not only an artefact associated with low-quality statistical test methodology (as suggested, e.g., by Brav/Geczy/Gompers (2000)), the question is: Why should IPOs and SEOs be associated with negative abnormal LTPP? Miller (1977) argues that heterogeneous expectations by investors will result in only optimistic investors buying IPOs. Pessimistic investors will not be able to short the IPO due to corresponding short-selling restrictions, and initial IPO prices will therefore be above their equilibrium. As more information becomes available in the time after the IPO, the optimistic investors will correct their expectations, leading to lower returns. Baker/Wurgler (2000) and Hirshleifer (2001) argue that market values regularly deviate from fundamental values. Firms exploit these misvaluations by issuing overvalued equity. These considerations form the behaviouralist fundament for the market timing hypothesis already brought forward by Ritter (1991). Similarly, Schultz (2003) argues that companies issue equity when prices are high, and consequently there are IPO waves around (ex-post) price peaks. As prices revert from their highs, the performance of IPOs becomes negative. Overall, there are some indications that while capital-raising activities such as IPOs, SEOs and secondary offerings tend to lead to underperformance of firms relative to appropriate benchmarks, corporate restructuring activities such as spin-offs tend to lead to an outperformance in the years following the event. Parent firms seem to time capital-raising activities to coincide with periods of high share prices. Results on the LTPP of ECOs are mixed, potentially for two reasons: First, the dual nature of ECOs as both a capital-raising and a corporate restructuring mechanism could lead to ambiguous results. Second, the disagreement about what constitutes an appropriate methodology for the analysis of abnormal LTPP increases the probability of finding methoddependent results.

stake retained by the parent firm: Initially the higher the stake retained by the parent the better the long-term price performance, but if the complete stake is sold long-term price performance is also positive.

120

5.4

General methodology

The hunt for anomalies has been an increasingly popular activity by academics since the early 1980s. 35~ This has been true despite both the widespread popularity of the notion of at least semi-strong informationally efficient capital markets TM, and the methodological challenges associated with LTPP analysis. The semi-strong form of the efficiency market hypothesis (EMH) is often studied in the context of the adjustment of prices to certain events. The question is simple: Do markets incorporate the news regarding an event with no time delay, or is there a post-event price drift? Assuming, as EMH does, a random arrival of new information to the markets, any systematic postevent price drift would be a violation of the semi-strong form of market efficiency. To ascertain the existence of post-event price drifts, one has to specify what the price development would have been without the event. As this is per definition unobservable, models of expected (or normal) returns are used. The basic idea is to match the firms in the sample to control firms which are subject to the same risk sources, and which are therefore assumed to produce the same return as the sample firms. However, there is currently no consensus about the number and nature of the risk factors involved in the determination of returns, and consequently there exists no model which is able to fully explain the cross-sectional variation of share price returns. 352 Fama (1970) is the first to describe this 'bad-model' or joint-hypothesis problem. It alone could suffice to argue that tests of long-term price analysis are not a sensible use of time and resources. In a review of various long-term return studies, Fama (1998) reproaches many papers for only casting doubt on market efficiency, and not producing viable alternatives. 353 According to Kuhn (1970), however, it is a common feature of science history that existing paradigms (here: market efficiency) are subject to an increasing number of anomalies. 354 These accumulate and build up into a crisis, followed by a revolution and a new paradigm. Consequently, the attempt to identify anomalies is a worthwhile endeavour in the development process of any science, and the search for these potential anomalies is a catalyst for future knowledge. The three key methodological questions to address in a LTPP analysis are How are returns calculated (as compound annual returns (CRs), as buy-and-hold returns (BHARs), or via the use of factor models)? What benchmark is used (market indices, portfolios of matched companies, or single matched companies?), and what are the matching criteria for selecting the benchmark portfolios and companies (size, book-to-market ratio (BTM), industry, or beta?) 350

351 352 353 354

Banz's (1981) paper on the small firm effect - an outperformance by a portfolio of small firms by about 1% per month for the period from 1931 to 1975 - was a seminal paper in the capital markets anomalies literature. See Dimson/Mussavian (1998), p. 101. See Fama (1998), p. 293. See Fama (1998), p. 284. See section 2.3.3.

121

How is statistical significance assessed (standard t-statistic, various amended t-statistics, using tabulated or empirically generated ('bootstrapped') critical values)? The following paragraphs review the literature relevant to these methodological issues.

Barber/Lyon (1997) prefer BHARs to CRs because CRs are biased predictors of BHARs (referred to as measurement bias). 355 For an individual security i, the BHAR is defined as 356

T

ggAgiT = ~-I (1 -+- Nit)- U t=l

(1 nt-

E(git ))

t=l

where E(Rit) is the expected return on security i at time t, measured as the return on some appropriately defined benchmark. Barber/Lyon (1997) find that reference portfolios used as benchmarks lead to misspecified test statistics, and they identify three biases responsible for this357: First, the (positive) new listing bias arises because reference portfolios include companies recently gone public. If these underperform as shown by Ritter (1991), there will be a negative bias in the reference portfolios, and consequently a positive bias in the abnormal return calculations. Second, the (negative) rebalancing bias arises because using an equal-weighted reference portfolio implicitly assumes regular rebalancing (since equal-weighted reference portfolios assume an equal money amount invested in each security, they must be rebalanced at the end of each period by partly selling securities with a positive performance in the last period, and by adding securities with a negative performance, to re-establish equal money weights). If there is a mean reversion effect of share prices 358, this leads to future good performers being added and future bad performers being removed from the portfolio, leading to a positive bias in the reference portfolio, and consequently a negative bias in the abnormal return calculations. Third, the (negative) skewness bias arises because individual share returns are more volatile than market returns, leading to an inflated estimate of the standard deviation, and thus to negatively biased test statistics. Since CRs are most affected by the new listing bias and BHARs are most affected by rebalancing and skewness bias, CRs tend to be positively biased and BHARs tend to be negatively biased. Barber/Lyon (1997) suggest using a matched-firm approach where single control firms are matched on size and BTM. In their simulation study they find that this method yields well-specified test statistics because it alleviates the three biases. 359 They 355 356 357 358 359

The sign of the bias depends on the actual return levels and is positive (negative) if BHAR is less than or equal to 0 (greater than 0). See Barber/Lyon (1997), p. 345. See Barber/Lyon (1997), p. 344. I.e., empirical rejection rates systematically deviate from theoretical rejection rates. As evidenced by DeBondt/Thaler (1985). See Barber/Lyon (1997), p. 343. Since the single control firm approach cannot cancel the measurement bias of CR, BHAR are preferable in general.

122

also test the Fama-French (1993) three-factor model, and find that while it has the advantage of not requiring size or BTM data on sample firms, it has the disadvantage of assuming that the regression estimates are stable over the estimation period. Kothari/Warner (1997) identify various sources for misspecification of test statistics. First, the degree of misspecification does not seem to depend on the model used to generate normal returns. Second, issues concerning the survival of companies (minimum data requirements leading to a pre-event survival bias; time-varying parameters influenced by whether a company survives the period or not; and the question of how to deal with firms that do not survive the period, i.e., 'drop-outs') can lead to misspecification of test statistics. 36~ Third, commonly used variance estimation methods underestimate the true variance, leading to upward-biased test statistics and therefore to higher rejection rates. Their simulation results find that test statistics increase with the time horizon, and CRs are on average positive for random samples. The latter finding is not driven by skewness, but by a sample-selection bias (similar to Barber/Lyon's (1997) new listing bias), and by calendar time-period effects. As a result, CRs have been rarely used in the past decade. 361 Without testing the procedure, they advocate the use of non-parametric tests such as the bootstrap procedures used by Ikenberry/Lakonishok/Vermaelen (1995). Ikenberry/Lakonishok/Vermaelen (1995) combine both CRs and BHARs with various benchmarks (market index, size-matched portfolios, and size/BTM-matched portfolios). However, unlike previous studies, they base their inference not on conventional tstatistics but on empirically generated cut-off points from bootstrapped portfolios: For each firm in their sample a firm matched on size and BTM is randomly selected from the Compustat universe and placed into a pseudo-portfolio. For this pseudo-portfolio LTPP is calculated, and the same procedure is repeated 1,000 times. The actual abnormal performance of the sample is compared to the empirically generated distribution of abnormal performances of the pseudo-samples. Lyon/Barber/Tsai (1999) suggest using carefully constructed reference portfolios free of the new listing and rebalancing biases, matched on size and BTM. Skewness is controlled for in two ways: First, by using skewness-adjusted t-statistics whose distribution is generated via a bootstrap procedure; and second, by determining empirical p-values from simulated distributions similar to Ikenberry/Lakonishok/Vermaelen (1995). Both methods, however, are unable to control for the cross-sectional dependence in sample observations arising due to the overlap of event windows. An alternative to the BHAR method is the calendar time portfolio method as suggested by Fama (1998). 362 All companies with an event in the last n (often n=36) months are 360 See Kothari/Warner (1997), p. 303-305. 361 See Ang/Zhang (2004), p. 254, and Kothari/Warner (1997), p. 308. 362 See Fama (1998), p. 295-296. In addition to offering a solution for the cross-sectional dependence problem, Fama (1998) favours the use of CR vs. BHAR for three reasons: First, tests for abnormal

123

placed into a portfolio. Abnormal returns are calculated for each firm in the portfolio using the matching firm or matching portfolio approach, and are averaged across all firms in each month. The effects of the return correlation across event firms are captured by the time-series variation o f the monthly abnormal portfolio returns. The (equally or value-weighted) returns on the portfolio are then regressed on the factors o f the Fama-French three-factor model. If the factor model is a valid method o f explaining the cross-section o f share price returns, and if there is no average abnormal portfolio return, the three factors should explain the variation in portfolio returns and therefore produce an intercept (a or 130) o f 0. If the intercept is significantly different from 0, its value is interpreted as the average monthly abnormal portfolio return. There are two drawbacks o f this approach: First, unlike BHARs, the abnormal return measure does not directly reflect investor experience. 363 Second, test statistics based on the calendar time method are generally well-specified only in non-random samples. They do not perform well when samples consist o f firms of non-average size, BTM, or in cases o f industry clustering. This is a consequence of the joint hypothesis/bad model problem highlighted by Fama (1970): If the model for producing normal returns does not explain the full cross-section o f share price returns, abnormal return calculations will be biased. Apparently, the Fama-French three-factor model is at least partially subject to this shortcoming. As a consequence, Lyon/Barber/Tsai (1999) recommend using both approaches due to advantages and disadvantages o f each individual approach. 364 Alternatives to dealing with the cross-sectional dependence problem have not caught on in empirical research. 365 Mitchell/Stafford (2000) also advocate the use of the calendar time portfolio method, and reject the bootstrapping of empirical abnormal return distributions because o f the violated assumption o f event-firm abnormal returns being independent (i.e., not clustered in time). 366 Taking into consideration that the Fama-French three-factor model

363 364

365

366

returns should use the same return metric as the model of expected returns, which are mostly shorter time intervals. Second, BHAR can imply abnormal returns for years in which there have been none. Third, CRs avoid problems (e.g., skewness) produced by compounding monthly returns. See Lyon/Barber/Tsai (1999), p. 198. Lyon/Barber/Tsai (1999) also suggest an alternative calendar time method, which is based on calculating mean monthly abnormal returns of the portfolio companies for each calendar month, and assessing their significance by a standard t-statistic (see p. 194-197). Brav (1997) suggests a Bayesian-based approach in which a model is fitted to observed firm residuals. Subsequently long-term returns are simulated taking into account the estimated residual variations and covariations. Similar to the pseudo-portfolio simulation approach, this is repeated, and the simulated averages are used to construct the null distribution of the sample mean. Statistical inference is based on a comparison of actual abnormal returns and the simulated distribution. However, Fama (1998) shows that in addition to the complexity of this approach, in practical situations a solution is not typically available because the number of covariances to be estimated is greater than the number of time-series observations. Also, Lyon/Barber/Tsai (1999) find that while this approach reduces misspecification, it does not eliminate it. As they point out (see Mitchell/Stafford (2000), p. 290-292), a large sample size may lead to a normally distributed sample, but not necessarily to an independently distributed one. Accounting for cross-correlation of abnormal returns, they show that t-statistics fall dramatically (resulting in a 3-year BHAR of 15% not being statistically significant). After accounting for cross-correlation, results for both methods are similar.

124

cannot explain the whole cross-section of returns, they refine the abnormal return calculation by determining that part of the intercept which is due to the mispricing caused by the inadequacy of the factor model, rather than the event. For this purpose, they generate 1,000 pseudo-portfolios with one company matching each sample company in size and BTM. They regress each pseudo-portfolio on the factors of the three-factor model, and calculate the average value of the intercept. By deducting this average intercept from the intercept based on the original sample they arrive at an "adjusted intercept ''367, thus partially alleviating the 'bad model' problem. Despite this progress in the use of calendar time methods, the BHAR methodology continues to be popular. Cowan/Sergeant (2001) differentiate two kinds of sample correlations: First, there is a correlation between a stock's return and its benchmark's return (pairwise dependence, inter-sample correlation36S). Second, there are correlations between the various stocks' returns in the sample (cross-sectional dependence, intrasample correlation) resulting from overlapping horizons. Using a standard paired difference t-test only accounts for inter-sample correlation and its positive bias on the sample variance. It does not account for intra-sample correlation and its negative bias on the sample variance. Therefore using a two groups difference of means test which accounts for neither dependence is argued to produce better results than using the standard paired difference t-test, because the two opposite-sign biases partially cancel each other out. They also suggest winsorizing abnormal returns at three standard deviations. In their simulations both methods alleviate but do not completely remedy the skewness bias. Ang/Zhang (2004) find that using Carhart's (1997) four-factor model instead of the Fama-French three-factor model leads to excessive rejection levels, which they attribute to overfitting of the model. They also document very low power for all methods over a time horizon of five years. Results are better when weighted least squares (WLS) rather than ordinary least squares (OLS) are used to estimate the model: WLS is better able to deal with the heteroscedasticity caused by changing portfolio sizes resulting from the addition to and removal of companies from the portfolio. They also test various alternatives of BHAR methods, and find that reference portfolios based on size and BTM matched firms produce positively biased abnormal returns, whereas single firm benchmarks are unbiased. They compare combinations of abnormal return calculations and significance test, and find that matching a single benchmark on size, BTM and beta, in conjunction with a standard sign test, produces the most powerful results. Jegadeesh/Karceski (2004) develop a test statistic incorporating an estimator of the variance-covariance matrix, which thus is autocorrelation and heteroscedasticity consistent. Their simulation shows that tests are well-specified in random samples (similar to Lyon/Barber/Tsai (1999)), but also in non-random samples (i.e., with either 367 Mitchell/Stafford (2000), p. 310. 368 The terminology of inter- vs. intra-sample correlation is based on Schikowsky/Schiereck/V61kle/Voigt (2005).

125

industry or event window clustering) where cross-sectional dependence poses a major problem. This allows them to recommend the use of this test statistic in conjunction with BHARs. To summarise, the methodological literature on LTPP is voluminous and leaves the researcher in some doubt as to which method is preferable. BHARs have been popular in the early and mid-1990s as evidenced by a series of studies using this methodology369, probably owing to their intuitive appeal and direct reflection of investor experience. Several methodological papers in the mid-and late-1990s (Barber/Lyon (1997), Kothari/Warner (1997), Lyon/Barber/Tsai (1999)) highlighted the methodological difficulties associated with BHARs, and attention subsequently turned to calendar time methods (Fama (1998), Mitchell/Stafford (2001)). These were seen as alleviating some of the most burdensome statistical issues (intra- and inter-sample correlation) regarding the test statistics used in conjunction with the BHAR methodology. However, in recent years BHARs seem to have regained some of their former appeal, as test statistics have been developed which are also able to partially cope with the statistical issues involved (Cowan/Sergeant (2001), Jegadeesh/Karceski (2004)). At present, both BHARs and calendar time methods seem to stand side by side as two fairly robust methods for the analysis of abnormal LTPP. The key take-way for empiricists therefore is that while admonitions from previous papers 37~ regarding the trickiness of LTPP analysis should certainly be considered when interpreting results, more recent methodological developments establish a pragmatic basis of trust in properly designed analyses.

5.5

Data and specific analyses

Based on the discussion above, the two methods of determining abnormal returns on which at least some level of consensus exists are implemented, with various variations, partially alleviating the joint-hypothesis problem. TM The identification of the ECO sample is described in section 2.1.5. Again, all financial companies are excluded from the analysis. Further companies are dropped because of limited data availability. Because different time horizons are considered, the number of companies varies across the analyses. The benchmark universe is constructed as described in section 4.5. For all benchmark companies, share prices, market value (market capitalisation, share price times shares outstanding) and book value of common shareholder's equity are obtained for the period from June 1983 to June 2005 on a monthly basis from Datastream. Local currencies are converted to Euros using month-end exchange rates also obtained from

369

See Ritter (1991), Cusatis/Miles/Woolridge (1993), Spiess/Affleck-Graves (1995). 37o "There is no standardized methodology" (Kothari/Warner (1997), p. 305), "The interpretation of longhorizon tests requires extreme caution" (Kothari/Warner (1997), p. 337), "Analysis of long-run abnormal returns is treacherous" (Lyon/Barber/Tsai (1999), p. 198). 371 See Vijh (1999), p. 281, and Clarke/Dunbar/Kahle (2004), p. 576, for a similar argument.

126

Datastream. 372 In total this dataset comprises 974,552 firm-month observations of prices and market values, and 764,371 firm-month observations for book equity. The analyses are implemented in Matlab.

5.5.1

BHAR and selection of benchmarks

For the implementation of the B H A R method, three different benchmarks are used: size/BTM matched control portfolios, size/BTM matched single peers, and Ang/Zhang matched single peers. Size/BTM portfolios are constructed similar to Lyon/Barber/Tsai (1999) and Jegadeesh/Karceski (2004). First, ten size deciles are formed by ranking all securities by market value in June of each year t. Second, each size decile is partitioned into five BTM quintiles by ranking all securities of each respective size decile by their B T M value, calculated as the book value of c o m m o n shareholder's equity at the end of the preceding December in t-1, divided by the market value of equity on the same date. This procedure assures that the book value of equity is known to investors in June of year t, when the size sorting takes place, thus preventing a foresight bias. Firms with a negative book value of equity are excluded. This procedure results in 50 size/BTM portfolios. 373 Monthly returns, both equally and value-weighted, are calculated for each portfolio from July t through to June t+ 1 of each year. The benchmark for each sample firm is the return of the portfolio in which the sample firm is placed at the end of June preceding the event. If the sample firm changes the portfolio during the return calculation period, the benchmark portfolio is also changed. When a sample firm gets delisted, returns for this sample firm and for its benchmark are calculated up to the month of delisting. TM Second, size/BTM matched single control firms are used, based on Lyon/Barber/Tsai (1999). To identify matched firms, in June of each year all firms with a market value of between 70% and 130% of each respective sample firm in the country of this sample

372

373

374

This was the relatively easiest way of obtaining this information. While the Thomson ONE Banker Interface allows downloading price and accounting data for many European companies already converted into Euro, this tool does not produce values for delisted securities. The Excel add-in of Datastream, Advanced Office, does yield values for delisted securities, but only in local currency. The latter tool was hence used and currencies converted to Euros. Lyon/Barber/Tsai (1999) and Jegadeesh/Karceski (2004) subdivide the smallest size decile further into five quantiles, resulting in 70 size/BTM portfolios overall. They do this because they create the deciles using NYSE firms, but also place AMEX and NASDAQ firms into the quantiles, where the latter are predominantly smaller firms. Consequently their smallest decile is relatively overpopulated. This is not an issue in this analysis. This is in accordance with Jegadeesh/Karceski (2004), p. 5. Lyon/Barber/Tsai (1999), p. 357-358, replace missing sample firm returns with returns from the corresponding size/BTM portfolio. When size/BTM portfolios are used as benchmarks, the results of either truncating or filling are necessarily the same. When single peers are used as benchmarks, the results of the two alternatives may differ. Lyon/Barber/Tsai (1999) only use equal-weighted portfolios (p. 353). Since this study employs both equally and value-weighted portfolios, filling rather than truncating would double the number of reported means, test statistics and p-values for the various single peer-based methodologies.

127

firm and in its two digits industry are identified. 375 If no matching company is found in the country of the sample firm and in its two digits industry, the industry definition is relaxed to the one digit industry level. If still no matching company is found, all firms with a market value of between 70% and 130% are identified, irrespective of country, in the same two digits industry. If still no matching company is found, the industry criteria is relaxed again to the one digit level. If still no matching company is found, all firms with a market value of between 70% and 130% are identified, irrespective of country and industry. From either selection, the firm with closest BTM in the previous December is chosen. The benchmark for each sample firm is the return of this matched firm. If returns on a matched firm are unavailable during any month of the abnormal return calculation period (e.g., because it gets delisted), returns of the next closest firm (in terms of BTM) are used. If the sample firm is delisted prior to the end of the period under consideration, returns are calculated up to the month of delisting. The number of sample firms used in the matched portfolio and matched single firm methodology decreases due to data restrictions resulting from the criteria described above: A sample firm must have been listed since at least the December preceding the June preceding the event for all required data to be available. The same restrictions are placed on the benchmark companies. When subsidiary firms are analysed, the portfolio assignment is based on the market value and BTM at the end of the month in which the IPO has taken place, where the BTM is calculated as the book equity from the December preceding the IPO date (taken from the IPO prospectus) divided by the market value of equity at the end of the event month. Third, another set of single control firms is identified based on the recommendations by Ang/Zhang (2004). Specifically, all firms from the same size/BTM portfolio as of the June preceding the event are identified. From this selection, the firm with the highest share price correlation in the 24 months preceding the event month is chosen as a matched firm, and its return serves as a benchmark. The implicit assumption is that if a control firm has had returns similar to the returns of a specific sample firm in the past, than returns are likely to be similar in the future. While this seems like a strong assumption, it seems justified if returns are driven by a set of unidentified factors which cannot be explicitly controlled for. Analogously to the algorithm above, if returns on a matched firm are unavailable during any month of the abnormal return calculation period, returns of the next most correlated firm are used. For all three benchmarks, BHARs are calculated both equal- and value-weighted, where weighting is based on the market value of the sample firms in the first month of the time period under consideration. When BHARs are equal- (value-)weighted using the 375

Lyon/Barber/Tsai (1999) in their simulation study only match on market value and BTM. However, they point out that "controlling for firm size and book-to-market ratios does not guarantee wellspecified test statistics" (Lyon/Barber/Tsai (1999), p. 173), and encourage researchers to match on additional characteristics when required. This study therefore adds country (due to the multi-country nature of the sample) and industry (due to the slight overrepresentation of certain industries in the sample, see descriptive statistics in section 2.1.6) as additional matching criteria.

128

size/BTM portfolio as benchmark, the returns on the portfolios are also equal- (value)weighted. 376 In total this results in six average BHARs. All BHAR distributions are winsorised at three standard deviations as suggested by Cowan/Sergeant (2001), thus alleviating potential skewness. Median BHARs based on all three benchmarks are also reported.

5.5.2

BHAR and statistical significance

Assessing statistical significance is of prime concern in LTPP analyses. The nature of the BHAR distribution prevents the unmodified use of conventional t-tests. A series of modifications have been developed in the methodological literature. The following section describes those variations which are implemented in this study. Test methodologies are based on the conventional t-test, the two groups difference of means test (Cowan/Sergeant (2001)), the skewness-adjusted t-test (Lyon/Barber/Tsai (1999)), a heteroscedasticity- and autocorrelation-consistent test statistic (Jegadeesh/Karceski (2004)), and a standard sign test. Conventional t-test: The conventional t-test is used as a reference. The variance in the denominator is the cross-sectional sample variance of the BHARs:

• 2

_

O'BHAR --

2

i=1

n-1

where BHAR i is the buy-and-hold-return of security i, and BHAR is the sample average of the n individual securities' BHAR. The test statistic is

BHAR t ...... tional

4

BHAIO'~R/H

The test statistic is likely to be misspecified because the distribution of the BHARs does not correspond to the distributional assumptions underlying the conventional t-test. In particular, as pointed out by Lyon/Barber/Tsai (1999), the positive skewness of the BHAR distribution leads to negatively biased test statistics. To address this violation of the distributional assumption, significance is also assessed by comparing the calculated t-value to an empirically generated distribution of t-values. 377

376

377

Two more BHAR could be added when value- (equal-)weighted size/BTM portfolios are used to calculate equal- (value-)weighted BHAR. As suggested by Jegadeesh/Karceski (2004), p. 25.

129

The algorithm used to create this distribution is based on the bootstrap-method. It proceeds by creating 1,000 pseudo-portfolios consisting of one matched firm for each firm in the original sample, where matching is based on the size/BTM portfolio assignment of the sample firm in the June preceding the event date. All BHAR calculations as described in section 5.5.1 are repeated for each pseudo-portfolio, and test statistics are calculated for each pseudo-portfolio, resulting in 1,000 t-values. The tvalue from the original sample is compared against this empirically generated distribution of t-values, and is assessed as significant at the a level of significance if it ranks in the upper or lower a/2-percentile of the empirical distribution.

Two groups difference of means test: As pointed out by Cowan/Sergeant (2001), there are two kinds of sample correlations: First, there is a (usually positive) correlation between a stock's return and its benchmark's return (pairwise dependence, inter-sample correlation378). If inter-sample correlation is ignored, the sample standard deviation is overestimated. To see why, consider a hypothetical sample where the buy-and-hold raw return (BHRRi) for each of the n sample firms is different but equal to its respective benchmark firm' s BHRRj (due to inter-sample correlation of 1). Each BHAR = BHRRi BHRRj is therefore 0, as is the standard deviation of the BHAR, assuming that it is calculated taking the correlation into consideration, i.e., using the variance of H R R ~ - BHRRj_ = 0. This is the method used in the standard

differences" cr = i=j=l n

paired difference t-test, which assumes that the observations from the two groups are dependent. However, when the observations are assumed to be independent, i.e., not showing inter-sample correlation, variance is calculated based on the individual variances of sample firms and control firms, and the standard deviation is O'--

~

1 8.ee

i=1 n

+_•

.eej

> 0.

This is the method used in the two groups

i=j n

difference of means test, which assumes that the observations from the two groups are independent. Consequently the sample standard deviation is overestimated if intersample correlation exists but is ignored, leading to the null hypothesis being rejected too rarely. Second, there are correlations between the various stocks' returns in the sample (crosssectional dependence, intra-sample correlation) resulting from overlapping horizons. If intra-sample correlation is ignored, the sample standard deviation will be underestimated. To see why, again consider a hypothetical sample where the BHRRi for each of the n sample firms is equal (due to intra-sample correlation of 1), and where also each respective benchmark firm's BHRRj is equal. Intra-sample correlation thus persists in the BHAR, and sample standard deviation is therefore 0, irrespective of whether it is calculated based on the standard paired difference t-test or the two groups difference of means test. This underestimates the true variance, which is very likely to 378

The terminology of inter- vs. intra-sample correlation is based on Schikowsky/Schiereck/V61kle/Voigt (2005). Cowan/Sergeant (2001) describe the two types of correlations (p. 745-746).

130

be > 0. Consequently the sample standard deviation is underestimated if intra-sample correlation exists but is ignored, leading to the null hypothesis being rejected too often. Because the standard paired difference t-test only accounts for inter-sample correlation and its positive bias on the sample variance, but not for intra-sample correlation and its negative bias on the sample variance, Cowan/Sergeant (2001) suggest using the two groups difference of means test, which accounts for neither of the two correlations. The hope is that the two opposite sign biases cancel each other out. While they admit that this method "admittedly is a crude approach to the issue of dependence ''379, results of their simulation study find that in many sampling situations using the two groups difference of means test produces better specified results than using the standard paired difference t-test. The test statistic for the two groups difference of means test is

BHAR t twogroups

2

0 ~ / g l + O'benchmark

/n

Statistical significance again is assessed either by comparing the calculated t-value to tabulated t-values, or to an empirically generated distribution of t-values, as described above. Skewness-adjusted test: Lyon/Barber/Tsai (1999) suggest using a t-statistic that explicitly accounts for the skewness of the BHAR distribution. The test statistics is calculated as

lys2 1 ) t skew = ~n * S + -~ + -~n y ' where

s S - BHAR~ , and y = i=1 O'BHARt

-BHARt~ FI * O'BHAR 3 t

A distribution of t-values is generated by a bootstrapping procedure, where 1,000 resamples of size n/4 are drawn without replacement from the original sample. Statistical significance is assessed by determining the percentage rank of the sample skewness-adjusted t-statistic within the distribution of bootstrapped skewness-adjusted t-values. This procedure works well in random samples, but may be misspecified in samples either showing industry clustering, or where sample firms have repeated events within the return calculation period. Heteroscedasticity- and autocorrelation-consistent test: Jegadeesh/Karceski (2004) suggest a more direct way of addressing the cross-sectional dependence/intra-sample 379

Cowan/Sergeant (2001), p. 746.

131

correlation issue. As pointed out above, intra-sample correlation will lead to a negative bias in the estimated variance, which increases the test statistic and leads to an overrejection of the null hypothesis. To prevent this, Jegadeesh/Karceski (2004) use a standard error which increases if the event firm returns are correlated, resulting from either industry clustering or overlapping event windows. 38~ Their t-statistic also allows for heteroscedasticity and autocorrelation in returns. Heteroscedasticity may arise because the number of sample firms in each month is likely to differ. Autocorrelation may arise if a sample firm has two or more events within the H-month period under consideration (i.e., the parent firm carves out more than one subsidiary firm within 36 months). TM The t-statistic is based on monthly cohort abnormal returns: For each month t in the sample period, the monthly cohort abnormal return

AR(t,H)

is defined as the

average o f the return-calculation-period abnormal returns o f firms with an event in that month (all firms in a given month are referred to as a monthly cohort):

t

ARi (t,H) if Flt ~1 AR(t,H) = f Lnt ~~ i=1 O, otherwiseJ where

ARi(t,H)

is the abnormal return 382 for sample firm i across the H-month period

(here" 36 months), n is the total number o f sample firms, and nt is the number of sample firms with an event in month t. T is the number o f months in the sample period. The sample average abnormal return

ARsamp~e(H)383 is

calculated as the average of the

T

monthly cohort abnormal returns,

AR(H)= ~--AR(t,H),

weighted by the number o f

t=l

firms in each monthly cohort:

ARsample(H)= w'*AR(H), where w is a vector o f length T, with w, =

n t /'7

The variance o f

AR,ample(H) is

0 -2-ARsampl(eH) = W' VW

380 381

382

383

See Jegadeesh/Karceski (2004), p. 3. Autocorrelation may also arise when returns on individual sample firms show a momentum effect. Jegadeesh/Titman (1993) find evidence for momentum in share prices over 6 to 12 month periods. ARi(t,H)is the difference between the sample firm's buy-and-hold return over the H-month period and the benchmark's buy-and-hold return over the same period. Note that the interpretation of ARsample(H) is equivalent to BHAR as used in the previous sections, i.e., it is simply the average of the abnormal returns of all sample firms. However, the method of arriving at the number is different. The notation is therefore kept different. This also allows the reader an easier cross-comparison with existing literature.

132

where V is variance-covariance matrix of A R ( H ) . It is this inclusion of V in the calculation of the standard variance of the abnormal returns which allows Jegadeesh/Karceski (2004) to account for intra-sample correlation: Intuitively, it is clear that V must reflect correlation among sample firms, and also that the value of the variance increases when sample firms are positively correlated. This increase in the variance is desired to prevent the problem of a negatively biased variance estimate as described above. Jegadeesh/Karceski (2004) suggest two different estimators for V. The first estimator is based on Hansen/Hodrick (1980), and allows for serial correlation of returns, but assumes homoscedasticity. The i,j th element of the matrix is"

1 T 0-2 : -~--nt~l - [A--R(t, = H) ARsample(H)~,ifi "--j n t >0

SC

__ V i , j

--

PJ = ~.,j t=~

( t , g ) * AR(t + j , g )

] , /flO n t +j>0

O, otherwise

where

ARsample(H) is

the sample average abnormal retum, T.,y is the number of times

where month t and month t+j both have at least one event (i.e., nt>O and nt+j>O), 0 -2 is the variance of monthly cohort 36-month average abnormal retums including only months with at least one event, and pj is the estimator of the jth-order serial covariance. For the calculation of each pj at least five observations (i.e., combinations of months with each at least one event) are required. This condition reduces the estimation error in p:. The test statistic is

tsc =

ARsample(O) ~/w' S C _ Vw

tsc thus differs from the conventional test statistic only in the denominator: The standard error used is ~w' S C _ V w instead of ~/0-sH~R 2 / n. The second estimator again allows for serial correlation, but also for heteroscedasticity of monthly returns. The i,j th element of the matrix is:

133

-R(i, H) 2, / f i= j

hsc_ Vi, j =

JAR(i,H)*A R ( j , H ) ,

t

if 1 < [i- Jl

[0, otherwise This estimator is based on White's (1980) heteroscedasticity-consistent estimator, and takes serial covariances into consideration when holding periods overlap (i.e., when two sample firms have an event within 36 months of one another). The test statistic is

ARsampte(H) tHsc = x/w' H S C _ Vw Statistical significance again is assessed either by comparing the calculated t-value to tabulated t-values, or to an empirically generated distribution of t-values. Finally, as suggested by Ang/Zhang (2004), a standard sign test is used to assess significance for their single firm benchmark. The use of a non-parametric test is justified by the likely skewness in abnormal returns as observed Ang/Zhang's (2004) simulation study. In total, these combinations of benchmarks, test statistics and distributions result in 19 statistical methods. Each method (apart from the sign test) can be applied to BHARs calculated either equal- or value-weighted. As pointed out by Brav/Geczy/Gompers (2000), the two weighting schemes can lead to dramatically different results. 384 If the sample contains a small number of large firms, as descriptive statistics for this sample show it does 385, those few firms will determine the average results for the whole sample. Brav/Geczy/Gompers (2000) argue that equal-weighted BHARs should be preferred if the researcher is interested in the managerial implications of abnormal performance, whereas value-weighted BHARs should be preferred if investor wealth considerations are the focus. 386 However, in this specific context, it can be argued that value-weighting does not represent an ex-ante implementable investment strategy because the weights (i.e., the market values) of future ECOs are unknown at the time of any given ECO. Equal-weighting BHARs thus seems more appropriate both from a firm and from an investor's point of view.

384 See Brav/Geczy/Gompers(2000), p. 212. 385 More than 52% of the total market value of all subsidiaries is due to five (out of 120) subsidiaries when analysing abnormal performance over the 12 month period. 386 See Brav/Geczy/Gompers (2000), p. 212.

134

5.5.3

Calendar time portfolio method

The main motivation for using calendar time methods lies in their ability to circumvent the difficulty of assessing statistical significance in the presence of cross-sectionally dependent abnormal retums. First, in each month, portfolios are formed consisting of those sample firms with an event in the last 36 months. Second, both equal- and value-weighted returns are calculated for each month. Third, these retums less the risk-free rate are regressed on the three factors of the Fama-French three-factor model. Any cross-sectional dependence is thus taken into account in the portfolio variance. A negative (positive) intercept Gt signals under-(over)performance, measuring the monthly abnormal return of the event portfolio companies relative to the market. Statistical significance is assessed based on the significance of the intercept. The regression parameters are estimated both by OLS and by WLS, with the number of event firms in the portfolio used as weights. 387 Weighting monthly observations takes account of heteroscedasticity caused by monthly changing portfolio sizes. The factors in the Fama-French three-factor model are meant to mimic common underlying risk factors driving retums. As Fama/French (1993) point out, small companies tend to have lower earnings on assets than large companies, as do high-BTM companies relative to low-BTM-companies. 388 These two relations seem independent and suggest two distinct common underlying risk factors, for which investors demand and receive a risk premium, reflected in the small-minus-big (SMB) and high-minuslow (HML) factors. This is similar to the risk premium associated with market risk reflected in the third factor of the model. In this study, the factors are created similarly to Fama/French (1993). First, all shares are grouped into two size quantiles (big (B) and small (S)) on the basis of their market value in June of each year t of the sample period. Second, all shares are grouped (independently of their size quantile assignment) into three BTM quantiles (low (L), medium (M), high (H)), with cut-off points at 30%, 40% and 30% of all ranked values, respectively, based on their BTM in the December of each year t-1. BTM is defined as described in section 5.5.1. Also, as for the BHAR calculations, firms with negative book equity are excluded from the analysis. Six portfolios (S/L, S/M, S/H, B/L, B/M, B/H) are formed from the intersection of the two independent groupings. Monthly valueweighted returns are calculated from July t until June t+l of each year, when the portfolios are resorted. The SMB factor is calculated as the monthly difference of the simple average retum on the three small-share portfolios and the simple average return on the three big-share portfolios. HML is calculated as the monthly difference of the simple average returns on the two high-BTM-share portfolios and the simple average return on the two low-BTM-share portfolios. The market factor is the difference of the simple average return on all six portfolios, less the risk-free rate of return, where the 387 As suggested by Ang/Zhang (2004), p. 258. 388 See Fama/French (1993), p. 6-10.

135

latter is proxied by the one month Euro/ECU interest rate, which is available from January 4 1995.389 Appendix 36 shows the resulting number of companies in each of the six portfolios in each of the years from 1985 to 2005. Since the three-factor model is known not to be able to fully explain the cross-section of returns across all size and BTM portfolios, the Mitchell/Stafford (2000) adjustment is implemented39~ A pseudo-portfolio is created consisting of one control firm for each sample firm i drawn randomly from sample firm i's assigned portfolio in the month of the event, and the intercept is calculated. This procedure is repeated 1,000 times. The average intercept from the 1,000 regressions is a measure of the three-factor model's incapability to adequately explain the cross-section of returns, and is the level of abnormal performance the sample firms would have experienced irrespective of the event under consideration. It is subtracted from the intercept calculated for the sample firms, and the remaining value ("adjusted intercept ''391) is the abnormal performance due to the event.

5.6

Empirical results

The following section discusses the results of the empirical analyses of post-event (section 5.6.1) and pre-event (section 5.6.2) LTPP for parent firms, and links the crosssectional distribution of results to various independent variables (section 5.6.3). Next, the post-event LTPP for subsidiary firms (section 5.6.4) is analysed, and again explained and interpreted as a function of event and firm characteristics (section 5.6.5).

5.6.1 Post-eventperformance of parent firms Table 15 shows the results of the LTPP analysis for parent firms using the BHAR methodology. Abnormal performance is calculated for 12, 24 and 36 months following the announcement of the ECO. The analysis is limited to a maximum of 36 months for three reasons: First, using periods longer than 36 months reduces the available sample size considerably due to the large number of ECOs announced in 2000. Second, Ang/Zhang (2004) show that most test statistics have very low power over time horizons longer than three years. 392 Third, the severity of the bad-model problem seems to increase with the length of the time horizon. 393 Figure 5 graphs the BHARs across 36 months following the end of the event month.

389

This leads to a slight decrease in sample size due to the non-consideration of sample firms with an event date more than 3 years prior to this date. 390 See Mitchell/Stafford (2000), p. 308-310. 391 Mitchell/Stafford (2000), p. 310. 392 See Ang/Zhang (2004), p.266. 393 See Drobetz/Kammermann/W~lchli(2003), p. 26.

136

Period

Size/BM portfolio (ew) Size/BM portfolio (vw) Size/BM portfolio (median) LBT single peer (ew) LBT single peer (vw) LBT single peer (median) AngZhang single peer (ew) AngZhang single peer (vw) AngZhang single peer (median)

12 m o n t h s

24 m o n t h s

36 m o n t h s

-1.1% 2.1% -10.3% -3.8% 3.1% -1.6% -4.2% -7.3% -8.9%

-5.9% -4.1% -19.2% -10.5% -8.2% -6.1% -13.8% -25.7% -9.1%

-11.3% -0.3% -21.9% -25.0% - 10.2% -22.6% -23.2% -25.4% -16.8%

Table 15: Parent firm BHAR across various periods

5.6.1.1

BHARs across various time periods

Across the 12 months following the end of the event month, the results using size/BTM portfolios and Lyon/Barber/Tsai (1999) (LBT) single control firms as benchmarks are similar: Mean performance is slightly negative when equal-weighting, and slightly positive when value-weighting. The distribution of abnormal returns using size/BTM portfolios is right-skewed (despite winsorizing), as indicated by the median abnormal return being smaller than the equal-weighted average. For LBT single peers, the median is slightly larger than the equal-weighted mean, indicating only minor (left) skewness. The difference to the size/BTM portfolio case is likely due to the fact that individual share prices are more volatile than portfolio prices. Abnormal returns, calculated as the difference between individual sample firm share prices and a portfolio benchmark, will therefore be right-skewed because of the unlimited upside potential for individual share prices. Abnormal returns, calculated as the difference between individual sample firm share prices and a single control firm, will not be right-skewed because the volatility of the sample firms is matched by the volatility of the benchmark firms; hence, the distribution of abnormal returns based on single control firm benchmarks is likely to be wider, but not skewed. Across the 24 and 36 months following the end of the event month, the results using size/BTM portfolios and Lyon/Barber/Tsai (1999) (LBT) single control firms as benchmarks are again similar: Mean performance is negative when equal-weighting, and less negative when value-weighting.

137

Using Ang/Zhang peer firms, both mean and median abnormal performance tend to be more negative than using either size/BTM portfolios or LBT single peers as benchmarks. A potential explanation for this is the exclusion of recently IPOed companies from the benchmark resulting from the requirement of calculating the correlation between sample and benchmark firms over the last 24 months. Assuming there is a negative long-term price performance of IPOs as suggested by Ritter (1991), it can be expected that the mean absolute performance of the Ang/Zhang benchmark is higher (and mean abnormal performance therefore lower) than the mean performance of the other two benchmarks.

Figure 5: Parent firm post-event BHAR

5.6.1.2 Statistical significance of BHARs Appendix 37 shows p-values for all 19 methodologies of assessing statistical significance. For the 12-month BHARs, none of the methods shows significance. For the 24-month period, equal-weighted BHARs are judged to be significantly negative by the standard paired different t-test (both when using tabulated and when using empirically generated test statistic distributions), by the LBT skewness-adjusted test statistic, and by the two Jegadeesh/Karceski (2004) test statistics. P-values for the hscstatistic are higher than for the sc-statistic, confirming Jegadeesh/Karceski's (2004) simulation finding that the sc-statistic may over-reject while the hsc-statistic may underreject. 394 The hsc-statistic is thus more conservative. For the 36-month period equalweighted BHARs, almost all test statistics produce significant p-values. The same pattern holds for the sc- and the hsc-statistic as in the 24-month period. Remarkably, all significant results disappear when value-weighted averages are considered. This is a strong indication that large firms perform better than small firms. Underperformance may thus be due to small sample firms. 394

See Jegadeesh/Karceski (2004), p. 17.

138

5.6.1.3

Calendar time method

The results from the BHAR analysis are now compared with the results of the second methodology used, the calendar time portfolio method. Table 16 shows the number of companies included in the calendar time portfolio method and the reasons why certain companies were excluded. Of the 142 non-financial companies, 18 have to be excluded because they have two or more events within 36 months of one another (i.e., the same parent firm carves out two subsidiary firms within three years). The remaining companies are excluded because of missing data, mostly because companies had not been listed in those months which were used as sorting months for the creation of the annual portfolios. One company is excluded because its event date lies more than 36 months prior to the first date for which the risk free interest rate is available (January 1995). Companies with an event within the 36 months prior to this date are included in the analysis as part of the portfolios formed from January 1995 onward. In total this leaves 110 firms in the analysis. Companies left after substracting...

Original sample Thomson NA Financial companies 36-month overlap Event date NA Negative common equity Company not listed in June preceding event Company listed in June but not in Dec preceding June preceding event Company listed in Dec preceding June preceding event, but no common equity data in that Dec Event/IPO date more than 36 months before first day of EURO interest rate

178 176 142 124 121 120 114 112 111 110

Table 16: Number of sample companies in calendar time method

As a side note, the exclusion of sample firms which carve out two or more subsidiary firms within 36 months is not required in the case of the BHAR methodology: Since the Jegadeesh/Karceski (2004) test statistic explicitly incorporates intra-sample correlation among event firms and autocorrelation resulting from overlapping event windows, multiple event firms need not be deleted. This reveals an advantage of the BHAR methodology over the calendar time methodology which has not been explicitly pointed out in literature: If the specific sample under consideration contains a non-trivial number of firms with multiple events within the return calculation period, using the BHAR methodology in conjunction with an appropriately defined test statistic allows the use of a larger set of companies, whereas the calendar time method requires that multiple event firms be excluded from the analysis. The disadvantage of having to exclude multiple event firms is that, as always, exclusion of sample firms may render the analysis subject to all types of biases. Table 17 shows the results for parent firms using the calendar time portfolio method. Using the unadjusted version with equal weighting and based on OLS regressions yields a monthly underperformance o f - 0 . 7 3 % , which is significantly different from zero (p=0.0515). However, the mean intercept from the 1,000 pseudo-portfolios i s - 0 . 4 4 % ,

139

implying that part of the seeming underperformance is due to model misspecification when using the unadjusted version. Subtracting the mean intercept from the original intercept yields an adjusted intercept/monthly underperformance o f - 0 . 2 9 % , which is not significantly different from 0. Using WLS regressions the adjusted intercept becomes significantly negative. This difference confirms Ang/Zhang's (2004) finding that estimating the regression via WLS yields more powerful test results than O L S . 395 The likely explanation is that the number of firms with an event in the last 36 months differs across the sample period. This causes heteroscedasticity in the time-series of portfolio returns, thus violating one of the assumptions for the application of OLS. WLS deals with heteroscedasticity by weighting each observation by the number of firms in the event month, potentially producing more reliable results. In the case of valueweighted portfolio returns, the significance of the results disappears. This implies that the negative performance is largely due to small sample firms. Results using the calendar time method are thus consistent with the result obtained from the B H A R methodology. Basic factor m o d e l

EW - OLS - Alpha EW - OLS - t stat EW - OLS - p

-0.0073 - 1.9670 0.0515

A v e r a g e of 1,000 pseudo portfolios -0.0044 - 1.4764 0.1424

Adjusted factor model -0.0029 -0.7842 0.4344

EW - WLS - Alpha E W - WLS - t stat E W - WLS - p

-0.0180 -6.1176 0.0000

-0.0078 -2.6978 0.0080

-0.0102 -3.4701 0.0007

V W - OLS - Alpha V W - OLS - t stat V W - OLS - p

-0.0016 -0.5911 0.5556

-0.0007 -0.1130 0.9102

-0.0009 -0.3451 0.7306

V W - WLS - Alpha V W - WLS - t stat V W - WLS - p

-0.0043 -1.5247 0.1299

-0.0014 -0.2626 0.7933

-0.0029 -1.0208 0.3094

Legend: ew = equal-weighted, vw = value-weighted OLS = ordinary least squares, WLS = weighted least squares Table 17: Calendar time method for parent firms

5.6.1.4 Interpretation of results The findings in this section offer some interesting comparisons with previous studies. First, the (method-dependent) result of a negative abnormal performance for parent firms partially contradicts Anslinger/Carey/Fink/Gagnon (1997), who find a positive 395

See Ang/Zhang (2004), p. 255.

140

abnormal performance of parent firms relative to market indices. However, their median abnormal return is also negative, which is consistent with this study. This study largely confirms Vijh's (1999) findings of limited underperformance 396 of US ECOs for a European sample. Vijh (1999) offers three different explanations for no significant underperformance, but none of them is confirmed in his empirical tests. 397 This study suggests an alternative possible explanation for no clear underperformance of ECO parent firms: The long-term positive effects of corporate restructuring (as witnessed by the long-term outperformance of spin-offs (Desai/Jain 1999)), are balanced against the negative effects of a capital-raising activity (as witnessed by long-term underperformance of IPOs (Ritter (1991)) and SEOs (Jegadeesh (2000)), leading to a normal overall performance. The dual nature of ECOs as both a restructuring and a financing mechanism may thus be mirrored in the LTPP. Second, Ritter (1991) finds that underperformance of IPOs is more pronounced for smaller IPOs (with size proxied by gross proceeds from the IPO). 398 Similarly, Brav/Geczy/Gompers (2000) find that IPO underperformance is mainly due to small firms and firms with high BTM. The results in the present study are consistent with these two studies in that underperformance is limited to equal-weighted schemes, and disappears when valueweighting, indicating the negative impact of smaller sample firms. Third, skewness in the abnormal returns distribution seems common in LTPP analyses. Clarke/Dunbar/Kahle (2004) find that the median abnormal performance for firms engaged in an SEO is far more negative than the mean abnormal performance across one, three and five years. Of course, any median-based abnormal underperformance cannot be exploited as a profitable investment strategy because it is not implementable ex-ante. Thus, market efficiency in the sense of Jensen (1978), whose refutation requires the establishment of a profitable trading strategy, is not endangered by a significantly negative median abnormal performance. One of the key insights from this section is that the assessment of parent firm LTPP depends on which weighting scheme is used. When equal-weighting, average BHAR is slightly negative; when value-weighting average BHAR is roughly neutral. This indicates that smaller firms perform relatively worse than larger firms, and is consistent with Brav/Geczy/Gompers's (2000) finding that underperformance in their sample of SEOs is due to smaller firms. Note that this relative worse performance does not imply that small sample firms perform less well than large sample firms, which would be contradictory to the small firm effect as described by Banz (1981), who finds that NYSE firms with a low market value of equity earn returns in excess of expected

396

397

398

Vijh (1999) finds negative abnormal performance, but significance depends on the benchmark used. Only equal-weightedresults are presented. See Vijh (1999), p. 293. First, parents seem to be less focussed in their respective businesses before the ECO, and the relative increase in focus followingthe event may have positive value consequences. Second, parents continue to exercise a monitoring function over the subsidiary by not selling the entire stake. Third, the reputation of the parent may prevent it from overpricing its subsidiary when selling it to the market. See Vijh (1999), p. 305-306. See Ritter (1991), p. 16.

141

returns as predicted by the CAPM. 399 It does imply, however, that the model used to produce normal returns (in this case, size- and BTM-matched benchmark firms) is misspecified for small firms in that it predicts a better performance than is actually achieved. This mirrors findings by Fama/French (1993), who find that their three-factor model mis-prices small-sized low-BTM firms. The results in the present study therefore confirm previous findings indicating that smaller firms are more difficult to price than larger firms. A potential explanation for this finding is that smaller firms are riskier than larger firms; if smaller firms hence tended to go bankrupt more often, there would be a bias in the way abnormal returns are calculated for small firms: While Datastream does give historic share prices for delisted firms (i.e., there is no classical survivorship bias), abnormal return calculations exclude firms which were delisted from the benchmark calculation from the time of the delisting and use the remaining shares to calculate the benchmark's return. Benchmark returns for smaller firms will therefore be upward biased relative to benchmark returns for larger firms (of which fewer are delisted and hence excluded from the analysis), and small firm abnormal returns will be downward biased. Negative performance of parent firms could be the result of market timing: Parent firms issue equity in their subsidiary firm at times when prices are high. The subsequent price reversal leads to a negative performance. To assess whether market timing indeed exists, it would be interesting to know whether parent firms have performed particularly well in the months prior to the ECO announcement. This question is analysed in the next section.

5.6.2 Pre-eventperformance of parent firms The following section analyses the abnormal LTPP of parent firms in the months leading up to the ECO announcement. From an investor's perspective, this may seem less relevant than post-event performance because any potential pre-event abnormal performance cannot be translated into a profitable trading strategy as per definition the event (i.e., the official ECO announcement) is the first time the broad public learns about the impending ECO. 4~176 Still, pre-event LTPP has been analysed for US samples of spin-offs (Desai/Jain (1999)), tracking stock (Billett/Vijh (2004)), and ECOs (Vijh (2002)).

5.6.2.1 Two hypotheses related to pre-event performance The analysis may help shed light on the two following questions: First, do parent firms market time their ECOs to occur in periods of high prices? If so, there should be a 399 400

See Schwert (1983) for a detailed review of the early literature on the small-firm effect. For some firms there may have been unconfirmed rumours in the market before the official announcement of an impending ECO, which on average seem to cause positive abnormal returns. See section 3.6.2.1.

142

systematic pattem of long-term pre-event and long-term post-event performance. DeBondt/Thaler (1985) notice negative serial correlation (i.e., trend reversal) in share prices over three to five years, whereas Jegadeesh/Titman (1993) find a positive serial correlation (i.e., momentum) in share prices over three to twelve months. If prices mean-revert in the longer run, and companies are able to assess their respective relative valuation level, this raises the opportunity for issuing firms to time their capital-raising activities to coincide with peaks in their valuation (or similarly, share price). This is the basis of the "windows of opportunity ''4~ hypothesis brought forward by Ritter (1991) to explain his finding of average underperformance of IPOs. The simple hypothesis to test therefore is that any post-event (negative) abnormal performance is a reversal of a directionally opposed (positive) pre-event abnormal performance. In the analysis of LTOP in chapter 4, it was found that operating performance is abnormally high around the time of the ECO. Two potential explanations - earnings management and market timing - were discussed. Some evidence of earnings management was found, but market timing could not be directly assessed. The present analysis addresses this issue by testing whether ECOs coincide with periods of positive abnormal share price performance. Finding a positive pre-event performance, in combination with the already established neutral to negative post-event performance, will support the idea of firms successfully timing their ECO to occur in phases of relatively high share prices. Second, is there a relationship between pre-event LTPP and the short-term announcement period abnormal return? If companies know that investors on average react positively to an ECO announcement, they could use this tool as a reaction to a negative share price performance aiming to positively surprise investors by carving out a subsidiary firm. Alternatively, positive pre-event LTPP could help to reduce informational asymmetry problems between the selling owners and the buying investors: Firms are able to signal their 'non-lemon' status by referring to positive past returns, and may thus be able to earn more positive returns to the announcement of an intended ECO. These two questions are addressed in the following analyses. Pre-event performance is measured analogously to post-event performance, and across three non-overlapping 12month periods beginning 12, 24 and 36 months prior to the event month, respectively. Figure 6 graphs the BHAR across the 36 months before the beginning of the event month.

401

Ritter (1991), p. 3.

143

Figure 6: Parent firm pre-event BHAR

5.6.2.2

BHARs across various time periods

Table 18 shows the results of the pre-event LTPP analysis for parent firms using the BHAR methodology, and Appendix 38 shows their statistical significances. Abnormal price performance is positive in the 12-month period immediately prior to the event relative to the size/BTM portfolio benchmark. The majority of test statistics finds this positive performance to be significant. In contrast to the results in the post-event performance analysis, significance is found both for equal-weighting and for valueweighting (value-weighted BHARs are again higher than equal-weighted BHARs, indicating that small firms again perform less well than large firms). P-values for the hsc-statistic are again higher than for the sc-statistic, indicating that the hscmethodology is more conservative. Using LBT single peers as benchmarks, the picture is dramatically different: BHARs are much lower, and none of the test statistics produces significant results. Two potential explanations for this difference are provided in the following subsection where results are interpreted. Period

Size/BM portfolio (ew) Size/BM portfolio (vw) Size/BM portfolio (median) LBT single peer (ew) LBT single peer (vw) LBT single peer (median) AngZhang single peer (ew) AngZhang single peer (vw) AngZhang single peer (median)

12 m o n t h s

24 m o n t h s

36 m o n t h s

22.2% 32.8% 7.5% 7.6% 7.4% 6.7% 10.4% 18.4% 12.5%

-5.3% 0.7% -8.7% -1.5% -13.0% -3.3% -4.2% 2.0% -1.3%

3.7% 8.9% -2.8% 0.2% -5.5% -1.5% -2.5% 9.1% -8.0%

Table 18: Pre-event parent firm BHAR across various periods

144

For the 12-month period beginning 24 months prior to the event, abnormal performance is negative, but mostly insignificantly so for the equal-weighted BHARs relative to both portfolios and single control peers, as well as for the value-weighted BHARs relative to portfolios. The only exception is value-weighted abnormal performance relative to LBT single peer firms, which is significantly negative based on a series of test statistics. A closer analysis of the sample shows that this result is driven by a very small number of large companies with negative performance. Excluding the three largest firms from the sample leads to a value-weighted BHAR of approx.-1% (instead o f - 1 3 % as indicated in Table 18). This finding highlights the limitations of this study: Despite using a larger sample than previous (European) ECO studies, the sample is still small and, above all, heterogeneous in terms of firm size: A few large firms dominate the sample. Hence, results (when using value-weighted averages) may be driven by these few large firms. A similar picture emerges for final period under consideration, the 12 months beginning 36 months prior to the event, where abnormal performance is not significant when equal-weighting observations, but significantly positive (relative to size/BTM portfolios) when value-weighting. In addition to a few large firms driving this result, positive abnormal performance relative to portfolios may be driven again (as in the case of the first 12-month period) by higher individual share price volatility and its consequences on the distribution of abnormal returns.

5.6.2.3 Interpretation of BHAR results Why is there positive pre-event performance relative to portfolios but not relative to single peer firms? There are at least two potential explanations. First, as indicated in section 5.4, individual share prices are more volatile than portfolios. Hence, abnormal returns based on portfolios are more likely to produce a small number of extremely positive values than abnormal returns based on single control firms (whose volatility partially neutralises the volatility of the individual sample firms). This explanation is supported by the median portfolio-based BHAR being considerably lower than the averages. Second, positive pre-event performance relative to size/BTM portfolio but not relative to single control firms could be indicative of how market timing functions: Firms will take advantage of high price levels and issue equity (in the subsidiary firm) when markets value them highly relative to some benchmark, e.g., firms in other industries: If investors consider a certain industry to have positive growth prospects, valuation multiples are likely to be high for firms in that industry. This has implications for abnormal performance measured relative to different benchmarks: If performance is measured relative to size/BTM portfolios, sample firms will show positive abnormal performance because the benchmark includes sample firms from other (lower valued) industries. If, on the other hand, performance is measured relative to single control firms, matched on size/BTM and industry, abnormal performance will be much less pronounced, because the control firms are likely to come from the same industry as the

145

respective sample •rm402, and hence have similar valuation levels as the sample firms. The difference in pre-event performance relative to portfolios and single peer firms is a first indication of market timing being related to industry valuation levels. This issue is further analysed in the paragraph on the market timing hypothesis in section 5.6.2.4. Previous studies have analysed the pre-event performance of (exclusively US) parent firms spinning off a subsidiary firm, issuing tracking stock, and carving out a subsidiary firm, respectively. Parent firms spinning off a subsidiary firm on average show no significant pre-event abnormal performance over the 24 and 36 months period to the event (Desai/Jain (1999)). Parent firms issuing tracking stock underperform both market- and size- and book-to-market adjusted benchmarks in the 12 months prior to the announcement by approx. 13.3% and 12.1%, respectively (Billet/Vijh (2004)). On the other hand, US ECOs on average seem to outperform benchmarks by approx. 14.9% in the 12 months preceding the announcement (Vijh (1999)). 4o3 The present study is consistent with the latter result, and confirms pre-event outperformance of a similar magnitude for a European sample. Why does pre-event performance differ for spin-offs, tracking stock, and ECOs? Assuming companies may freely choose between the various corporate restructuring methods, differences in pre-event performance may reflect different sets of economic circumstances companies are facing when choosing the method. When past performance is positive, the portfolio restructuring aspect of the method may be less relevant than the financing considerations associated with an ECO: High prices allow the parent firm to raise cash by selling highly valued equity. Also, a positive past performance may help to overcome informational asymmetry problems between parent firm management and potential investors: If the latter take the parent firm's past performance as indicative of future performance of the subsidiary firm, they may be more willing to buy shares in the subsidiary firm. In contrast, such an information asymmetry is not relevant in the case of spin-offs, because future subsidiary firm shareholders are already parent firm shareholders, receiving shares in the subsidiary firm without having to pay for them. Consequently parent firms do not require a positive abnormal performance before a spin-off because a positive signalling towards investors is not necessary. Also, a company which does not believe that its current share price is high may decide to forego the more costly option of an ECO and simply distribute shares in its subsidiary firm to existing shareholders, i.e., do a spin-off. Tracking stocks, on the other hand, seem to be issued after a period of negative share price performance. Parent firms expect that markets are likely to undervalue the subsidiary firm, and therefore decide to 'pseudo-float' it via a tracking stock structure. This allows the parent firm to meet two aims not achievable with either a spin-off or an ECO: First, the parent firm signals to the market its willingness to establish a more 402

403

They may not, depending on whether at least one control firm of an appropriate size relative to the respective sample firm is available. See section 5.4 for further details regarding the matching procedure. See Vijh (2002), p. 169.

146

market-orientated corporate structure with all the associated benefits (e.g., a management compensation which is more aligned with market-based economic performance of the subsidiary firm). Second, the parent firm stays in legal possession of the subsidiary firm, and prevents the sell-out of its subsidiary firm for less than the subsidiary firm's intrinsic value.

5.6.2.4 Testing the first hypothesis: Market timing To test the hypothesis of whether companies successfully manage to time the ECO with periods of positive abnormal share price performance, the 12, 24 and 36 months postevent BHARs are regressed on the 12-month pre-event BHARs. If the hypothesis is true, regression coefficients should be negative, because a positive pre-event performance should be followed by (i.e., cause) a negative post-event performance. Table 19 shows the coefficients and p-values of the respective OLS regressions. All coefficients for equal- and value-weighted abnormal returns relative to size/BTM portfolios are negative, and significantly so for the 12-month period for both weighting schemes, and for the 24-month period using equal weights. The remaining coefficients are marginally not significant (from p=0.1162 to p-0.1655)). Coefficients using LBT single peers are also negative, and significantly so for the 24-month period, while again marginally not so for the other two periods (p=0.1232 and p=0.1577). Coefficients using Ang/Zhang single peer firms are not significantly different from 0. Overall these results support the hypothesis of a trend reversal following the event. Parent firms seem at least partially able to time the ECO to occur in periods of high market prices, thus taking advantage of "windows of opportunity ''4~ suggested by Ritter (1991) for IPOs, and as supported by Hirshleifer (2001), who argues that market values regularly deviate from fundamental values, and firms exploit these misvaluations by issuing overvalued equity. Size/BTM Size/BTM portfolio ew portfolio vw

Coefficient -(12 months) Coefficient- (24 months) Coefficient - (36 months) p-value - (12 months) p-value - (24 months) p-value - (36 months)

-0.1369 -0.1413 -0.1151

-0.1684 -0.1570 -0.1814

0.0292 0.0734

0.0558 0.1431

0.1655

0.1162

LBT single peer

AngZhang single peer

-0.1430 -0.2212 -0.2525 0.1232

-0.0299 0.0192 -0.3197 0.8092 0.9098 0.2481

0.0389

0.1577

Table 19: Regression of LTPP on 12-month pre-event L T P P

5.6.2.5 Additional test of the market timing hypothesis In addition to positive parent pre-event LTPP, it would be desirable to find evidence on a subsidiary level supporting the market timing hypothesis. It was previously found that 4O4

Ritter (1991 ), p. 3.

147

abnormal performance in the 12 months prior to the event is significantly more positive relative to size/BTM (not industry) matched portfolios, whereas it is not significantly different from 0 relative to size/BTM and industry-matched single peers. This result can be interpreted as implying that market timing refers to the ability of the parent firm to assess relative industry valuation levels, and time the ECO to occur in phases of high relative valuations in the parent firm's industry. However, it could be argued that the relevant industry is the subsidiary firm's industry, rather than the parent firm's industry. 4~ Hence, the question is whether the parent firm is able to time the ECO to occur in periods of high relative valuations in the subsidiary firm's industry. This is analysed next. The aim of the following analysis is to compare the development of the valuation levels of the subsidiary firm's industry with valuation levels of other industries. For each subsidiary firm in the sample, the market-to-book ratio is calculated in the IPO month for all firms which are both in the subsidiary firm's industry (defined on the two digits industry code level), and which come from the subsidiary firm's country. The median market-to-book ratio is defined as the subsidiary firm industry's valuation level at the IPO date. 4~ Similarly, valuation levels for all other two digits industries are calculated as the median market-to-book ratio of all firms in each respective industry within the subsidiary firm's country. In addition to the IPO date, industry valuation levels (both for the subsidiary firm's industry and for the other industries) are calculated 12, 24 and 36 months prior to the IPO date. For each sample firm, the percentage change in the valuation level from 12, 24 and 36 months before the IPO to the valuation level on the IPO date is calculated. Similarly, the percentage changes in the valuation levels of all other industries are calculated. The mean and median percentage change for sample firms is compared to mean and median percentage changes across all industries for all three time periods. The key result as shown in Table 20 is that valuation levels in the subsidiary firm's industry increase more from previous periods to the IPO period than valuation levels in other industries.

Mean change (sample industry) Mean change (all other industries) p-value (difference) Median change (sample industry) Median change (all other industries) p-value (difference)

12 months

24 months

36 months

26.1% 7.2%

55.9% 8.8%

48.3% 17.8%

0.0003

0.0000

0.0008

8.9% 7.9%

11.3% 9.2%

23.0% 8.4%

0.0030

0.0092

0.0092

Table 20: Change in valuation levels of subsidiary firm's and other industries

405 104 of the 178 sample cases are represented by parent and subsidiary firm's from different industries. 406 The key results of the analysis are robust to using P/E ratios instead of market-to-book ratios.

148

The difference in the change is significant across all three time periods, and is robust to using both mean and median percentage changes. The result represents strong support for the hypothesis that ECOs are timed to occur in periods when valuation levels in the subsidiary firm's industry have developed more positively than valuation levels in other industries. Parent firms hence seem able to assess that the valuations for firms in their subsidiary firm's industry have improved more relative to valuations in other industries, and consequently take advantage of this positive relative development to sell equity in the subsidiary firm. This result is very intuitive and adds to the evidence supporting the market timing hypothesis .It could be argued that rather than comparing valuation levels of industries, the subsidiary firm's valuation relative to its industry peers should be analysed. There is a methodological and a conceptual reason why inter-industry rather than intra-industry valuation level differences are analysed. First, methodologically, comparing valuation levels to benchmark firms does not make sense when the matching of those benchmark firms is performed along the same dimensions on which the valuation level is assessed. Since both benchmark portfolios and single peer firms are selected (among other) on the basis of similar book-to-market ratios, and valuation is assessed as the reverse of this ratio (i.e., market-to-book ratio), analysing the relative valuation of subsidiary to benchmark firms will not yield meaningful results. Alternative measures of valuation could be used (e.g., P/E ratios), but these are likely to be correlated with market-to-book ratios. Second, even if matching of benchmarks was solely based on industry and not on book-to-market ratios to prevent the previous criticism, the analysis would be flawed on conceptual grounds: Comparing the subsidiary firm's valuation to its industry peers requires using a market value of the subsidiary firm's equity, e.g., the market value of equity on the first trading day. However, the parent firm does not know this value ex ante when the decision to carve out the subsidiary firm is taken. Hence, finding a high valuation level of the subsidiary firm relative to industry peers does not allow concluding that market timing occurs, because the only value that the parent firm knows at the time when the ECO decision is made is the valuation level of firms in the subsidiary firm's industry which are already listed. For that purpose, the present analysis uses this industry valuation level, thereby reflecting the knowledge of the parent firm at the time of the announcement.

5.6.2.6 Testing the second hypothesis: STPP andpre-event LTPP To assess the second question of whether there is a relationship between pre-event LTPP and announcement period STPP, the various event window announcement period abnormal returns from the STPP analysis are regressed on 12 months pre-event BHARs. If companies use ECOs as a reaction to a negative share price development, coefficients should be negative. Alternatively, if companies manage to convey positive signals to potential investors through a past positive share price development, coefficients should be positive. Appendix 39 shows the coefficients and respective p-values of the OLS

149

regressions. Coefficients across most event windows are positive, and significantly so for three out of the four benchmarks across the three shorter event day windows [0], [0,1] and [-1,1]. This result refutes the hypothesis that ECOs on average are carried out as a reaction to negative share price developments, and support the hypothesis that companies which have performed well in the 12 months prior to the ECO announcement are likely to profit from this positive development in the form of higher announcement period abnormal returns. Potentially, this result can be explained by a decreased informational asymmetry between the parent firm and future investors resulting from the positive signalling in the form of positive past returns. This result also ties in with the considerations on the differences between spin-offs, for which there is no pre-event abnormal performance because there does not need to be any (since informational asymmetry issues do not play a major role), and ECOs, for which there is a positive pre-event abnormal performance (since informational asymmetry issues do play a major role).

5.6.3 Explanation of cross-sectional results forparent firms 5.6.3.1 Two hypotheses related to post-event performance Given the findings with respect to abnormal STPP and abnormal LTOP, the following two hypotheses offer themselves: First, it was previously found that operating performance is positive around the time of the event, negative in the first year after the event, and neutral in the second and third year after the event. ECO are thus impacting the level of operating performance in the years following the event. Market participants will update their expectations (Bayesian updating) based on this new information, potentially leading to an abnormal LTPP of the same sign. The hypothesis therefore is that a positive (negative) long-term operating performance leads to a positive (negative) long-term price performance. The underlying idea is that the fundamental value of a company (as shown in operating performance measures) should determine its market value (as shown in price performance measures) in the long run. Second, market participants may realise that their initial reaction to the announcement of the ECO has been inadequate. Specifically, Daniel/Hirschleifer/Subrahmanyam (1998) predict that investors initially under-react to news, leading to post-event price drifts of the same sign as the announcement period return. Eventually investors realise that they have 'overcorrected' their initial under-reaction, leading to a price drift in the opposite direction. Daniel/Hirschleifer/Subrahmanyam's (1998) model thus predicts short-term trend continuation (i.e., momentum) and long-term trend reversal. The hypothesis therefore is that a positive (negative) short-term price performance causes a positive (negative) mid-term price performance; and a positive (negative) short-term price performance causes a negative (positive) long-term price performance. Mid-term is defined as the 12-month period following the event, whereas long-term comprises the 24- and 36-month periods following the event. The cut-off point between mid- and long-term roughly corresponds to Jegadeesh/Titman (1993), who find momentum in

150

share prices over 3 to 12 months, and DeBondt/Thaler (1985), who find trend reversal in share prices over 36 to 60 months.

5.6.3.2

Testing the first hypothesis: LTOP causes LTPP

To test the first hypothesis of a causal link from LTOP to LTPP, LTPP is regressed on LTOP. Given the various alternatives when measuring LTOP and LTPP, there is a multitude of possible combinations between measures for both performances. LTOP can be measured across three different time horizons (as the mean of the yearly operating performance measure in the IPO year and the one, two and three years following the IPO year, respectively), based on two different matching methods (BL4 and BL5), and four different profitability measures (ROA, ROCAA, ROS and ROMVA). 4~ LTPP is the dependent variable and can be measured across three different time horizons (one/two/three years post-event) and based on four different benchmark measures (equal-weighted size/BTM portfolio, value-weighted size/BTM portfolio, LBT single peer, Ang/Zhang single peer). This results in at least 3*2*4*4=96 combinations of dependent and independent variables. Another set of 96 combinations is created when using level (as opposed to profitability) measures of performance (sales, EBIT, assets and capex growth). Appendix 40 shows the combinations of various alternative LTPP and LTOP level measures, for which the coefficients are statistically significant at least at the 5% level. Out of the 96 coefficients, 31 are significantly positive. In particular, 17 coefficients based on sales growth as the level measure are significantly positive. As the table indicates, the result is robust to a variety of alternative time horizons, matching methods and benchmarks. Similar results hold for asset and capex growth. Apparently, there is a positive link between a company's ability to grow its operations following an ECO and its share price performance. Appendix 41 shows the combinations of various alternative LTPP and LTOP profitability measures, for which the regression coefficients are statistically significant at the 5% level. Out of the 96 coefficients, 20 are significantly positive at least at the 5% level. This result implies a link between a company's ability to increase the profitability of its operations following an ECO and its share price performance. The two analyses suggest that the cross-sectional variation in LTPP can be partially explained by the post-event development of the operating performance of parent firms.

407

These are the benchmarks and operating performance measures as suggested by Barber/Lyon (1996). From the five benchmarks suggested, only two (BL4 and BL5) are used in this study because they are shown by Barber/Lyon (1996) to produce the best-specified results. See section 4.6.1.1 for further details on how these benchmarks are constructed.

151

Companies able to grow their operations and become more profitable in the years following the ECO also tend to perform well in the stock market. 4~

5.6.3.3

Testing the second hypothesis." STPP causes LTPP

To test the second hypothesis, various LTPP measures are regressed on the APAR from the ten different event windows. STPP is the independent variable and can be measured across ten different event windows. LTPP is the dependent variable and can be measured across three time horizons and based on four different benchmark measures, resulting in at least 10"3"4=120 combinations of dependent and independent variables. Table 21 shows all combinations of LTPP and STPP measures which produce significant regression coefficients (at least at the 5% level). Significantly negative coefficients are found in ten out of 120 cases, of which four lie in the [0,1] day window. Of the ten significant cases, seven refer to 12-month LTPP measures, and three refer to 24-month LTPP measures, respectively. There are no occurrences of significantly positive coefficients. This result refutes the underreaction hypothesis based on Daniel/Hirshleifer/Subrahmanyam's (1998) model: There is no evidence of positive mid-term price momentum following the event. In contrast, companies with a positive STPP tend to have negative LTPP, and vice-versa. This implies that market participants overreact to the initial ECO announcement, and correct their overreaction in the 12 and 24 months subsequent to the event. This result may also be a reflection of a general trend reversal in share prices: If parent firms announce an ECO in times of high prices (as evidenced by positive pre-event performance), returns in the following months are more likely to be lower, and consequently the relationship between STPP (which occurs in the high price period) and LTPP will be negative. Years Event window

1 1 1 1 1 1 1 2 2 2

[- 10,0] [0,1] [0,1] [-1,1] [-5,5] [- 10,5] [- 10,5] [0,1 ] [0,1] [- 10,5]

LTPP Benchmark

Single peer Single peer AngZhang MC Single peer Single peer Single peer AngZhang MC Single peer AngZhang MC Single peer

p-value

Coefficient

0.0272 0.0238 0.0099 0.0432 0.0479 0.0069 0.0487 0.0090 0.0274 0.0108

- 1.8942 -3.6271 -5.4663 -2.9978 - 1.4205 - 1.4561 - 1.5681 -4.8710 -5.2022 - 1.5911

Table 21: OLS-regression of LTPP on STPP

408

Clearly, this does not establish a causal link between the ECO and the operating performance development, and a more complete analysis of the phenomenon could compare the growth of parent companies to the growth of non-ECO companies across similar time-horizons

152

5.6.3.4 Multivariate case To see whether the above findings from the separate tests of the two hypotheses also hold in the multivariate case, an additional model is estimated, using measures of profitability, growth and STPP as independent variables. The choice of the first two is somewhat arbitrary, given two different benchmarks, four measures of profitability and four measures of growth. Based on the findings of the univariate analysis, ROS and sales growth are chosen as the profitability and level measure, respectively. To prevent a potential small-firm effect from driving results, firm size is controlled for (measured as the log of the market value of the parent firm's equity in the month prior to the event). As pointed out by Ritter/Welch (2002), long-term price performance may be driven by data from the Intemet bubble period. 4~ Therefore a hot market dummy (equalling 1 when the event occurred between 1998 and 2000, and 0 otherwise) is also included in the regression. Appendix 42 shows various models used to estimate LTPP. The LTOP level measure coefficients have the expected positive sign Oust as in the univariate case) and are significant. In addition, the LTOP profitability measure coefficients also have the expected positive sign, and are also significant in all two-year models, and one of the one-year models. The coefficients on the STPP measure are all negative as expected, and significant in two two-year and two three-year models. The regression coefficient on size is not significant, although marginally not so in two of the three-year models. The regression coefficient on the hot market dummy is significantly negative for some of the models, indicating that performance of parents firms who announced an ECO between 1998 and 2000 performed worse than parent firms who announced an ECO in other years.

5.6.3.5 Interpretation of results The findings imply that companies managing to grow their operations and make them more profitable, presumably as a result of the ECO, tend to perform well in the stock market. This result seems very intuitive, and is consistent with the view that in the long run market values are determined by fundamental values. Additionally, there is some evidence that post-event performance is at least partially a correction of the announcement period return, indicating an overreaction by investors to the announcement of an ECO. This could be a reflection of the market timing hypothesis discussed earlier: Positive returns before and around the announcement period are followed by negative returns in the subsequent months. The finding is also consistent with the STPP analysis, where it was found that returns for parent firms are on average negative in a two-day event window following the positive returns on the announcement date.

409

See Ritter/Welch (2002), p. 1822.

153

5.6.4 Post-eventperformance of subsidiaryfirms Table 22 shows the results of the LTPP analysis for subsidiary firms using the BHAR methodology. Abnormal performance is negative relative to all benchmarks. All valueweighted results are less negative than equal-weighted results, again implying that negative abnormal performance is mainly due to smaller companies. The question of which result is 'correct' rests on the underlying assumptions about whether equal- or value-weighting represents the appropriate methodology: As pointed out in section 5.5.2, the sample contains a small number of large firms, and these few firms may drive average results when value-weighting. Equal-weighting thus seem more appropriate if one wants to make a statement on the likely performance of any particular ECO, and also if, as pointed out by Brav/Geczy/Gompers (2000), one is interested in the managerial implications of abnormal performance. Also, value-weighting does not represent an ex-ante implementable investment strategy because the weights (i.e., the market values) of future ECOs are unknown at the time of any given ECO. Equalweighting BHARs thus seems more appropriate both from a firm and from an investor's point of view. The Ang/Zhang single peer methodology is not applicable for subsidiary firms because market prices (and thus correlated companies) are not available prior to the ECO. Period

12 m o n t h s

24 m o n t h s

36 m o n t h s

-6.7% -1.4% -12.4% -4.0% -0.2% -3.9%

-14.8% -8.8% -24.7% -22.1% -17.6% -11.9%

-14.6% 0.0% -29.7% -20.1% -1.6% -12.7%

Size/BM portfolio (ew) Size/BM portfolio (vw) Size/BM portfolio (median) LBT single peer (ew) LBT single peer (vw) LBT single peer (median)

Table 22: Subsidiary firm BHAR across various periods

5.6. 4.1 Statistical significance of BHARs Appendix 43 shows p-values for all 18 methodologies of assessing statistical significance. Again as in the case of parent firm abnormal performance, most significant results refer to equal-weighted abnormal returns in the 24- and 36-month periods following the event. The level of the negative abnormal performance over the 24 period is roughly -0.5% to -1% per month. Given this magnitude, it is not surprising that conventional t-tests judge this result to be statistically significant: Both the dependent (standard paired difference) and the independent (two groups difference of means) t-test find significance, with the independent t-test producing higher p-values than the dependent t-test. This relationship implies that intra-sample dependence is an issue, and not accounting for it (using the dependent t-test) biases p-values downward. Hscstatistic p-values are higher than sc-statistic p-values, again supporting the idea that the hsc-methodology produces conservative results. The skewness-adjusted test statistic also yields significant p-values for both 24- and 36-month periods. This result may be

154

caused by the non-random nature of the sample leading to cross-correlation, and it is a manifestation of the problem addressed by Lyon/Barber/Tsai (1999). 41~

5.6. 4.2

Calendar time method

Table 23 shows the results of the LTPP analysis for subsidiary firms using the calendar time portfolio method. Again, as in the case of the parent firms, using the unadjusted version with equal weighting and based on OLS regressions yields a considerable monthly underperformance of the subsidiary firms of roughly 1.31% per month. However, the mean intercept from the 1,000 pseudo-portfolios is roughly 0.66%, implying that some of the seeming underperformance is due to model misspecification. Using equal-weighting the adjusted intercept is significantly negative for both OLS and WLS. Again as in the analysis of parent firms, this significant underperformance vanishes when value-weighting is applied, both under OLS and WLS: Abnormal performance is then indistinguishably different from 0. Result using the calendar time method are thus broadly in line with those obtained using the B H A R methodology.

Basic factor model

Average. of 1,000 pseudo portfolios

Adjusted factor model

EW - OLS - Alpha EW - OLS - t stat EW - OLS - p

-0.0131 -3.9227

-0.0066 -2.2014

-0.0065 - 1.9441

0.0001

0.0296

0.0542

EW - WLS - Alpha EW - WLS - t stat E W - WLS - p

-0.0174 -4.8778

-0.0091 -3.1064

-0.0083 -2.3344

0.0000

0.0024

0.0212

V W - OLS - Alpha V W - OLS - t stat V W - OLS - p

-0.0009 -0.2019 0.8404

-0.0003 -0.0537 0.9573

-0.0006 -0.1441 0.8857

V W - WLS - Alpha V W - WLS - t stat V W - WLS - p

0.0015 0.3114 0.7560

0.0001 0.0958 0.9239

0.0014 0.2824 0.7781

Legend: ew = equal-weighted, vw = value-weighted OLS = ordinary least squares, WLS = weighted least squares Table 23: Calendar time method for subsidiary firms

410

See Lyon/Barber/Tsai (1999), p. 188 and p. 190.

155

5.6.4.3 Interpretation of results The results offer interesting comparisons to previous studies. Again, as in the case of parent firms, Ritter's (1991) result that underperformance is caused by small firms finds some support in the current study in that significant results are restricted to equalweighted schemes. Drobetz/Kammermann/WNchli (2003) find no underperformance for their sample of Swiss IPOs in the first three years after the IPO 411, consistent with this study. Schikowsky/Schiereck/VNkle/Voigt (2005) find an underperformance of ECO subsidiary firms over a two-year period following the event when using equalweighted observations, but not when using value-weighting. The results in the current study are consistent with this finding in that significance is found for equal-weighted schemes, but not for value-weighted schemes. Appendix 44, which graphs the BHAR across the 36 months following the end of the event month, reveals that there is a positive average performance in the first few months immediately following the event month. This finding is consistent with previous studies on IPOs and SEOs. Ritter (1991) finds positive abnormal returns in the first two months following the I P O . 412 Drobetz/Kammermann/WNchli (2003) find some positive abnormal performance in the first months following the IPO in their Swiss sample. 413 Alvarez/Gonzales (2005) find positive abnormal returns of Spanish IPO companies in the first year following the event, and negative abnormal returns over a three- and fiveyear time horizon. Clarke/Dunbar/Kahle (2004) find a positive median abnormal return for companies engaging in an SEO in the first year, whereas median abnormal returns are negative over three and five years. While most of these results are not significant, they raise the possibility of a pattern which does not seem to have attracted much attention in the LTPP literature: While the focus of many studies (including this one) is on potential underperformance in multi-year windows following the event, there is at least some support for a positive performance in the months immediately following the event. Why should mid-term post-event performance be positive? Potential explanations for a positive performance in the first few months following the ECO can be based on behavioural models. First, Daniel/Hirshleifer/Subrahmanyam (1998) develop a model which predicts short-term positive autocorrelation (i.e., momentum) in share prices following an event: If an investors buys a stock based on his conviction that it will rise (following a private signal, in the terminology of Daniel/Hirshleifer/Subrahmanyam (1998)), and it subsequently does rise following the announcement of positive news (public signal), the investor is confirmed in his belief of having a superior ability to assess the future direction of the stock (self-attribution bias), and may continue to buy it, causing momentum. The findings in this study seem consistent with this model. Second, Rangan (1998) points out that earnings management witnessed in IPOs may be 411 412 413

They do find significant underperformance in the followingyears, but warn that this result may be due to the bad-modelproblem (see Drobetz/Kammermann/W~ilchli(2003), p. 26.) See Ritter (1991), p. 10.

See Drobetz/Kammermann/W~ilchli(2003), p. 47.

156

likely to continue in the two quarters following the IPO, to prevent investors from filing lawsuits against a company which surprises investors negatively immediately following the IPO. Also, given the frequent use of lock-up agreements whereby the owners agree not to sell any further shares during a 90-180 days period, earnings management in the period following the event is encouraged to ensure an attractive price once the lock-up period expires. If investors are at least partly led astray by earnings management, positive earnings management in the first two quarters following the ECO may support the share price and lead to positive abnormal returns. Third, both the subsidiary firm and the underwriting investment bank may have an intrinsic interest in a positive share price development following the ECO: A negative share price performance in the immediate aftermath of an ECO may make a return to the capital market for future SEOs more difficult. Depending on the size of the issue, both subsidiary firm and investment bank may have some discretionary power to support the share price by buying the newly issued shares in cases where supply presses on the price.

5.6.5 Explanationof cross-sectional results of subsidiaryfirms Whereas the explanation of parent firm price performance has been linked to the parent firm's operating performance, the subsidiary firm's price performance is explained as a function of the characteristics of the ECO. The reason is that the characteristics of ECO are more likely to be relevant for the subsidiary firm's price performance than the parent firm's price performance, given the potentially low significance of subsidiary for the parent firm as a whole. 414 Simultaneously, it is important to control

the for the for

other factors previously shown to influence the cross-sectional distribution of LTPP measures in IPOs. 415 Similar to the analysis of STPP, a multiple regression framework is used. The LTPP measures are the dependent variables, and independent variables include the following event and firm characteristics:

414

415

Unreported results of analyses attempting to explain subsidiary price performance as a function of its operating performance show results similar to the parent company case for LTOP efficiency, which is significantly positively related to price performance. The results on LTOP level, however, are mixed and sometimes even negative. Overall adjusted RE's are very low (1-5%). In addition to the variables described subsequently, Roosenboom/Goot/Mertens (2003), using a sample of Dutch IPOs, find that companies with a high level of discretionary accruals at the time of the IPO perform worse than companies with lower levels of discretionary accruals. This variable is not used in the present analysis because sufficient data for its construction could not be made available. Alvarez/Gonzales (2005) using a sample of Spanish IPOs find that long-run performance is positively related to the volume of funds raised through SEOs subsequent to the IPO, and to the initial underpricing. The first variable (volume of funds) is not used here because the required data for the ECO sample was unavailable. The second variable (underpricing) is not used because, as pointed out by Ritter/Welch (2002), results regarding this variable depend on whether IPOs from the Intemet bubble period are included in the sample. Since Alvarez/Gonzales (2005) do not control for the time period in which the IPO occurred, their finding regarding underpricing cannot be generalized.

157

5.6.5.1

Variables in regression

Remaining stake: Annema/Fallon/Goedhard (2002) find that ECO subsidiary firms which are fully disposed outperform while subsidiary firms remaining under the control of the parent firm underperform the S&P500 index. In contrast, Powers (2003) finds that subsidiary firm performance increases with the stake retained by the parent firm. The expected sign of the regression coefficient on the remaining stake variable is therefore undetermined. Financial distress: Madura/Nixon (2002) find that if the parent firm has been financially distressed prior to the ECO, both parent and subsidiary firm on average underperform. Financial distress is measured by a dummy variable equalling one if the interest coverage ratio (EBIT/total interest expense) is less than one, as suggested by Asquith/Gertner/Scharfstein (1994). 416 Although there are alternatives for the definition of distressed companies 41v, it seems appropriate to use the dummy approach for the purpose of the present study. This method has found widespread application in empirical literature. 418The coefficient is expected to have a negative sign. Region: In the STPP analysis it was found that parent firms announcing an ECO on average have higher abnormal returns when they are located in countries with higher shareholder rights. A similar relationship could hold for LTPP: Firms in higher shareholder rights countries can be expected to perform better than firms in countries with lower shareholder rights. The coefficient on a dummy variable equalling 1 for firms from higher shareholder rights countries is thus expected to have a positive sign. Motivation: In the STPP analysis it was found that parent firms announcing an ECO with the motivation to develop the business of the subsidiary firm on average earn higher abnormal returns. This motivation may also positively influence the subsidiary firm's LTPP. The coefficient on a dummy variable equalling 1 for firms with this motivation is thus expected to have a positive sign. Industry: Desai/Jain (1999) find that spin-offs where parent and subsidiary firm come from different industries ('cross-industry') perform better than same-industry spin-offs. Hence, a dummy is created equalling 1 when parent and subsidiary firm share the same two digits industry SIC code, and 0 otherwise. The sign of the regression coefficient is expected to be negative. Number of segments: The higher the number of segments, the larger potentially the relative gain of the subsidiary firm, and the higher the potential for eliminating negative synergies. 419 The regression coefficient is therefore expected to have a positive sign.

416 See Madura/Nixon (2002), p. 174. 417 See discussion in section 4.6.4.1. 418 Other studies employing the dummy approach include Wagner (2004), p. 18; Hovakimian/Titman (2003), p. 4; and Molina (2005), p. 1439. 419 See Wagner (2004), p. 17.

158

Size and BTM: Brav/Gecy/Gompers (2000) find that underperformance of IPOs in their sample is explained by small firms with low BTM ratios. The regression model therefore controls for these two variables. Size is measured as the market value of the subsidiary firm's equity on the first day of trading. BTM is calculated in the standard way as the sum of the book value of common equity and debt, divided by the sum of the market value of equity and the book value of debt. In analogy to Brav/Gecy/Gompers (2000), the expected sign on the coefficients is positive for both size and BTM: Larger firms and firms with a higher BTM are expected to perform better. 'Hot market': As pointed out by Ritter/Welch (2002), an evaluation of IPO long run performance needs to account for the Internet bubble period. A dummy variable is created, equalling 1 when the IPO occurred between 1998 and 2000, and 0 otherwise. The expected sign of the regression coefficient is negative: IPOs from the bubble period are expected to perform worse than IPOs from non-bubble periods. Profitability: Yi (2001) finds that firms with positive earnings at the time of the IPO have a better LTPP than firms with negative earnings. Similar to Yi (2001), a dummy variable is created, equalling 1 when the subsidiary firm has positive earnings (defined as ROA>0) in the last fiscal year prior to the IPO, and 0 if earnings are negative. The expected sign of the regression coefficient is positive.

5.6.5.2 Interpretation of results Appendix 45 shows various models used to explain subsidiary firm LTPP. The coefficients on the industry dummy are negative in all models, and significantly so in all two- and three-year models, mirroring results from the LTOP profitability analysis. The finding indicates that subsidiary firms arising out of cross-industry ECOs perform better than subsidiary firms arising out of same-industry ECOs. Again, the likely explanation is that cross-industry parent/subsidiary firm combinations are subject to a higher level of negative synergies, and their separation leads both to a higher level of operational performance as found in section 4.6.5.2, as well as to a more positive share price development. The distress dummy is significantly negative in all two-year models, which is a direct confirmation of the results by Madura/Nixon (2002) who find a lower price performance for subsidiary firms carved out of distressed parent firms. They interpret their result as the consequence of distress symptoms being carved out along with the subsidiary firm. 42~Again this result is consistent with the LTOP profitability analysis in section 4.6.5.2: If distress is, for example, the ultimate result of bad management practices or other organizational issues, it seems plausible that the same effects will be observed in the subsidiary firm, leading both to lower profitability and a worse share price performance. 420

See Madura/Nixon (2002), p. 181.

159

The period dummy is significantly negative for almost all models, again in congruence with the LTOP profitability analysis. In that context a possible explanation proposed was that subsidiary firms carved out before 1998 were carefully selected for high performance potential, whereas later ECOs were more motivated by costly internal capital markets. Consequently subsidiary firms may on average not have been as well suited for a positive post-ECO performance. An alternative (and mutually not exclusive) explanation for a more negative performance for post-1998 ECOs is the higher number of ECOs carried out for presumably opportunistic market timing reasons in the boom years 1998 to 2000: These ECOs may have failed to materialise the benefits they were initially expected to produce. The market mood in the boom years was such that investors were likely to invest in (and companies thus able to sell) shares in companies which in other periods may not have passed going-public scrutiny tests. The coefficient on the profitability dummy is significantly positive in some of the oneand three-year models. This confirms Yi's (2001) finding that firms with positive earnings at the time of the IPO have a better LTPP than firms with negative earnings for the case of ECOs. The result is intuitive and suggests that fundamental value has an important impact on market value, a notion consistent with section 5.6.3.4, where it was found that operating performance is an important driver of parent firm LTPP. Finally, the size coefficient is significantly positive in the three-year models, implying that larger subsidiary firms perform better than smaller subsidiary firms. This result reflects the finding in section 5.6.4.1 that negative price performance seems driven by smaller firms as indicated by the equal-weighted BHAR being significantly negative, and the value-weighted BHAR n o t being significantly different from 0. All remaining coefficients are not significant. In particular, the coefficient on the remaining stake dummy is insignificantly different from 0. There are two potential explanations. First, the result could be due to the two conflicting effects highlighted in section 4.6.4.1: According to the incentive alignment hypothesis, performance should increase with the stake held, while according to the entrenchment hypothesis, performance should decrease with the stake held. The relationship between stake held and performance could also be non-linear as suggested by Morck/Shleifer/Vishny (1988). Second, the result may arise because of the different way the variable is measured: Both Annema/Fallon/Goedhard (2002) and Powers (2003) use the stake retained in the initial ECO, whereas the present study employs the stake owned by the parent firm at the end of each year under consideration. Arguably the latter method better captures the empirical fact that the parent firm's stake continues to change following the initial ECO, and taking these changes into consideration when analysing various time horizons is likely to produce more meaningful results.

160

5.7

Conclusion

This study analyses abnormal LTPP for parent and subsidiary firms involved in an ECO. It has become an established admonition that "conclusions from...long-horizon studies require extreme caution ''421. For that reason, a considerable number of test designs are implemented to increase robustness of results, including numerous variations of the BHAR methodology and calendar time portfolio methods. ECOs have characteristics of both portfolio restructuring activities (e.g., spin-offs), commonly associated with a positive abnormal LTPP; and of capital-raising activities (e.g., IPOs and SEOs), commonly associated with a negative abnormal LTPP. Results in this study indicate that ECOs behave more like IPOs and SEOs than like spin-offs: Both mean and median abnormal return measures tend to be negative for parent and for subsidiary firms. However, the assessment whether these results are significant rests on one's view of what constitutes an appropriate testing methodology: Equal-weighting schemes are more likely to produce significantly negative results, particularly over longer time periods, whereas value-weighting does not yield significant results. This result is robust to the use of the Mitchell-Stafford adjusted calendar time method. It suggests that underperformance is more likely for smaller firms. However, the fact that value-weighted results are less negative than equal-weighted results is also driven by the sample's heterogeneity in firm size, where a few large firms impact value-weighted results. This confirms Brav/Geczy/Gompers (2000), who argue "how fragile certain results in the [LTPP] literature actually are ''422. Similarly, Sapusek (2000) demonstrates for a sample of German IPOs how different benchmarks can lead researchers to conclude that IPOs outperform, underperform, or perform in line with benchmarks. 423 Applying, as the present study does, a whole arsenal of testing methodologies highlights some of the difficulties researchers face when assessing abnormal LTPP. It also visualises the prevalence of the bad-model problem addressed by Fama (1998): Unless a more powerful and commonly accepted model producing normal returns is found, LTPP studies are likely to continue to produce method-dependent results. For subsidiary firms, there is some indication of an outperformance in the first few months immediately following the ECO. This may be the result of a self-attribution bias affecting investors, continued earnings management following the IPO, and supporting share purchases by the ECO firms and the underwriting investment bank. For parent firms, there are three indications supporting the idea of market timing. First, the analysis of pre-event LTPP finds that parent firms seem to outperform in the 12 months prior to the event. Second, there is a negative relationship between pre-event and post-event LTPP, indicating that firms performing well prior to the ECO tend to perform worse following the ECO. Third, the analysis of relative changes in industry valuation levels reveals that valuations in the subsidiary firm's industry develop more 421 422 423

Kothari/Wamer (1997), p. 302. Brav/Geczy/Gompers (2000), p. 212. See Sapusek (2000), p. 384.

161

positively over the 36 months preceding the ECO than valuations in other industries. All three findings suggest that parent firms time the ECO to occur in periods of high (and supposedly unsustainable) prices, and take advantage of periods of high relative valuations in the subsidiary firm's industry by selling equity in the subsidiary firm. There is also a positive link between pre-event outperformance and announcement period abnormal returns, suggesting that parent firms may use positive past performance as a positive signal to investors to reduce informational asymmetry. Multivariate analyses of the cross-section of LTPP results reveals that there is strong positive relationship between both a parent firm's profitability and growth measures on the one hand, and its LTPP on the other hand. This result is intuitive and implies that companies managing to grow profitably, potentially as a result of the ECO, also enjoy a positive market valuation. Additionally, post-event performance is partially a correction of the announcement period return, indicating an overreaction by investors to the initial ECO announcement. Subsidiary firms tend to perform best when they are carved out of non-financially distressed parent firms; when the parent firm is financially distressed, the subsidiary firm seems to inherit some of the causes of this distress, hampering its future share price development. Subsidiary firms arising out of cross-industry ECOs fare better than those arising out of same-industry ECOs, indicating that the removal of negative synergies is an important driver of share price performance. Also, subsidiary firms which have been profitable prior to the ECO perform better, suggesting that market value is driven by fundamental values. Finally, subsidiary firms carved out in the Internet bubble period have performed worse than subsidiary firms carved out outside of this period. This results either from a lower average subsidiary firm quality in the bubble period because of less public scrutiny of firms going public, or from increased opportunity costs of internal capital markets as external capital markets develop 424, and companies consequently decreasing their requirements on the suitability of subsidiary firms to be carved out. In all, many of the determinants of subsidiary firm profitability as identified in section 4.6.5.2 are also found to be determinants of subsidiary firm share price performance. This finding parallels the results from the analysis of parent firm share price performance, which also links market values to fundamental values.

424

See Khanna/Palepu (2000b), p. 281.

162

W h a t do we learn about internal capital markets from equity carve-outs? 6.1

Abstract

There is no consensus among academics whether internal capital markets (ICMs) are efficient or not. This study extends current literature by analysing two related questions with the help of a sample of European equity carve-outs (ECOs): First, what is the investors' average judgement on the existence of ICMs? Second, what are the conditions for ICM efficiency? Investors' opinion is assessed via the share price reaction to the announcement of an ECO, and ICM activity is measured using a series of variables constructed to reflect the size and the efficiency of a firm's ICM. First, investors seem to react more positively to the announcement of an intended ECO (i.e., higher parent firm abnormal retums) when the announcing firm's ICM is large, and when its size decreases following the ECO, according to the constructed size measures. These findings imply that investors generally consider the existence of an ICM to be value destroying. Investors react less positively to the announcement of an intended ECO (i.e., lower parent firm abnormal returns) when the announcing firm's ICM has been working efficiently prior to the announcement, and when the efficiency of the firm's ICM decreases following the ECO, according to the constructed efficiency measures. These findings indicate that investors are at least partially able to discern efficient from inefficient ICMs. Interpretability of results rests on the assumption that the measures of ICM size and efficiency are valid proxies for the characteristics of a firm's ICM. Second, there appear to be two distinct scenarios in which ICMs are efficient and investors value their existence: First, when business segment cash flows are positively correlated and the segments are hence related, potentially allowing the firm to crosssubsidise related activities; and second, when business segment cash flows are negatively correlated and the segments are hence unrelated, thereby increasing the value of the option to switch investment between business segments depending on the resolution of uncertainty. There seems to be little value in ICMs when neither of these conditions holds, as evidenced by the positive market reaction to ECO announcements by firms whose business segments display no systematic positive or negative relationship.

6.2

Introduction

Internal capital markets (ICMs) are defined as "markets in which corporate headquarters allocate capital to their business units ''425. An important aspect of ICMs is "crosssubsidization ''426 of business segments by channelling cash flows from one segment into 425 426

Gertner/Scharfstein/Stein (1994), p. 1211. Chevalier (2000), p. 3.

163

investment projects of another segment. This can be either value-creating, when capitalconstrained segments with good investment opportunities receive financing from capital-rich segments with low investment opportunities; or value-destroying, when cash flows in the opposite direction, potentially resulting from principal-agent problems. The role of ICMs in creating value is thus contested. Existing literature offers theoretical arguments and empirical evidence for both positive and negative value contributions by ICMs, and academics are far from reaching a consensus on this issue. In an ECO, an ICM is partially closed because the ties between parent and subsidiary firm are loosened or cut, depending on the stake retained by the parent firm after the event. Because of this change in the structure of the ICM, ECOs represent useful objects of analysis to address general questions about ICMs. Specifically, ECOs allow analysing how investors react when an ICM is partially closed, and the reaction can be studied as a function of both the structure of the ICM before the event, as well as the change in the ICM structure from the time before to the time after the event. Previous studies have applied similar research designs for US samples of spin-offs and tracking stock. Gertner/Powers/Scharfstein (2002) use a sample of US spin-off subsidiary firms to analyse whether investment behaviour improves following the separation from the parent firm. They describe spin-offs as "natural experiments ''427 suitable for their specific research question. Billett/Mauer (2000) use a sample of US parent firms issuing tracking stock to analyse the relationship between firm value and the efficiency of the ICM. Both studies do not focus on the specific portfolio restructuring activity (spin-off or tracking stock), but rather use the occurrence of this activity to analyse more general questions regarding the functioning of ICMs. The present study follows this path. As pointed out by Gertner/Powers/Scharfstein (2002), one caveat of such an approach is the potential for a self-selection bias428: Companies choose to undertake the portfolio restructuring activity, and generalizability of results may thus be hampered. 429 This admonition will need to be borne in mind when interpreting results in the present study. This study adds to the existing literature through the construction of a wide array of ICM measures, both of size and of efficiency, for a previously untested sample of European ECOs. Specifically, this study addresses two questions: First, do investors favour the existence of ICMs or not, and are they able to distinguish efficient from inefficient ICMs? Second, what are the conditions for ICM efficiency? To answer the first set of two questions, announcement period abnormal returns (APARs) are regressed on various measures of ICM size and efficiency, controlling for previously identified variables partially explaining the level of abnormal returns. The hypothesis is that the size and efficiency of the ICM influence the investors' reaction to 427 428 429

Gertner/Powers/Scharfstein (2002), p. 2481. See Gertner/Powers/Scharfstein(2002), p. 2482. For example, Gertner/Powers/Scharfstein (2002) find that investment behaviour improves following spin-offs. However, if only firms with suboptimal investment behaviour engage in a spin-off in the first place, one cannot deduce that ICMs work inefficiently in general.

164

an ECO announcement. If the hypothesis is true, the implication is that investors take firm-specific ICM characteristics into consideration when forming their view (expressed in the APARs) on the level of value creation resulting from the ECO. Whether investors favour the existence of an ICM is assessed on the basis of the sign of the coefficients of the ICM size variables: First, assume investors have a positive view of ICMs. In this case, they will be negatively surprised by the announcement that the ICM will be partially closed. The degree of disappointment is likely to increase with the size of the ICM. Hence, the sign of the regression coefficient will be negative. Alternatively, assume investors have a negative view of ICMs. In this case, they will be positively surprised by the announcement that the ICM will be partially closed. The degree of positive surprise is likely to increase with the size of the ICM. Hence, the sign of the regression coefficient will be positive. Whether investors are able to distinguish efficient from inefficient ICMs is assessed on the basis of the sign of the coefficients of the ICM efficiency variables: First, assume investors are able to distinguish efficient from inefficient ICMs. Assuming investors value the existence of an efficient ICM, their reaction to the announcement of the partial closure of an efficient ICM will be negative, and increasingly so in the efficiency of the ICM. Equivalently, the reaction to the announcement of the partial closure of an inefficient ICM should be positive, and increasingly so in the inefficiency of the ICM. Thus, a negative regression coefficient is expected. Alternatively, assume investors are not able to distinguish efficient from inefficient ICMs. In this case, there will be no systematic relationship between ICM efficiency measures and the APAR (i.e., the regression coefficient will not be significantly different from 0). A significantly positive regression coefficient is not expected, because it would imply that investors consistently appreciate that efficient ICMs are closed. To summarise, a positive (negative) coefficient when regressing APAR on ICM size measures implies that investors are happy (unhappy) about ICMs being partially closed, and hence have a negative (positive) view of ICMs. A negative (non-significant) coefficient when regression APAR on ICM efficiency measures implies that investors are (not) able to discern efficient and inefficient ICMs. The second question- under what conditions are ICMs efficient? - is a logical followup question arising if investors are indeed able to distinguish efficient from inefficient ICMs. Extant literature suggests two contradictory conditions for ICM efficiency. On the one hand, Rajan/Servaes/Zingales (2000) see the relatedness of business segments as a prerequisite of ICM efficiency. Related business segments allow management to use its capacities more effectively, and focus its attention on similar opportunities. Diversity in resources and investment opportunities, on the other hand, is costly for firms. The reasoning goes as follows: Managers can choose between 'efficient' and 'defensive' investments. 43~An efficient investment benefits the firm as a 430

See Rajan/Servaes/Zingales (2000), p. 37.

165

whole while the defensive investment benefits the divisional manager more directly. Divisional managers will only undertake efficient investments when they don't have to share the surplus with the remaining divisions. Unequal investment opportunities therefore distort the incentives of divisional managers towards defensive investments. Company headquarters, anticipating this behaviour, attempt to equalise investment opportunities across the group's business segments by allocating capital from larger divisions with higher investment opportunities to smaller divisions with lower investment opportunities. The result is an inefficient (and hence costly) ICM. Since diversity in resources and investment opportunities is more likely to arise when business segments are unrelated, relatedness of business segments is seen as a prerequisite for ICM efficiency. Similarly, Stein (1997) argues that a firm ranks its investment projects on a relative basis. If estimation errors are uncorrelated, they influence the quality of the ranking; however, the impact of the estimation errors on the quality of the ranking decreases as their correlation increases. Errors are more likely to be correlated when investment projects are similar, and consequently a firm with related business segments will have a more efficient ICM than a firm with unrelated business segments. 431 This view is also supported by the finding that improvements in operating performance in spin-offs is limited to cases where parent and subsidiary firms come from different industries (Daley/Mehrotra/Sivakumar (1997)). This implies the existence of negative synergies between unrelated business segments, which are removed through the spinoff. In a similar spirit, Vijh (2002) finds that firms announcing an ECO earn higher APAR when parent and subsidiary firms come from different industries. This implies that markets welcome the dissolution of cross-industry parent/subsidiary firm combinations, and prefer focussed firms operating in related business segments to unfocussed firms operating in unrelated business segments. On the other hand, ICMs can be seen as creating real options for companies. Trigeorgis (1993) analyses the options an oil company has when undertaking oil extraction and refinery projects. One of the options is to switch either the input or the output of the machinery, in order to either produce the same product with different inputs, or to produce a different product with the same inputs. The company will exercise this option depending on the resolution of uncertainty regarding the price development of the various inputs and outputs. The fact that it can choose/switch between various alternatives is thus valuable. Similarly, Triantis (2004) interprets ICMs in a real option framework, where the value of the option to switch investment between various segments decreases in the correlation of a firm's segments' cash flows. This result follows from an analogy between financial and real options: The value of a financial option increases as the volatility of the underlying increases. In real option terms, as pointed out by Dixit/Pindyck (1994), the greater the uncertainty about the profitability of an investment, the higher is the value of an option on this investment. 432 Intuitively,

431 432

See Stein (1997), p. 113. Dixit/Pindyck (1994) describe an option to wait until an investment decision is taken. The specific type of option described by Triantis (2004) in the context of an ICM is an option to switch

166

companies with more diverse business opportunities have a better chance of creating value in more states of the world, thanks to their flexibility to shift investments between diverse business opportunities. Similarly, Matsusaka/Nanda (2002) in their model of ICMs as real options posit that the value of ICMs increases as the variability of investment opportunities increases. 433 To the degree that correlation of segments' cash flows proxies for relatedness of segments, these two views seem mutually exclusive and are thus subject to empirical analysis. A straightforward approach to decide the issue is to regress the A P A R on the correlation between parent and subsidiary firm cash flows in the years prior to the ECO. A negative regression coefficient indicates that investors are unhappy 434 about an ICM consisting of related business segments being closed, supporting Rajan/Servaes/Zingales' (2000) view: The induction is that if investors are able to tell when an ICM is efficient 435, their unhappiness indicates that the ICM (consisting of related business segments) closed was efficient. Conversely, a positive regression coefficient indicates that investors are happy about an ICM consisting of related business segments being closed, supporting Triantis' (2004) view: The induction is that if investors are able to tell when an ICM is efficient, their happiness indicates that the ICM (consisting of related business segments) closed was inefficient. The remainder of this chapter is organised as follows: Section 6.3 surveys the relevant literature on ICMs and the conglomerate discount. Section 6.4 discusses the required data, the construction of various measures of size and efficiency of ICMs, and the research design. Section 6.5 presents and interprets the empirical results, and section 6.6 concludes.

6.3

Literature review

This section describes the relevance of ICMs in the context of ECOs (section 6.3.1). It reviews existing literature arguing for the inefficiency and the efficiency of ICMs, respectively (section 6.3.2). Potential biases in previous studies are discussed (section 6.3.3), as well as the dependency of results on the time period and geography of the sample (section 6.3.4).

investments between various business segments. Fundamentally, however, both option types represent exchange options (see Hommel/Mtiller(2000), p. 72-74). 433 In Matsusaka/Nanda's (2002) model, the value of the ICM also depends on the quantity of intemal resources: If no resources are available, the real option is worthless because there is no capital to invest. As resources increase, the value of the real option increases. Beyond a certain point, the value of the real option begins to decrease as resources increase, because enough capital is available for all valuable projects to be financed, and the value of the option to switch begins to decline. 434 Here and in the remainder of the analysis, happy investors produce positive abnormal retums, and unhappy investors produce negative abnormal returns. 435 Previewing results (for better understanding), there is some indication that investors are partially able to tell when ICMs are efficient. This is crucial for the interpretation of the results of the analysis of conditions of ICM efficiency.

167

6.3.1

ICMs in the context o f equity carve-outs

An ECO is "the initial public offering of some of the stock of a wholly owned subsidiary ''436. ECOs are characterised by their dual nature as both a corporate restructuring and a capital-raising mechanism. The positive APAR evidenced in previous studies 437 are explained by two sets of explanations: The divestiture gains hypothesis explains the positive market reaction with the increased business focus, increased financial flexibility, the pure-play character of the carved-out subsidiary firm and a better alignment of managers' and investors' interests. 438 The asymmetric information hypothesis states that issuing shares in the subsidiary firm signals an undervaluation of the larger parent firm assets and an overvaluation of the smaller subsidiary firm assets, leading to a price increase in the former. 439 Both hypotheses have found some empirical support. 44~ An ECO leads to a partial closure of an ICM. 'Partial closure' refers to the fact that the subsidiary is separated from the parent firm, and the remaining parent firm's ICM is smaller than it was prior to the ECO. 'Partial closure' also refers to the fact that the level of influence of the parent firm on financing and investment decisions by the subsidiary firm depends on the stake retained. While a complete review of the legal settings of all countries in the sample to determine the thresholds required for certain levels of influence is beyond the scope of this study, a simple example suffices to appreciate the issue. If the parent firm continues to own a majority stake in the subsidiary firm, it is very likely that it will continue to exert a considerable influence on financing and investment decisions. For example, since it can subscribe to any capital increase by the subsidiary firm, the parent firm can still effectively use its own resources to finance capital investments by the subsidiary firm. However, there are two important differences to the time before the ECO. First, since the subsidiary firm is public, there is a heightened level of investor scrutiny with regard to such transactions. Miles/Rosenfeld (1983), in the context of spin-offs, assume that this monitoring function by external capital markets increases the quality of the investment decisions. 441 The same argument can be made for ECOs. Similarly, in Holmstr6m/Tirole's (1993) model a decrease in the concentration of ownership in a firm (e.g., through a partial public floating) increases the benefits of market monitoring because the amount of information contained in the share price increases. An ECO is thus likely to increase the quality of the market's control over the investment decisions of the subsidiary firm. Second, many legal systems are likely to have a certain level of minority protection, limiting the parent firm's ability to take decisions unfavourable for the subsidiary firm. This minority 436 437 438 439

440

441

Schipper/Smith (1986), p. 154. See chapter 3 for a literature review. See Schipper/Smith (1986), p. 169-175 and Vijh (2002), p. 164-165. See Nanda (1991) and Slovin/Sushka/Ferraro (1995) for a detailed description of the asymmetric information hypothesis. See Vijh (2002) for empirical support of the divestiture gains hypothesis, and Michaely/Shaw (1995) for empirical support of the asymmetric information hypothesis. See Miles/Rosenfeld (1983), p. 1598.

168

protection serves as a further rationality check for financing and investment decisions initiated by the parent firm. In sum, a majority-controlled subsidiary firm is still under the influence of the parent film regarding its financing and investment decisions, but the presence of new shareholders creates limits to this influence. In this sense, an ECO 'partially closes' an ICM between parent and subsidiary firm. Two findings from the analysis of the cross-sectional variation in APAR are of particular interest in the context of ICMs and ECOs. First, APAR are higher on average when parent and subsidiary firms are from different industries. 442 This result is interpreted as indicating that the separation of cross-industry parent/subsidiary firm combinations removes a higher level of negative synergies443, and that therefore such combinations are inefficient on average. This inefficiency may arise from ICMs not allocating capital in a value-creating manner. Consequently a partial closure of a firm's ICM through the removal of one of the firm's subsidiaries (either through a sale, a spinoff, or an ECO) is considered to be beneficial to parent firm shareholders. Second, APAR are higher on average when the ECO is conducted as a primary (rather than a secondary) offering444: Apparently, investors prefer if the subsidiary firm, rather than the parent firm, receives the capital raised. Equivalently, investors seem to prefer if the ICM of the remaining firm decreases in size. Both findings suggest that the removal of a subsidiary firm, and thus a reduction in the size of the ICM, contributes positively to parent firm shareholder value.

6.3.2

Internal capital markets in the literature

Diversified conglomerates became popular in the third merger wave towards the end of the 1960s445: Growth by acquisition of either related or unrelated businesses was seen as a viable way of corporate expansion. Among the first to support this development from an academic perspective were Alchian (1969) and Williamson (1975), who both argued that ICMs help to overcome information and incentive problems associated with external capital markets (ECMs). Myers/Majluf (1984) argue along similar lines, and believe that companies diversify as a reaction to inefficient ECMs: ICMs overcome these inefficiencies by allocating capital between divisions to finance projects which would otherwise have not been financed due to informational asymmetries between the firm and ECMs. 446 Hubbard/Palia (1999) support this positive assessment of ICMs by finding that gains from acquisitions are highest when financially constrained firms acquire financially unconstrained targets. 442 See Vijh (2002), p. 177, and also chapter 3. 443 See Daley/Mehrotra/Sivakumar (1997), p. 269-272. 444 See Kaserer/Ahlers (2000), p. 562-564. 445 See Gaughan (2002), p. 32-37. 446 In addition to overcoming informational asymmetries, two popular arguments for the creation of diversified conglomerates are their increased debt capacity and associated value-gains through tax shields (Lewellen (1971)), as well as the creation of economies of scope (Teece (1980)). These are of less relevance in the context of this study.

169

Opponents to diversification, on the other hand, argue that managers diversify for reasons other than shareholder value creation, including increased private benefits (Jensen (1986)) and the better employment of their own capacities in other industries (Morck/Shleifer/Vishny (1990)). Since the beginning of the 1990s, the discussion on the benefits and downsides of corporate diversification is linked to the debate about a potential conglomerate discount. The existence of the latter is explained by some authors as evidence for the inefficient nature of conglomerates in general, and inefficient capital allocation in ICMs in particular. The following section surveys the relevant literature on these issues.

6.3.2.1 Argumentsfor inefficiency of lCMs The negative view is supported by a series of empirical studies: Lang/Stulz (1994) find that diversified firms have lower Tobin's Q ratios 447 than single-segment firms, and conclude that a conglomerate discount exists. Berger/Ofek (1995) quantify this discount to be 13-15% in the time period from 1986 to 1991. They do this by estimating the implied value of conglomerate companies relative to their market value. The implied value is calculated as the sum-of-the-parts of the individual business segments, and the value of each business segment is calculated on the basis of sales, EBIT and asset multiples derived from the market value of single-segment firms in the same industry as each respective business segment. 448 They find some evidence that the source of this discount is overinvestment by the individual business segments of conglomerate companies, partially offset by interest tax shields resulting from the higher debt capacity of conglomerate companies, as well as tax savings from the offsetting of profits and losses from various parts of the conglomerate. They also find that the discount is lower for conglomerates made up of same-industry divisions (defined on the two digits SIC code level). The notion of inefficient investment as being a major source for the discount is supported by Lamont (1997): He shows that conglomerates owning oil divisions significantly reduced investment in their non-oil divisions following the oil price shock in 1986. This result is inconsistent with the idea of efficient segment investment, which should be independent of the other segments' cash flows. Shin/Stulz (1998) find that a segment's investment depends both on its own cash flow and the cash flow of the other segments, and the segment with the highest investment opportunities has the same 447

448

Tobin's Q is defined as the market value of a firm's assets divided by the replacement value of the firm's assets. It can be operationalised in variety of ways. Lang/Stulz (1994) use the Smirlock/Gilligan/Marshall (1984) algorithm, explicitly modelling acquisition and depreciation schedules and correcting for price inflation to calculate the asset replacement values. An alternative is to assume that asset replacement values roughly equals book values. Q can then be calculated as total market value of equity plus book value of total debt, divided by book value of total equity plus book value of total debt (see, e.g., Wagner (2004), p. 13). The advantage of the latter approach is that all required data is readily available from commercial databases (e.g., Datastream). This multiple approach is a common valuation tool of market practitioners.

170

sensitivity to other segments' cash flows as the remaining segments. However, in an efficient ICM, the segment with the highest investment opportunities should have a lower sensitivity towards other segments' cash flows. The implication therefore is that ICMs are not efficient. Lamont/Polk (2002) find that an exogenously caused increase in the diversity of investment opportunities of a conglomerate's divisions reduces firm value. This finding is also inconsistent with the idea of efficient ICMs, which should be associated with an increase in firm value if investment opportunities become more diverse, because investments can be channelled into segments with high investment opportunities. Finally, in a German context, Kaserer/Ahlers (2000) find that ECOs in the form of primary offerings experience more positive abnormal announcement returns than ECOs in the form of secondary offerings, which they interpret as evidence for markets preferring a replacement of less efficient ICMs with more efficient ECMs. The empirical work on the inefficiency of ICMs is underpinned by a series of theoretical models attempting to explain the sources of the apparent inefficiency. Rajan/Servaes/Zingales (2000) develop a power model of internal resource allocation where divisional managers are more likely to invest in poorer projects as the diversity in the investment opportunities between the divisions increases. Firm headquarters anticipate this behaviour and transfer resources with the aim of equalizing investment opportunities between divisions to incentivise divisional managers to invest into efficient projects. As Rajan/Servaes/Zingales (2000) point out, this leads to a "secondbest attempt to head off a third-best outcome ''449. Scharfstein/Stein (2000) develop a two-tiered agency model of capital allocation within afirm. Divisional managers engage in rent-seeking activities to strengthen their bargaining position vis-a-vis the CEO. Since the opportunity costs of spending time on rent-seeking are lower for managers of poorly performing divisions, and since the CEO will attempt to limit rent-seeking of divisional managers, more resources will flow to lower performing divisions. These resources come not in the form of cash (as would be preferred by external investors), but in the form of capital allocations, because the CEO himself is not only a principal but also an agent (vis-a-vis the external investors). In De Motta's (2003) model, the value of an ICM is determined by a trade-off between two effects: On the one hand, divisional managers profit from the external assessment of the firm because the latter is allocated capital by external investors based on this assessment. Divisional managers are tempted to free-ride on this public good, and may undertake projects which benefit themselves more directly. On the other hand, divisional managers are incentivised to produce divisional profits, which are used as a basis for the internal capital allocation decision by the group's HQ. The resolution of the trade-off depends on the quality of information available to external investors and to HQ: If information quality for external investors is low, a multidivisional structure in which divisional managers are incentivised by HQ is preferable. If information quality for HQ is high, however, divisional managers won't need to produce divisional profits to signal qualification for capital allocation (as they do when information quality is low), and consequently a 449

Rajan/Servaes/Zingales (2000), p. 38.

171

single-segment structure is preferable in which incentives are placed on divisional managers by external investors. 45~ Another strand of research attempts to find evidence for the assumed value-destroying nature of corporate diversification, compared to the assumed value-creating nature of corporate refocusing. There is an overall trend of companies to increase their focus since the 1980s, and this focus increase is related to an increase in stock price returns (Comment/Jarrell (1995)). Companies announcing an asset sale on average experience positive APARs and show an increase in operating performance in the three years following the asset sale (John/Ofek (1995)). Johnson/Klein/Thibodeaux (1996) find that spin-off subsidiary firms grow faster, and spin-off parent firms also grow faster and have a higher cash flow margin following the spin-off. Consistently, Daley/Mehrotra/Sivakumar (1997) localise operational improvements in parent (not subsidiary) firms, and refine the finding by limiting it to cross-industry (vs. sameindustry) spin-offs. Desai/Jain (1999) confirm the finding of an increased operating performance, and also find evidence for an abnormally positive LTPP. Gertner/Powers/Scharfstein (2002) document an increase in sensitivity of investment in spun-off companies to investment opportunities (as measured by Tobin's Q) compared to the time before the spin-off. While this supports the notion of an increase in the efficiency of ICMs resulting from an increased industrial focus, the authors caution that the result may be also be driven by a self-selection bias: Firms with an inefficient ICM may choose to engage in a spin-off. Interpretability of results is therefore limited to the specific case of spin-off companies, rather than to ICMs in general. Ahn/Denis (2004) find that companies invest more heavily into high-Q segments following the spin-off, and the conglomerate discount decreases as a function of the increase in investment efficiency. Very similar findings are presented by Dittmar/Shivdasani (2003) for parent firms engaging in asset sales: Investment efficiency of the remaining divisions improves, and the reduction in the conglomerate discount is associated with this improvement.

6.3.2.2 Argumentsfor efficiency of lCMs Khanna/Tice (2001) study the reaction of diversified and single-segment firms to the market entry of Wal-Mart. Prior to the competitor's entry, the discount divisions of the diversified firms in the sample are more productive than the single-segment firms in the sample. After the entry, diversified firms react more quickly by either exiting the business, or investing more into it. If they decide to remain in the business, their investment decisions are more sensitive towards the productivity of their businesses. ICMs thus seem to function well in that resources are allocated in a timely manner from low- to high-productivity divisions. Khanna/Tice (2001) interpret this finding (in the spirit of Stein (1997)) as showing that the benefits from "winner-picking ''451 are more 450 451

See De Motta (2003), p. 1194. Stein (1997), p. 111.

172

valuable than the costs associated with the existence of an ICM. Their result that diversified firms are more productive than single-segment firms is confirmed more generally by Schoar (2002), who finds support for the hypothesis that conglomerate firms are more productive than single-segment firms. When conglomerates acquire new plants, the productivity of the latter increases, whereas the productivity of the conglomerate's existing plants decreases, with the total net effect being negative ("new toy effect"452). Stock prices reflect productivity differences between firms. Also, conglomerates seem to pay relatively higher wages and fringe benefits to employees, at the expense of shareholders. The conglomerate discount is thus interpreted as a function of firm productivity and relative wealth distribution, rather than inefficient ICMs. Campello (2002) analyses the workings of ICMs within financial conglomerates, and finds that (large) conglomerates do not subsidise poor-performing divisions at the expense of high-performing divisions in the case of an exogenous shock (tightening of interest rates by the Fed). 453 Billett/Mauer (2000) analyse the relationship between a measure of the efficiency of an ICM and the market reaction to the announcement of a tracking stock structure. They find a positive relationship: Markets seem to favourably judge the company's decision to increase the subsidiary firm's flexibility while at the same time maintaining the ICM and its associated benefits. Peyer (2002) analyses the frequency with which firms use external capital markets. He finds that companies with large and efficient ICMs use more external capital, and they are also able to do so at a lower cost of capital, compared to both single-segment firms and firms with inefficient ICMs. He interprets his finding as resulting from a lower informational asymmetry: Investors observing an efficient ICM realise that capital is allocated in a value-creating way inside the company's capital market, and are therefore willing to provide capital at a cost lower relative to cases where ICMs do not work efficiently. Investors are also more willing to supply capital to multisegment firms whose ICMs work efficiently than to singlesegment firms: Over- and underinvestment problems in multisegment firms are less pronounced than in single-segment firms, because management can choose to employ the capital in the most profitable segments. McNeil/Niehaus/Powers (2001) find that subsidiary firm managers are three times more likely to be replaced following a drop in their division's performance than are group CEOs, indicating a stronger disciplining function of ICMs vs. ECMs. This result holds even when other known corporate governance characteristics related to CEO turnover are controlled f o r . 454 However, the relationship between manager turnover and performance weakens as the diversification of the parent firm increases, indicating that 452

Schoar (2002), p. 2380. For small conglomerates, investment sensitivity for poor-performing divisions declines relative to high-performing divisions following exogenousshocks. 454 See McNeil/Niehaus/Powers (2001), p. 16. Control variables include the percentage of directors that are insiders (Weisbach (1988)), a dummy for whether the same person holds the CEO and the board chair positions (Goyal/Park (2000)), and a dummy for significant blockholder ownership (Denis/Serrano (1996)). 453

173

there is an optimal level of diversification beyond which ICMs become less efficient as a disciplining mechanism. The result implies that the efficiency of a firm's governance structure is subject to a trade-off between internal controlling mechanisms and external monitoring by shareholders and capital markets. Hadlock/Ryngaert/Thomas (2001) find that the negative market reaction to the announcement of an equity issue by a diversified firm is less pronounced compared to the same announcement by a single-segment firm. They argue that since market estimation errors for the performance of unrelated divisions are less than perfectly correlated, the absolute value of the percentage error will be smaller for diversified than for single-segment firms. As in the case of the negative view, the positive view of ICMs is supported by a series of theoretical models. Gertner/Scharfstein/Stein (1994) argue that ICMs offer increased monitoring possibilities and better asset re-deployability from poor to well-performing divisions, at the cost of reduced managers' entrepreneurial incentives. In Stein's (1997) model top management is able to finance otherwise constrained projects through "winner-picking ''455, i.e., the selection of the relatively highest value-adding projects. The optimal scope of the firm is determined by the trade-off between an increased ability to select profitable projects (increasing in the number of segments) and managements' capacity to do so (decreasing in the number of segments). Management is also assumed to do a better job in selecting profitable projects when segments are related rather than unrelated. In Matsusaka/Nanda's (2002) model conglomerate firms profit from the real options character of an ICM, which allows them to avoid having to resort to external financing (assumed to be associated with higher transaction costs) in more states of the world. The cost of an ICM lies in the managers' potential use of the additional financial flexibility for private benefits. In the model the only differentiating factor between multi- and single-segment firms is the distribution of control rights regarding the allocation of resources. This difference (rather than agency or power conflict considerations) is the source of the above trade-off, and of the resulting (in)efficiency of the ICM.

6.3.3

Biases in studies on ICMs and the conglomerate discount

The validity of findings on ICMs and the conglomerate discount has been called into question by a series of papers pointing to methodological difficulties when performing the required analyses. Hyland/Diltz (2002) find that only 72% of the reported business segment changes in the Compustat database correspond to actual changes as indicated by companies' annual reports. In a similar spirit, Martin/Saryak (2003) point out that companies have a considerable amount of flexibility in forming business segments, thus limiting cross-firm comparability of business segments. 456 Whited (2001) suggests that 455 Stein (1997), p. 111. 456 See Martin/Saryak (2003), p. 53.

174

the average Tobin's Q of all single-segment firms in a particular industry (often defined as 'industry Q') is a poor proxy for the investment opportunities of a conglomerate's business segment active in that particular industry, and shows that many of the results in the literature are not robust towards using a measurement-error-consistent estimator controlling for possible biases in Q. Villalonga (2004a) identifies a series of biases in the Compustat database arising from the noise in segment data and managerial discretion in segment reporting. Using establishment-level data from an alternative database (BITS) she finds that diversified firms trade at a significant premium. Chevalier (2000) calls into question the findings on value-destroying crosssubsidization within conglomerates by analysing a sample of stand-alone firms merging at a later point in time. These pairs of independent firms display some of the characteristics (e.g., sensitivity of one firm's investment to the other firm's cash flow, higher investment in low-Q business segments), usually attributed to crosssubsidization, before the merger, i.e., in a constellation in which these characteristics are not expected to occur. The conclusion is that similar findings for conglomerates cannot be only due to the conglomerate nature of the companies, since the findings are also found to occur between companies which are not part of the same conglomerate. As pointed out by Lamont/Polk (2002) 457 and Martin/Saryak (2003) 458, the findings on the existence of a conglomerate discount can have two (mutually non-exclusive) explanations: First, diversification itself could destroy value. This is the explanation implied by most studies finding a conglomerate discount and linking it to the diversified nature of the companies involved. Second, firms already discounted in value could decide to diversify. Closely linked to this is the realization that companies choose to diversify/refocus, introducing a self-selection bias into any empirical analysis for which few studies explicitly control. Campa/Kedia (2002) find a strong negative correlation between a company's tendency to diversify and its value. They show that when the endogenous nature of the diversification/refocus decision is taken into account, the conglomerate discount disappears and sometimes even turns into a premium. Firms react to their changing external environment and seem to take value-maximizing decisions: Campa/Kedia's (2002) model shows that if companies had not diversified, they would have performed even worse. Colak/Whited (2003) question the evidence on the improvement of investment behaviour following refocusing restructuring activities. Using estimators which control both for the endogeneity of the refocusing decision and the measurement bias in Q, they find no significant improvement in investment behaviour following the restructuring measure. Graharn/Lemmon/Wolf (2002) find that the decline in excess value of acquiring firms is largely because the businesses acquired traded at a discount before the acquisition. Similarly, Villalonga (2004b) finds that diversified companies traded at a discount prior to becoming diversified.

457 See Lamont/Polk (2002), p. 52. 458 See Martin/Saryak (2003), p. 53.

175

Another line of literature attempts to explain the conglomerate discount as the result of forces other than the ICM. To the degree that these attempts are successful, the case for the inefficiency of ICMs loses some of its force. Sanzhar (2003) analyses "pseudoconglomerates ''459, defined as multi-divisional firms with all divisions operating in the same industry. He finds a discount similar to unrelated conglomerates, i.e., multidivisional firms with divisions operating in different industries. Since investment opportunities within pseudo-conglomerates are assumed to be identical across segments, and also identical to the investment opportunities of single segment firms operating in the same industry as the pseudo-conglomerate's segments, inefficient ICMs cannot be the cause of the discount: Firms with identical investment opportunities should not be faced with distorted investment incentives as modelled by Rajan/Servaes/Zingales (2000); consequently capital does not need to (and also cannot) be allocated from highinvestment opportunity segments to low-investment opportunity segments. Sanzhar (2003) also finds that conglomerates starting to report a higher number of business segments (resulting from a new segment reporting rule) begin to trade at a discount. Doukas/Kan (2004) find a link between excess value losses and declines in cash flows following acquisitions. The relationship is more pronounced in the case of unrelated acquisitions. The existence of a conglomerate discount is thus linked to the fundamental firm value (as proxied by cash flows). Bernardo/Chowdry (2002) develop a company lifecycle model based on a real options framework. In their model, young firms undertake investments to learn about their skills and capabilities, and, depending on the outcome, develop either into multisegment firms (when they find that they possess general skills) or into single-segment firms (when they find they possess specialised skills). Bernardo/Chowdry (2002) conclude that young firms about whose skills there exists uncertainty are more valuable than mature firms whose skills are better known, if both firms have the same tangible resources. This is equivalent to a financial option being worth more than an otherwise identical option whose underlying is less volatile. The conglomerate discount is thus explained as the result of the lower value of learning options. Similarly, Matsusaka (2001) develops a model in which a firm's diversification is a search process for the right match of firm capabilities to business requirements. If a company has not (yet) found the right match, it will likely trade at a discount and continue to diversify; hence, the idea of diversification as a value-maximizing strategy is compatible with the existence of a conglomerate discount. This idea relates to the previously discussed endogeneity problem: Diversification does not cause the discount; rather companies trading at a discount will diversify (potentially aiming to remove the discount). Maksimovic/Phillips (2002) find that plants of conglomerate firms are less productive than those of single-segment firms. They also find value-maximizing behaviour of conglomerates as indicated both by large plants in the largest segments being the most productive, and segments growing as a function of their relative investment 459

Sanzhar (2003), p. 2.

176

opportunities. Their findings indicate conglomerates invest into those businesses where they have comparative advantages. Investment decisions do not seem to be subject to agency problems as implied by supporters of inefficient ICMs. Consequently, the conglomerate discount is not the result of inefficient ICMs, but rather of less productive plants, which can exist in equilibrium because of decreasing returns to scale. Ang/Cheng/Nagel (2003) compare related and unrelated acquisitions. They find that overvaluation and premiums paid in unrelated acquisitions are higher than in related acquisitions, and they explain the conglomerate discount as being the result of the acquisition of an overvalued target. Apparently, companies doing a related acquisition are better able to estimate a fair price for the target than are companies doing an unrelated acquisition. A potential explanation is that the ability to appropriately value an acquisition requires knowledge of the industry; such knowledge is more likely to exist when the acquiring firm is in the same industry as the target. Lack of this knowledge leads to a discount for the newly combined entity as the overvaluation of the acquired target is corrected through a price reversal of the combined entity.

6. 3.4

ICMs and the conglomerate discount across time and geography

Although the evidence is inconclusive, there are indications that the level of efficiency (and thus the value) of ICMs is non-stationary across time and geography. The conglomerate discount seems also to have existed in the 1960s when many of the conglomerates were formed (Servaes (1996)), and declined to almost zero in the 1970s, with the reason for the decline being unclear. In contrast, Matsusaka (1993) finds positive abnormal returns to the announcement of diversifying acquisitions in the 1960s. Hubbard/Palia (1999) confirm this finding and explain it by the higher relative value of ICMs in the presence of less informationally well-developed ECMs. Consistently, Deloof (1998) produces evidence for the value of ICMs in Belgium, where complex holding structures substitute for the role of a (non-existing) well-functioning ECM. Geographical (as opposed to industrial) diversification of US companies is associated with a valuation premium (Bodnar/Tang/Weintrop (1998)), partially as a result of higher growth opportunities in non-domicile countries. For non-US companies, a discount of the size similar to the one in the US (13-15%) has been found in Japan (10%) and UK (15%), but not in Germany (Lins/Servaes (1999)), where concentrated ownership in the hands of insiders seems to support valuations and prevent a conglomerate discount from materializing. The same authors in a later paper (Lins/Servaes (2000)) find an average conglomerate discount of 8% across seven emerging markets, limited to companies with an insider ownership of between 10% and 30%. They find that the discount is larger in countries with less developed ECMs, rejecting the hypothesis of a higher value of ICMs in countries with less developed ECMs. Instead, they interpret their findings as supporting the notion of minority shareholders being expropriated in a diversified firm structure. In contrast, Khanna/Palepu (2000a) in their sample of Indian firms find a quadratic relationship

177

between group diversification and firm performance: Firms affiliated with small and medium-sized groups perform worse than single-segment firms, but those firms affiliated with the largest groups show the best performance of all firms. The authors interpret this result as being evidence for the high value of ICMs (which only the largest groups can offer) in countries where ECMs are less developed. In addition to the inter-temporal and geographic non-stationarity of the conglomerate discount, there is a growing realization that the (in)efficiency of ICMs is attributable to certain organizational and financial characteristics of the firms. Through the design of compensation schemes firms are able to create incentives for divisional managers to provide more accurate information, thus enabling a more efficient allocation of resources in ICMs (Wulf (2002)). High leverage leads to distorted capital allocations and hinders the functioning of ICMs because investment sensitivity towards cash flow (rather than towards Q) increases as a result of the pressure to service the outstanding debt (Peyer/Shivdasani (2001)). The value of an ICM for a firm increases with the number of valuable trade secrets and with the degree to which the segments' investment programs are sensitive to delay or interruption (Liebeskind (2000)). To summarise, the case for (in)efficiency of ICM remains open. In general, ICMs create value when capital is allocated to segments which are both financially constrained and possess good investment opportunities. ICMs do not create value when capital is allocated to financially unconstrained segments. ICMs destroy value when capital is allocated to segments without good investment opportunities. 46~Progress is likely to be made along two lines: First, Steven Kaplan suggests that future analyses should use more detailed data, rather than "large Compustat-type data sets ''461, potentially at the cost of a smaller sample. Second, Stein (2003) suggests that rather than considering the question of ICM efficiency as a yes/no dichotomy, researchers should analyse the crosssection of results, and ask "under what conditions is an internal capital market most (or least) likely to add value relative to an external capital markets benchmark? ''462. This study broadly follows the second suggestion, as described in the following section.

6.4

Data and specific analyses

6.4.1 Sources of data All data required for the construction of the ICM size and efficiency measures is from Datastream. Local currencies are converted to Euro using fiscal-year end exchange rates. Business segment industry classification is based on SIC codes, which are taken from Datastream where available. 463 The remaining cases are hand-filled based on 460 See Billett/Mauer (2003), p. 1193-1195. 461 Villalonga (2003), p. 17, there attributed to Steven Kaplan. 462 Stein (2003), p. 141. 463 The reason why SIC codes are used, rather than NACE codes as in the LTOP and LTPP analyses, is that Datastream only provides SIC codes for a firm's business segments. This data is required to identify single-segment firms.

178

personal judgement following consultation of the respective firm's annual report. The various size and efficiency measures are based on existing literature, which has used these measures in different contexts. Measures of ICM size include excess capital expenditure based on Billett/Mauer (2000), sales and asset-based Herfindahl indices as suggested by Peyer (2002), and diversity as developed by Rajan/Servaes/Zingales (2000). Measures of ICM efficiency include profitability-weighted excess capital expenditure as developed by Billett/Mauer (2000), relative value added by allocation (RVA) and absolute value added by allocation (AVA) as constructed by Rajan/Servaes/Zingales (2000), Q-sensitivity of investment as developed by Peyer/Shivdasani (2001), and cash flow-sensitivity of investment as constructed by Peyer (2002). A critical component of sOme of these measures is Tobin's Q (henceforth referred to alternatively as Tobin's Q or simply Q). Tobin's Q is defined as the market value of a firm's assets divided by their replacement v a l u e . 464 Lang/Poulsen/Stulz (1995) use Q as a proxy for a firm's investment opportunities. 465 The economic intuition is that if the market value of assets exceeds their replacement value, this excess part of the firm's value must reside in less tangible assets such as real options and investment opportunities 466, which are hoped to contribute to cash flows in later periods. Q can be estimated in two different ways. The more complex approach employs detailed balance sheet data and adjustments for deviations of accounting data from replacement values. An example of this approach are Lang/Stulz (1994), based on Smirlock/Gilligan/Marshall (1984), who explicitly model asset acquisition and depreciation schedules and correct for price inflation to calculate the asset replacement values. The simpler approach is suggested by Chung/Pruitt (1994), and assumes that asset replacement values roughly equal book values. Q can then be calculated as total market value of equity plus book value of total debt, divided by book value of total equity plus book value of total debt. 467 The advantage of the simple approach is that all required data is readily available from commercial databases (e.g., Datastream). Chung/Pruitt (1994) find the Q values based on the simple approach approximate Q values based on the more complex approach very well. DaDalt/Donaldson/Garner (2003) find that the more complex approach, besides being costly to implement in terms of time and resources, can introduce a selection bias due to non-availability of some of the data required for the calculations. They suggest using the simple approach "except

464 See DaDalt/Donaldson/Gamer (2003), 535. 465 See Lang/Poulsen/Stulz (1995), p. 12. 466 These are intangible assets. However, they are different from other intangible assets (such as goodwill) in that they are not recorded on the firm's balance sheet. The denominator of Tobin's Q includes only intangible assets recorded on the balance sheet, and the numerator of Tobin's Q includes both intangible assets recorded on the balance sheet, and intangible assets not recorded on the balance sheet (i.e., investment opportunities). Tobin's Q is therefore a measure of the latter. 467 An example of a practical implementation is Wagner (2004), p. 13.

179

when extreme precision of the Q estimate is the paramount consideration ''46s. This study therefore employs the simple approach to estimate Q. The idea behind and construction of each of the ICM measures is described next.

6. 4.2

Size measures

Firm excess capital expenditure (FECE): This measure is a modification of the ICM measure used by Billett/Mauer ( 2 0 0 0 ) 469. FECE is defined as the sum across segments

of segment capital expenditure minus segment cash flow, with each component weighted by segment sales. Segment cash flow can be either before-tax cash flow (CF) or after-tax cash flow (ATCF), resulting in FECE1 and FECE2, respectively. Intuitively, using before-tax cash flow seems a better comparison because it directly compares investment-related (as opposed to financing-related) cash flow streams. However, it could also be argued that taxes are a cost of the business segment which has to be taken into account. Consequently both before and after tax cash flows are used. Two additional restrictions are placed on the measure: First, a segment's excess capital expenditure is not allowed to exceed the segment's total capital expenditure. Second, when segment cash flow exceeds segment capital expenditure, the segment's excess capital expenditure is set to 0. The relevance of these two restrictions will be illustrated shortly. The two measures are

N I ( Capex~ Sale& ' Sales~

FECE1 = Z Max Min ~=1

Salesi

II

F E C E 2 = ~-' Max Min .=

Sales i

,0

)

ATCCaexi) / S a l e s i , Salesi

,0 ,

where Salesi, Capexi and A TCFi are the sales, capital expenditure and after-tax cash flows of segment i, respectively.. Because A TCFi is not commonly reported in commercial databases, it must be imputed. Imputation is based on median values for single-segment firms from the same industry as segment i. The assumption is that segment i is economically equivalent to single-segment firms in the same industry. This assumption has been called into question 4v~ but is common in the literature on the 468 See DaDalt/Donaldson/Garner (2003), p. 551. 469 See Billett/Mauer (2000), p. 1467-1468. The original measure is used later as a proxy for the efficiency (rather than size) of an ICM. The modification consists in not weighting excess capital expenditure by its profitability, as is done in the original proxy. This allows constructing a measure of the size (rather than the efficiency) of an ICM. Apart from this, all calculations and rules are identical to Billett/Mauer (2000). 470 See Whited (2001), who raises doubts as to whether a segment's investment opportunities are identical to investment opportunities of single-segment firms in the same industry.

180

conglomerate discount and ICMs. 471 Median (rather than mean) values are used to prevent the impact of outliers. is defined as

ATCFi

ATCFi I EBITi - I Salesi*lndik(I"nterest~]llnd(Tax)Jmf ~]+Oisal*es 1=

where

t~,EBTJJmf

Indi( Interest'] is the median of the interest to sales ratios of all single-segment k. ms

firms in segment i's industry (where industry as usual is defined on the two digits SIC codelevel),

Indi( )EBT Tax

isthemedianofthetaxtoeamingsbeforetaxratioofallfirms

EBITi

Di

in segment i's industry, is segment i's EBIT, and is segment i's depreciation and amortization expense. FECE3 and FECE4 are identical to FECE1 and FECE2, except that weighting is based on total assets, rather than sales. A numerical example will provide some economic intuition for the two restrictions in the previous paragraph. Consider a firm with two business segments A and B (case 1 in Appendix 46). Each segment has a CF of 100. Segment A has an ATCF of 60 while segment B has an ATCF of 70. The effective tax rate can differ across segments because tax payments are imputed based on the median tax rate in the respective business segment's industry. Capital expenditure is 110 and 20 for business segment A and B, respectively. According to the above formula, excess capital expenditure for segment A is 50, (=Min(ll0-60,110), and 0 for B (=Max(20-70,0)). Total excess capital expenditure is thus 50. The first restriction states that excess capital expenditure cannot exceed total capital expenditure. This restriction is required for cases where CF and ATCF are negative. To understand restriction one, consider case 2a, the same example with the signs for the CF and ATCF reversed. Excess capital expenditure for A and B is now 110 (=Min(110+60,110)) and 20 (=Min(20+70,20)), respectively, and total excess capital expenditure is 130. Case 2b doubles CF and ATCF. Again, excess capital expenditure for A and B is 110 (=Min(110+120,110)) and 20 (=Min(20+140,20)), respectively. If the first restriction were not used, excess capital expenditure for A and B would be 230 (=110+120, no min-function) and 160 (=20+140, no min-function), respectively. The point is that in case 2b, losses for the two business segments are double in size compared to case 2a; however, this should not increase the size of the ICMs because the same capital expenditures are financed (presumably through external financing). While losses are double, the workings of the ICM are the same in the two cases. The size of the losses therefore should not impact the ICM measure. This is guaranteed by the first restriction. The second restriction defines a segment's excess capital expenditure as 0 when its capital expenditure is less than its cash flow. To 471

Berger/Ofek (1995) calculate the imputed value of a conglomerate based on multiples derived from the median of single-segment firms. A similar approach is used by Campa/Kedia (2002). Billett/Mauer (2000) use median industry values to calculate segmentafter-tax cash flows.

181

understand restriction 2, consider again the numbers from case 1. If segments with capital expenditure less than their own cash flow were not defined to have 0 excess capital expenditure (see case 3 in Appendix 46), excess capital expenditure for segment B would b e - 5 0 (=Min(20-70,20)). Total excess capital expenditure would thus be 0, since the positive excess capital expenditure of A would be cancelled by the negative excess capital expenditure of B. Intuitively, this would not make sense because there is clearly some cross-subsidization occurring. Restriction 2 therefore assures that excess capital expenditure cannot become negative for any segment. Excess capital expenditure is a measure of the degree to which investment in one segment is financed by cash flows from the firm's remaining segments, and therefore of the total activity in the ICM. A larger value for the variable thus indicates a larger ICM.

Herfindahl-index of segment sales (assets):

As suggested by Peyer (2002), the index measures the concentration of sales (assets) across the various segments of the firm. The lower the index, the lower the concentration and therefore the larger the potential for cross-segment subsidization and thus the size of the ICM.

Diversity: Diversity

as developed by Rajan/Servaes/Zingales (2000) is calculated as the standard deviation of a firm's asset-weighted Q's divided by the equal-weighted 472

average Q"

Diversity =

i=l

n- 1

N

~.~qi i=l /7

where n is the number of business segments, and wi is the ratio of segment i's assets to total firm assets. The measure is intended to proxy for the difference in investment opportunities and resources across a firm's various business segments. A basic assumption, as indicated above, is that Q is a valid proxy for investment opportunities, which are assumed to account for the difference between market value and replacement value of assets as measured by Q. If Q proxies for investment opportunities, then a higher dispersion of the segments' Q's around their weighted average is an indicator of more diverse investment opportunities across the segments. A larger value for this variable therefore indicates a larger ICM.

472

Rajan/Servaes//Zingales do not justify why the numerator is asset-weighted while the denominator is equal-weighted. A potential reason is that if the denominator were also asset-weighted, than an increase in a high-asset segment's Q would lead to a lower increase in the diversity measure, compared to when the denominator is equal-weighted (denominator would increase more because of higher asset weight, compared to equal-weighting). Therefore the ICM measure would be less sensitive to firm differences in investment opportunities, the estimation of which is the aim of the measure.

182

6.4.3 Efficiency measures Profitability-weighted excess capital expenditure (PECE): PECE

is defined as the "sum across segments of the product of each segment's excess capital expenditures and its industry-adjusted return on investment ''473. Thus, in addition to capturing the size of the ICM as described in the FECE measures, this variable also reflects the profitability of cross-segment investment. Again, excess capital expenditure is defined as segment capital expenditure minus segment cash flow, and segment cash flow can be either before-tax cash flow (CF) or after-tax cash flow (ATCF), resulting in PECE1 and PECE2, respectively. The same two restrictions are applied as in the case of PECE: First, a segment's excess capital expenditure is not allowed to exceed the segment's total capital expenditure. Second, when segment cash flow exceeds segment capital expenditure, the segment's excess capital expenditure is set to 0. To allow cross-firm comparisons, all variables are scaled by sales.

PECE~ =

N

Salesi

- lndil SalesiCFil l * Max(Min(Capexi - CFi' Capexi

i=1

PECE2 =

1

~ Sales i

,~-~lSalfF~si_lndil CFilI,Max(Min(fapexi_ATfFi,fapexi),O ) ~Salesi mf i=l

i=1

PECE3 and PECE4 are identical to PECE~ and PECE2, except that weighting is based on assets, rather than sales.

Absolute

value added by allocation (A VA): AVA as developed by Rajan/Servaes/Zingales (2000) measures the value created by a firm's ICM. Business segment i's investment ratio is its capital expenditure divided by its book value of total assets,

Capexi. t~Ai

The investment ratio is

industry-adjusted by

subtracting the industry

investment ratio, defined as the median capital expenditure to total assets ratios of all single-segment firms in segment i's industry,

CapexSiS Each BASS 9

segment's i n d u s t r y -

adjusted investment ratio is weighted by the difference of the segment's Q and one. Using these weights (as opposed to using the difference of the segment's Q to the average Q of all segments, which is done in the RVA measure presented next) aims to capture another aspect of value addition by a diversified firm: If a more diversified firm

473

Billett/Mauer(2000), p. 1467.

183

is able to raise more funds than a single-segment firm, it may invest more capital on average across all segments. 474 Segments contribute positively to the A V A measure if they invest more than single-segment firms and have a Q higher than 1, or if they invest less than single-segment firms and have a Q lower than 1. They contribute negatively if they invest more than single-segment firms and have a Q lower than 1, or if they invest less than single-segment firms and have a Q higher than 1.

i=l

AVA=

BA i (q~ - 1 Capexi BAi

CapexSS B Ass

M

BAj j=l

Relative

value

added

by

allocation

(R VA):

RVA

as

developed

by

Rajan/Servaes/Zingales (2000) is a measure based on the investment ratio of a firm's business segments. A business segment's investment ratio is industry-adjusted as in the case of AVA. Further, the investment ratio is firm-adjusted by subtracting the average industry-adjusted investment ratio across all segments of the firm, ~w i i--1

BAi

BA ss

. Firm adjustment serves to analyse whether a company

invests more in high-Q segments than in low-Q segments. To assess whether crosssegment transfers are efficient, a segment's firm- and industry-adjusted investment ratio is weighted by the difference between the segment's Q and the firm's equal-weighted average Q. The sum across all segments of these weighted firm- and industry-adjusted ratios is the RVA:

~-~BAi (qi RVA=

CapexiBA i

CapexSi y - Z

CapexiBAi Capexss

M

BAj j=l

where BAi is the book value of segment i's assets,

CapexSi s is the median capital BA ss

expenditure to assets ratio of single-segment firms in segment i's industry, and wi is segment i's book value of assets divided by the firm's total book value of assets. R V A is positive if a firm invests more in segments with relatively high investment

474

See Rajan/Servaes/Zingales (2000), p. 69. RVA tends to underestimate the value created by diversified firms in cases where the total amount of investment increases as a result of the diversified nature of the firm. They suggest an additional measure, the absolute value added by allocation (AVA), to account for this.

184

opportunities (controlling for industry investment levels), and less in segments with relatively low investment opportunities.

Q-sensitivity of investment: This variable proposed by Peyer/Shivdasani (2001) measures the deviation of sales-weighted segment capital expenditure to the same ratio on the firm level. If Q proxies for investment opportunities, this measure will be positive if segments with above-average Q receive more investments than segments with below-average Q, indicating an efficient ICM.

M

N ZSalesi(qi

Capexi

--q)*

- -

M

Sales i

i=1

Zi=ICapexi

Z Sales i i=1

Q - Sens =

N

Z Sales i i=1

Cash flow-sensitivity of investment: This measure proposed by Peyer (2002) is similar to Q-sensitivity except that the weighting of the firm-adjusted capex-to-sales ratio is performed using the difference of the segment cash flow-to-sales ratio and the firm cash flow-to-sales ratio. Analogously, the measure will be positive if segments with aboveaverage cash flow-to-sales ratio receive more investments than segments with belowaverage cash flow-to-sales ratio, indicating an efficient ICM.

U ZSalesi /=1

CF - Sens =

CFi _

_

Sales i

ZCFi i=1 , Ca-exi M ~ ~ Z Sales i i=1

ZCapexi i=1 M E Sales i i=1

M

Z Sales i i=1

The ICM measures are constructed using accounting data from the year prior to the ECO. The required business segment level data (segment level sales, EBIT, depreciation, total assets, capital expenditure) is retrieved manually from annual reports since an electronic database containing segment level data could not be made available. Segment level cash flows are approximated as E B I T D A minus capital expenditure. Since publication of segment level data is severely limited 475, the construction of the various ICM measures is only possible for 20-40% of all sample firms, depending on the specific ICM measure. The number of companies included in the regressions is 475

The legal publication requirements differ across countries. In Germany, for example, companies have only been required to publish segment level data since 1999.

185

further reduced by the exclusion of all firms with at least one business segment whose SIC code begins with 6 (financial institutions). 476 In accordance with previous studies, business segment Q is proxied by the median Q of single-segment firms in the same industry. 477 Industry is defined on the two digits SIC code level. If there are less than three firms in the same two digits industry, the median Q of single-segment firms in the same one digit industry is used. The same algorithm is used for the calculation of industry ratios (cash flow-to-sales, taxes-to-EBT, interest-to-sales, capex-to-total assets). In addition to constructing ICM measures with data from the fiscal year prior to the announcement of the ECO, the same measures are constructed using data from the first fiscal year following the implementation of the ECO. This allows the use of the change in the measures as an independent variable. Two additional hypothesis can be based on these change measures (in addition to the two hypotheses based on the ICM measures from the fiscal year prior to the ECO): First, if investors react positively to the announcement of an ECO by a parent firm whose ICM becomes smaller following the ECO, this implies that investors view ICMs as value-destroying. Alternatively, if investors react positively to the announcement of an ECO by a parent firm whose ICM becomes larger following the ECO, this implies that investors view ICMs as valuecreating. Second, if investors react positively to the announcement of an ECO by a parent firm whose ICM becomes more efficient following the ECO, this implies that investors are able to distinguish efficient from inefficient ICMs. Equivalently, if investors do not react systematically to the announcement of an ECO by a parent firm whose ICM becomes larger following the ECO, this implies that investors are not able to distinguish efficient from inefficient ICMs. As in the case of the ICM measures from the fiscal year prior to the ECO announcement, a systematically positive reaction of investors is unlikely, as this would imply consistently positive reactions to increases in ICM efficiency.

6.5

Empirical results

The basic assumption in the hypotheses developed in previous sections is that the size and the efficiency of the ICMs cause the level of APARs to change. ICM measures are used both as absolute values and as change values, where the change is calculated as the difference in the respective ICM measure from the fiscal year prior to the ECO to the first fiscal year following the ECO. The hypotheses are tested in a multivariate regression framework similar to the one described in chapter 3 on the STPP.

476

477

Financial segments are excluded because their accounting data is not comparable to non-financial segments. Of the 89 firms for which business segments can be identified, seven are excluded because at least one segment has a financial SIC code. For a discussion of this point see Whited (2001), p. 1667-1668.

186

6.5.1

Investors'perspective on ICMs

The APAR is the dependent variable, and the size and efficiency ICM measures are the independent variables. Other variables found to have significance in the explanation of the cross-section of APARs are used as control variables, including the level of preevent informational asymmetry, a dummy variable indicating whether parent and subsidiary firm are from the same industry, relative transaction size, a period dummy indicating whether the ECO occurred before or after 1998, a regional dummy, a motivational dummy, return on assets, and a dummy variable for the stake retained. 4v8 Appendix 47 shows various models explaining APAR as a function of ICM size and efficiency measures from the year prior to the announcement of the ECO. Model I is the base model incorporating only those variables found to have explanatory power in the STPP analysis. Models II-XVI represent the base model augmented by each one of the 15 ICM variables, respectively: Models II-VIII incorporate the seven ICM size measures, and models IX-XVI incorporate the eight ICM efficiency measures.

6. 5.1.1

Measures of ICMsize from the year prior to the ECO

For the ICM size measures, all coefficients for the firm excess capital expenditure (FECE) variables have a positive sign, which is statistically significant for the FECE1 variable at least at the 10% level (p=0.0544). This finding indicates that investors react more positively to ECO announcements by firms with larger ICMs, as measured by the FECE variable. An increased amount of excess capital expenditure before the ECO, possibly subsidised by the remaining business segments, is seen as negative: Partially closing the ICM, thereby potentially limiting further excess capital expenditure, leads to a positive abnormal return. Thus, investors seem to view the existence of ICMs as negative. One caveat regarding this interpretation arises from the construction of the FECE measure: It is assumed that the excess capital expenditure measured by the FECE variable for one business segment is financed by the cash flow coming from the remaining business segments. However, it could equally be the case that the excess capital expenditure is financed by external capital raised by the firm and allocated towards its business segments. Intuitively, setting off 'positive' and 'negative' excess capital expenditure against each other could help determine whether the capital comes from internal or external sources. However, as indicated in the numerical example in the previous section, this would further weaken the interpretability of the FECE measure: Company A with no excess capital expenditure would have the same value for FECE as a company B with an excess capital expenditure of +x in one business segment a n d - x in another business segment, although an ICM clearly exists in B but not in A. Interpretability of results is therefore limited and based on assumptions regarding the ability of the variables to correctly measure the size of the ICM. The coefficients of all remaining size variables are insignificant.

478

See section 3.5.3 for more details on these variables.

187

6.5.1.2 Measures of lCM efficiency from the year prior to the ECO For the ICM efficiency measures, models IX to XII use the profitability-adjusted excess capital expenditure (PECE) measure. All four models have the expected negative sign, and all four coefficients are significant at least at the 5% level. The PECE variable essentially measures the excess capital expenditure of business segments, weighted by their respective profitability. A higher value for PECE in combination with a lower value for the APAR therefore indicates that investors accord lower abnormal returns to the announcement of ECOs by parent firms whose ICMs have been working efficiently prior to the announcement. While this result is intuitive, it is simultaneously surprising because it indicates that investors actually take the efficiency of ICMs into account when forming their judgement on the potential value creation in an ECO as expressed in the APAR. This is remarkable because investors have been shown to fail at what seem to be analytically much easier tasks to perform. 479 That investors would go through the lengthy process of constructing the fairly data-intensive measures of ICM efficiency as is done in this study seems doubtful. A potential explanation is that there exist other measures, more heuristic in nature and thus more readily available to investors, which are correlated to this study's ICM measures and therefore produce results for investors similar to the results in this study. This is not a negative judgement on the validity of this study; it merely indicates that simpler proxies of ICM efficiency may exist, and identifying these may prove a valuable avenue for future research. Results on the two value-added measures, RVA and AVA, are insignificant, as is the coefficient on Q-sensitivity. The coefficient on the cash flow-sensitivity variable is significantly negative at the 10% level. Cash-flow sensitivity is positive if segments with an above-average cash flow-to-sales ratio receive more investments than segments with a below-average cash flow-to-sales ratio, which is assumed to be a signal for an efficiently working ICM. A significantly negative coefficient for the cash-flow sensitivity measure therefore confirms the result found using the PECE measures: Investors accord lower abnormal returns to an ECO announcement by a firm whose ICM has been working efficiently prior to the ECO. This supports the notion of firms with more efficient ICMs profiting less from ECOs, and concurrently the idea that investors are able to differentiate between efficient and inefficient ICMs. A caveat with regard to the cash-flow sensitivity measure is that again there may be doubts that it correctly measures what it intends to measure: A higher cash flow-to-sales ratio may be indicative of higher investment opportunities, which is the underlying assumption of the above interpretation of results; however, other constellations are thinkable: Consider a company consisting of two business segments. Segment A is a high-technology segment with higher margins, and segment B is an industrially more mature segment with lower 479

The literature abounds with corresponding examples; see section 2.3.4 for a list of such pricing anomalies. A particularly striking case in point in the context of ECOs is presented by Lamont/Thaler (2003), who find that in the US pre-ECO combined parent/subsidiary companies are valued less than the subsidiary to be carved out in the 1998 to 2000 period. According to the authors, this casts doubt on the market's ability to add and subtract (part of the phenomenon may be explained by short-selling restrictions preventing effective arbitrage).

188

margins. Cash flow-to-sales ratio in the former may be higher (depending among others on the segment's capital expenditure) than in the latter, but the cost of capital may also be higher due to higher business risk. If the company then invests more in segment A than in segment B, the cash-flow sensitivity measure will be positive, indicating an efficient ICM, but value is actually destroyed because the economic value added in segment A is lower, due to relatively higher costs of capital for A compared to B. The essential assumption for an interpretation of results as performed above therefore is that the cash flow-to-sales ratio is a valid proxy for value-adding investment opportunities.

6.5.1.3 Measures of changes in ICMsize Appendix 48 shows various models explaining APAR by the change in ICM size and efficiency measures from the year prior to the announcement of the ECO to the year in which the ECO occurred. Again model I is the base model incorporating those variables found to have explanatory power in the STPP analysis, and again models II-XVI represent the base model augmented by each one of the 15 ICM change variables, respectively. For the variables indicating changes in the size of the ICM, the coefficient on the Herfindahl-index of total assets is significantly positive at least at the 5% level (p=0.0263): An increase in the concentration of total assets in the firm is associated with higher APARs. The economic interpretation is that investors approve of a reduction in the size of the ICM. The assumption is that the ECO has caused the concentration index to increase, because the firm's assets are concentrated in fewer business segments following the separation of the subsidiary from the parent firm. This is equivalent to a decrease in the size of the ICM, because the parent firm is less diverse following the ECO, compared to the time prior to the ECO. A caveat here is that obviously there may have been other reasons for a change in the index: A company may have shifted assets away from lower-asset segments to higher asset segments, increasing the index. Arguably, this would also represent a reduction in size in the ICM, but the point is that this reduction may have nothing to do with the ECO. The critical assumption therefore is that the increase in the concentration of the assets results from the separation of the subsidiary from the parent firm. The remaining variables on the change in ICM size measures have insignificant coefficients.

6. 5.1.4 Measures of changes in ICM efficiency For the variables indicating changes in the efficiency of the ICM, the coefficient on Qsensitivity has a positive sign, significant at least at the 5% level (p=0.0221): An increase in ICM efficiency is associated with higher APARs. It supports the hypothesis that investors can anticipate changes in the efficiency of ICMs likely to result from an ECO, and also that they take these changes into account when forming their judgement on the potential value creation in an ECO as expressed in the APAR. The same caveat

189

as described in the above paragraph on the PECE measure applies here: It seems questionable whether investors construct this specific measure of ICM efficiency when reacting towards ECO announcements, and simpler proxies may be available and more suitable. The remaining coefficients are statistically not different from zero.

6.5.1.5 Interpretation of results There is some evidence that investors react more positively to the announcement of an intended ECO when the size of the announcing firm's ICM is large. Similarly, there is some support for the idea that investors react more positively when the size of the ICM decreases following the ECO. These findings imply that investors generally consider the existence of an ICM to be value destroying: Investors are happy about the announcement of a partial closure of ICMs, and prefer a decrease in size following the ECO. At the same time, investors react less positively to the announcement of an intended ECO when the announcing firm's ICM has been working efficiently prior to the announcement. Also, investors react more favourably to an ECO announcement when the efficiency of the firm's ICM increases following the ECO. These latter findings indicate that investors are at least partially able to discern efficient from inefficient ICM. At least three caveats are in order with respect to the results of this analysis. First, a considerable part of the original sample firms is excluded from the analysis because of insufficient data for the construction of the ICM measures. This is due to the limited publication of segment level data by firms, particularly prior to the mid-1990s, and introduces the possibility of a selection bias. (Non-)significance of results may hence be driven by the fairly small number of firms in the analysis. Second, the ability of the constructed variables to measure correctly the size and efficiency of ICMs may be limited. All results are based on the assumption that the measures of ICM size and efficiency capture what they set out to capture, which requires some restrictive assumptions as described above. Evidence of how specific capital allocation decisions are taken within firms is severely limited for outside investors. Even if such information were available, a benchmark would need to be established to be able to judge the efficiency of any particular investment decision: The discussion of what valuation method is appropriate for investment projects 48~ highlights the intrinsic difficulty of judging what constitutes an efficient investment. Consequently, researchers must resort to more or less crude proxies. As always in empirical research, it is hoped that finer measures will be developed, possibly using more detailed data. This will come at the likely cost of not being able to use commercial databases, which will increase research time and resource requirements dramatically, if a statistically sufficient number of firms is to be analysed. At the same time, investors may find some of the ICM measures to complex to model in everyday investment decisions, and therefore may not employ the 480

See Dixit/Pindyck (1994) for a discussion of the benefits of an options-based approach to investment project valuation vs. the more traditional NPV approach.

190

specific ICM measures used in this study: However, while investors may use simpler measures which are different from those used in this study, the fact that this analysis produces significant results supports the notion that the size and the efficiency of a parent firm's ICM prior to the ECO is indeed taken into account in some form or other. Third, many results of the analysis are derived under the critical assumption that Q is a good proxy for investment opportunities, and in particular that the Q of single-segment firms is a good proxy for the Q of a conglomerate firm's business segments. This assumption, while widespread in the literature regarding the conglomerate discount, has been severely criticised: Whited (2001) suggests that industry Q is a poor proxy for the investment opportunities of a conglomerate's business segments. Further research seems required before a final judgement can be passed. In particular, applying the algorithms in this study to a set of US ECOs, which are both more numerous and for which business segment level data is more easily available (e.g., in the Compustat Business Segment Information database), could yield valuable insights on whether the hypotheses brought forward are further supported.

6.5.2

Conditions of lCM efficiency

One of the key results of the above analysis is that investors seem at least partially able to discern efficient from inefficient ICMs. A natural follow-up question is to ask under what conditions ICMs are efficient. Literature offers two contradictory views: In Rajan/Servaes/Zingales's (2000) power struggle model of ICMs, diversified firms with business segments similar in resources and opportunities will channel investments from low-opportunity segments to high-opportunity segments. ICM thus work efficiently. However, when there is a high degree of diversity of resources and opportunities between business segments increases, investment flows reverse from high-opportunity to low-opportunity segments, thereby decreasing firm value. The reason for this reversal is that divisional managers do not have an incentive to undertake efficient investments if diversity is high, because they would have to share their surplus with the remaining divisions. They will rather pursue projects which benefit them more directly (defensive investments481), but which may not be efficient from the firm's perspective. Company headquarters anticipate this behaviour and attempt to equalise investment opportunities across segments. They do so by allocating capital from divisions with high opportunities to divisions with low opportunities. In Rajan/Servaes/Zingales's (2000) model, this behaviour is rational because it leads to a second-best outcome by preventing a thirdbest outcome. Unrelatedness of business segments is thus negative for firm value, and consequently relatedness of business segments is considered a prerequisite of ICM efficiency. On the other hand, Triantis (2004) considers ICMs as incorporating an optionality to allocate funds between different projects. The value of this switching option decreases as the correlation of a firm's segments' cash flows increases: If two projects have perfectly correlated cash flows in all states of the world, the value to 481

See Rajan/Servaes/Zingales(2000), p. 37.

191

switch between them is worthless. In contrast, the option to switch between projects with not or even negatively correlated cash flows is valuable because it allows the company to profit across a multitude of states of the world. The following analysis makes a critical assumption: Business segments whose cash flows are correlated are also related. Intuitively, this assumption makes sense because if the cash flows of two segments develop similarly across multiple years, one reason for this may be that they are subject to similar industrial and competitive forces. If a firm's management ability consists in adequately responding to different competitive forces, then business segments which are subject to similar competitive forces can be considered to be related. If this assumption is true, it follows from Triantis (2004) that unrelatedness of business segments is a prerequisite of ICM efficiency. Rajan/Servaes/Zingales (2000) and Triantis (2004) therefore make opposite predictions on the conditions of ICM efficiency. This study proposes a simple test of these predictions. The basic idea is to analyse the relationship between the cash flow correlation of parent and subsidiary firm (proxying for their relatedness), and the reaction of capital markets to the announcement of an ECO. At its simplest, if the correlation is positive (=relatedness) and capital markets react positively to the ECO announcement, then relatedness is a bad thing (hence, Rajan/Servaes/Zingales (2000) are wrong and by induction Triantis (2004) is right). If the correlation is zero or even negative (-unrelatedness) and capital markets again react positively to the ECO announcement, then unrelatedness is a bad thing (hence, Triantis (2004) is wrong and by induction Rajan/Servaes/Zingales (2000) are right). Specifically, the analysis progresses as follows: First, the correlation between the subsidiary firm's cash flow and the remaining parent firm's cash flow is calculated. Second, the APAR is regressed on the correlation measure. A positive regression coefficient indicates that a positive APAR is caused by a positive correlation between parent and subsidiary firm cash flows. A positive APAR implies that investors are happy because an inefficient ICM has been closed. Inefficient ICMs are therefore characterised by a positive correlation coefficient (i.e., related business segments). This rejects Rajan/Servaes/Zingales' view and by induction supports Triantis' view. On the other hand, a negative regression coefficient indicates that a positive APAR is caused by a negative correlation between parent and subsidiary firm cash flows. Again, a positive APAR implies that investors are happy because an inefficient ICM has been closed. Inefficient ICMs are therefore characterised by negative correlation coefficients (i.e., unrelated business segments). This rejects Triantis' view and by induction supports Rajan/Servaes/Zingales' view. Appendix 49 shows the regression coefficients and p-values of tests for the significance of these regression coefficients. There is no clear trend in results. Some of the coefficients are negative, while others are positive, and none is significant. This indicates that there is no linear relationship, either positive or negative, between the parent/subsidiary firm cash flow correlation and the APAR. However, a closer look at the distribution reveals an interesting picture. Appendix 50 plots the individual firm's correlation coefficients against the respective [0] day window APAR. Two aspects are

192

noteworthy: First, the distribution of correlation coefficients seems left-skewed, also indicated by the mean (median) of the correlation coefficient being 0.3994 (0.6812). This indicates that a positive correlation between parent and subsidiary firm cash flows is more frequent than a negative correlation. The result is intuitive and implies that the majority of parent firms consist of business segments which are related. Second, the image suggests a quadratic relationship between the correlation coefficients and the APAR: While the coefficient resulting from a linear approximation is close to 0 (i.e., the fitting line is almost horizontal), a quadratic approximation results in a negative coefficient for the x 2" Both very positive and very negative correlation coefficients are associated with lower APARs, whereas only slightly positive and only slightly negative correlation coefficients are associated with higher APARs. A two groups difference of means test of the APARs of firms with correlation coefficients between -0.5 and +0.5, and firms with correlation coefficients in excess of +0.5 and below-0.5, confirms this idea: Firms in the former group have an average APAR of 2.44%, whereas firms in the latter group have an average APAR of 0.65%, with the difference being statistically significant at the 10% level (p-value of 0.0935). The result of the analysis thus suggests that both views held in the literature are correct to a certain extent: Rajan/Servaes/Zingales (2000) suggest that relatedness of business segments is a prerequisite for ICM efficiency. The negative reaction to the ECO announcement by firms with related business segments (i.e., business segments whose cash flows are very positively correlated) can be interpreted as investors being unhappy about an efficient ICM being partially closed. This supports Rajan/Servaes/Zingales' (2000) view. Triantis (2004) suggests that unrelatedness of business segments is a prerequisite for ICM efficiency. The positive reaction to the ECO announcement by firms with unrelated business segments (i.e., business segments whose cash flows are not correlated) can be interpreted as investors being happy about an inefficient ICM being partially closed. This rejects Triantis' (2004) view. However, there is an important caveat: There is a negative reaction to the ECO announcement by firms whose business segments' cash flows are negatively correlated. Again, a negative reaction is interpreted as investors being unhappy about an ICM being partially closed, and hence ICMs with negatively correlated cash flows seem to be efficient. This supports Triantis' (2004) view. The underlying question is: When are business segments unrelated? Is it when cash flows are not correlated, or when they are negatively correlated? The analysis reveals clear differences in the results for the two alternative definitions. To the degree that uncorrelated cash flows signal unrelated business segments, Triantis' (2004) view must be rejected. There thus appear to be two distinct scenarios when investors value the existence of ICMs. First, when business segment cash flows are positively correlated, which potentially allows the firm to cross-subsidise related activities; and second, when business segment cash flows are negatively correlated, thus increasing the value of the option to switch between activities depending on the resolution of uncertainty. There seems to be little value in ICMs when business segment cash flows are not uncorrelated, as evidenced by the positive market reaction to ECO announcements by such firms.

193

6.6

Conclusion

The case for efficiency or inefficiency of ICMs is still open. A fairly large number of studies on this subject exists, having produced both theoretical and empirical evidence for and against ICM efficiency. This diversity in results suggests that a generalizing answer to the question of ICM efficiency is unlikely to be found. This study therefore extends current literature by asking two related questions: First, what is investors' judgement on the existence of ICMs? Second, what are the conditions for ICM efficiency? First, investors' reaction to the announcement of an ECO is analysed as a function of various measures of ICM size and efficiency. Investors seem to react more positively to the announcement of an intended ECO when the size of the announcing firm's ICM is large, and when the size of the ICM decreases following the ECO. These findings imply that investors generally consider the existence of an ICM to be value destroying. Investors also react less positively to the announcement of an intended ECO when the announcing firm's ICM has been working efficiently prior to the announcement, and more positively when the efficiency of the firm's ICM increases following the ECO. These findings indicate that investors are at least partially able to discern efficient from inefficient ICM. Second, having established investors' ability to recognise efficient ICMs, the logical follow-up question is: Under what conditions are ICMs efficient? Rajan/Servaes/Zingales (2000) see the relatedness of business segments as a prerequisite of ICM efficiency, whereas Triantis (2004) interprets ICMs in a real option framework, where the value of the option to switch investment between various segments decreases in the correlation of a firm's segments' cash flows. If correlation of segments' cash flows proxies for relatedness of segments, these two views are mutually exclusive. Regressing the APAR on the correlation of parent and subsidiary firm cash flows in the years prior to the ECO, both views are found to be right to a certain extent: Negative reactions occur both when business segments' cash flows are either very positively (supporting Rajan/Servaes/Zingales (2000)) or very negatively (supporting Triantis (2004)) correlated, and positive reactions occur when cash flows are weakly or uncorrelated. Triantis' (2004) prediction is therefore only supported when it refers to negatively correlated cash flows; it is not supported when it refers to uncorrelated cash flows. In conclusion, there appear to be two distinct scenarios when investors value the existence of ICMs: First, when business segments are related, allowing the firm to cross-subsidise related activities; and second, when business segment cash flows are negatively correlated, thereby increasing the value of the option to switch between activities depending on the resolution of uncertainty. There seems to be little value in ICM when neither of these conditions holds, as evidenced by the positive market reaction to ECO announcements by such firms.

194

Determinants of the nature of the second event in European equity carve-outs 7.1

Abstract

Equity carve-outs (ECOs) are temporary structures in almost two thirds of all cases. This chapter pursues two objectives. First, it analyses the frequency of the various types of second events for a sample of European ECOs. Sell-offs occur more often in Europe than in the US, potentially driven by differences in the development states of European vs. US financial markets: European markets may have a lower market capacity and may also produce lower quality information for the parent firm before the ECO, necessitating a partial float of the subsidiary firm even if a sell-off is intended from the outset. Second, this study analyses the determinants of the second event decision. Subsidiary firms reacquired by their parent firms are more likely to have been undervalued by the market relative to their intrinsic value, and subsidiary firms completely sold off are more likely to have been overvalued by the market prior to the second event. Such behaviour is found to be more likely in countries whose financial markets are less developed, and where shareholder rights are consequently assumed to be less pronounced. Firms in industries where synergies between parent and subsidiary firms are more uncertain wait longer until a second event, and are more likely to reacquire compared to firms in more mature industries. No evidence is found that either the parent firm's debt burden or the efficiency of the ICM prior to the initial ECO influence the decision on whether to reacquire or sell off the subsidiary firm. Results are confirmed in two logit models assessing the probability of a sell-off vs. a reacquisition, and the probability of any second event vs. no second event, respectively. Given previous findings about abnormal returns to the announcement of the intended second event, these results may form the basis of a profitable trading strategy.

7.2

Introduction

ECOs are sometimes argued to be temporary structures, which are either completely sold off or reacquired by the parent firm in the years following the ECO. The reasons why companies choose to pursue either of these types of a second event, or alternatively leave the initial stake outstanding with little or no further change, have not been systematically analysed in literature. This is surprising, given the observed tangible value consequences of the second event decision for shareholders of both parent and subsidiary firms. This study analyses the link between various event and firm parameters, and the decision about the nature of a potential second event. In particular, what role does the valuation level of the subsidiary firm, the parent firm's leverage, the previously combined group's internal capital market, parent and subsidiary firm industry association, and the financial home market of the parent firm play in this decision? Intuitively, subsidiary firms undervalued by capital markets should more likely be

195

reacquired, whereas subsidiary firms overvalued should more likely be completely sold off. Parent firms with lower leverage and higher coverage ratios should more likely reacquire their subsidiary firms, whereas parent firms with higher leverage and lower coverage ratios should more likely sell them off. The probability of a buy-back should also increase in the ICM's efficiency of the previously combined entity prior to the ECO, because companies profit from the re-established ICM, as do investors if the ICM is efficient. Same-industry parent/subsidiary firm combinations should more likely experience a reacquisition, whereas cross-industry parent/subsidiary firm combinations should more likely experience a sell-off: According to Kranenburg/Perotti/Rossetto (2004), the decision on the nature of the second event is a function of the resolution of uncertainty regarding the positive synergies between parent and subsidiary firms, and positive synergies are more likely between same-industry combinations than between cross-industry combinations. Finally, parent firms in countries with more developed capital markets should take less time until a second event, because markets are better at producing the information required by the parent firm to take its decision. This last analysis is made possible by the multi-country nature of the sample of European ECOs, in comparison to previous studies using US-only samples. This chapter is organised as follows: Section 7.3 reviews the relevant literature on second events. Section 7.4 analyses the relationship between various event parameters and the nature of the second event. Section 7.5 incorporates the main results of these analyses into a logit-regression based decision model, aiming to evaluate the probability of a parent firm choosing either a reacquisition or a complete sell-off as the second event. Section 7.6 concludes.

7.3

Literature review

A reacquisition and a complete sell-off are not the only two options for the parent firm: First, it may choose to spin off its remaining stake to existing shareholders. Second, it may also have carved out 100% of the subsidiary firm at the initial ECO, making a reacquisition very unlikely. 482 Third, there is the zero-option, i.e., the parent firm decides not to take any action, and to maintain the partial ownership in the subsidiary firm. 483 Fourth, there is a range of other possibilities, including cases where either the parent or the subsidiary firm go bankrupt, and cases where the subsidiary firm is merged with another company, diluting the parent firm's stake, which subsequently may also be

482

483

As will be discussed later, some studies choose to not include this category in their original sample of ECOs. For consistency reasons, this study does include the category. Not including the relevant 17 cases in the sample does not impact the main results of the analysis. A decision to classify a sample firm as belonging to the 'no second event' category is to a certain extent arbitrary in that it is also a function of the time period considered for the analysis: Obviously, leaving more time for a second event to happen c.p. raises the probability of it happening. Existing studies deal with this issue in different ways as discussed in the following paragraphs.

196

sold off. While second event categories are not consistent across studies 484, they can be summarised under one of the six categories discussed: Reacquisition by the parent firm, complete sell-off (either to parent firm shareholders or to a third party), spin-off, 100% sold in initial ECO, no second event, and other. Appendix 51 summarises the results of previous studies along these six categories. On average reacquisitions occur about as often as complete sell-offs (25% vs. 24%). 7% of subsidiary firms are spun off. About one third of companies have not experienced a second event in the time period under consideration. Given that the geographical coverage of all studies is the same (i.e., the US) while the time periods covered differ, a trend across time can be observed: The two studies covering the period from the mid-1960s to the mid-1980s find a higher percentage of subsidiary firms reacquired by their parent firm than do the three studies covering the period from the beginning of the 1980s to the mid-1990s. The percentage of companies sold off and spun off does not seem to deviate systematically across time. Two of the three later studies find a substantially higher percentage of companies with no second event, compared to the two earlier studies. Inter-study comparability of results is limited for two reasons. First, the studies differ in their definition of ECOs, leading two studies 485 to not include cases where the parent firm sells off its entire stake at the time of the initial ECO. In two later studies 486 it is implied that these cases are included in the sample, but it is left unclear whether they are included in the 'Complete sell-off' or in the 'No second event' category. Also, some of the studies have an 'Other' category, which includes various cases of liquidated/bankrupt firms, and firms delisted from CRSP in the years following the ECO. These again could be included in the 'No second event' category of the remaining studies without an explicit 'Other' category, or they may have been excluded from their original sample. Second, the nature of the time periods considered differs across the studies: While Allen/McConnell (1998) and Hulburt (2003) consider time periods relative to the ECO (three and six years, respectively), Klein/Rosenfeld/Beranek (1991) and Vijh (2002) count all events up to a fixed date (five and one year from the end of the sample period, respectively). While the former method allows a better crosscompany comparability of results because each firm has the same time to engage in a second event, the latter method allows for a bigger sample size because recent ECOs do not need to be excluded. Given the limited number of European ECOs, this study therefore chooses the second method. In addition to counting the occurrence of various types of second events, existing studies also analyse the frequency across relative event time (i.e., how many years after the ECO does a given second event occur), as well as the relationship between the nature of the second event and abnormal announcement period returns. 484

485 486

This is because existing studies may have different objectives. While some, like this study, explicitly analyse the nature of the second event (e.g., Klein/Rosenfeld/Beranek (1991), others merely use a certain type of second event as an independent variable (e.g., Allen/McConnell (1998)). Schipper/Smith (1986) and Klein/Rosenfeld/Beranek (1991). See Vijh (2002) and Hulburt (2003).

197

Klein/Rosenfeld/Beranek (1991) find that within three years of the ECO sell-offs are more likely to occur than reacquisitions, whereas sell-offs are less likely than reacquisitions in the following years. Also, parent firms retaining more than 80% of the subsidiary firm following the initial ECO are more likely to reacquire than to sell off, whereas subsidqary firms in which less than 50% is retained are very unlikely to be reacquired. They interpret this as parent firms showcasing their subsidiary firms to the markets as potential acquisition candidates, and reacquiring them when not enough interest is shown. Klein/Rosenfeld/Beranek (1991) also analyse the share price reaction of both parent and subsidiary firms to the announcement of the second event. For parent firms, they find a positive abnormal return for sell-off announcements, and insignificant abnormal returns for reacquisition announcements. 487 Investors thus seem to prefer a complete separation of ties between parent and subsidiary firm to a reintegration of the subsidiary firm. For subsidiary firms, abnormal returns are positive both when the parent firm announces a sell-off and when it announces a reacquisition. The subsidiary firm's positive reaction to an announced sell-off is explained as the result of the buyer firm acquiring all outstanding shares at the price for which the parent firm sells. Since the parent firm reacts positively to a sell-off announcement, the selling price per share must be higher than the market value per share; consequently all shares of the subsidiary firm are traded at a price higher than the current market price, and the reaction to the sell-off announcement is hence positive. 4ss The subsidiary firm's positive reaction to an announced reacquisition is consistent with the general finding in literature of acquisition targets earning positive abnormal returns. 489 Hulburt (2003) finds that carved-out subsidiary firms are more likely to be taken over in the six years following the ECO, relative to a sample of size- and industry-matched benchmark firms. ECO announcements by parent firms whose partially floated subsidiary firm is eventually taken over by another company are accompanied by higher announcement period returns than similar announcements by parent firms whose subsidiary firms are not taken over. Investors seem able to anticipate the second event, and to prefer a complete separation of ties between parent and subsidiary firms. 490 Hulburt (2003) also finds that parent firms retaining between 10-50% of the subsidiary firm in the initial ECO show the highest abnormal announcement period returns. Vijh (2002) finds that ECO announcements followed by a spin-off earn positive announcement period returns whereas ECO announcements followed by no second event earn no abnormal returns, with the difference between these two sub-groups also being significant. Differences between the reacquisition vs. no second event, and complete sell-off vs. no second event subgroups, are not significant. Thus, while their is some discrepancy between Vijh (2002) and Hulburt (2003) as to which method is 487 See Klein/Rosenfeld/Beranek (1991), p. 454. 488 See Klein/Rosenfeld/Beranek (1991), p. 45 5. 489 See Jensen/Ruback (1983) and Jarrell/Brickley/Netter (1988). 490 In contrast, Allen/McConnell (1998) find that APARs in the initial ECO are not higher when the subsidiary firm is subsequently acquired by a third party (see Allen/McConnell (1998), p. 181).

198

preferred (spin-off vs. sell-off), both studies agree in finding that a complete separation is preferable to a reacquisition. Kranenburg/Perotti/Rossetto (2004) suggest an alternative explanation of the source of the value gains in ECOs by viewing an ECO as the creation of a real option allowing the parent firm to either reacquire or sell off its partially floated subsidiary firm. The decision on whether to reacquire or sell off will depend on whether positive synergies from keeping the two businesses together either exceed or fall short of the associated negative synergies arising in the context of conglomerates. This decision is not known at the time of the initial ECO, and is only taken as more information becomes available. Kranenburg/Perotti/Rossetto (2004) also find that an ECO "appears to be an optimal intermediate step even when the likely outcome is a sell-out ''491. They argue that a partial floating prior to a full disposal allows the generation of new information regarding the value of the asset by the market, which in turn improves the management and transparency of the subsidiary firm, thereby increasing the value of the asset to future new investors and maximizing disposal proceeds for the parent firm. Zingales (1995) assumes that in divesting a subsidiary firm, a company must trade off the proceeds from the disposal of cash flow rights (which is maximised when selling to small investors) and proceeds from the disposal of control rights (which is maximised when selling to a large third party). In his model a firm will choose a sell-off if the future owner of the subsidiary firm is likely to decrease the value of the cash flow rights, whereas it will choose a spin-off (assuming private benefits are zero) or an ECO if the future owner will likely increase the value of the cash flow rights. An ECO can thus be a means of maximizing combined proceeds from the sale of both types of rights, if the initial ECO is followed by the sale to a third party. In the case of a share price decline, a firm may also decide to reacquire its partially floated subsidiary firm because it is more likely to be able to increase the value of the cash flow rights associated with the subsidiary firm's ownership, relative to the (undervalued) market price of the subsidiary firm. To summarise, in two out of three cases ECOs are temporary structures. Following the initial ECO, sell-offs tend to occur sooner whereas reacquisitions tend to occur later. In some cases markets are able to anticipate the nature of a potential second event, and a complete separation of the subsidiary firm (either through a sell-off or a spin-off) is preferred. In theory, a parent firm can maximise the value creation from the planned disposal of a subsidiary firm through the two-step procedure of an ECO, which creates valuable real options for the firm.

491

Kranenburg/Perotti/Rossetto (2004), p. 29.

199

7.4

Data and univariate analyses

7.4.1

Data sources

Second events are identified based on the parent firm's equity stake in the subsidiary firm as indicated in the annual reports of either parent or subsidiary firm published in the years following the initial ECO. The obvious question arises which percentage thresholds should be used to categorise the cases in the sample as a particular second event. While choosing a particular level will always contain an arbitrary element, there are at least two sensible limiting values: First, a 100% threshold for a reacquisition and a 0% threshold for a sell-off: These are the extreme values corresponding to the intuitive notion of reacquisition and sell-off. However, requiring a 100% stake in the subsidiary firm to qualify as a reacquisition and a 0% stake to qualify as a sell-off is unlikely to capture all economically relevant cases: A company may own less than 100% and still economically control the subsidiary firm. Similarly, ownership levels following an announced sell-off may be larger than 0% if a limited number of shares in the subsidiary firm continue to be held as a (readily sellable) financial investment. To capture these cases, an alternative threshold definition is applied: 75% for a reacquisition and 25% for a sell-off. These values are likely to lie within the proximity of relevant threshold values in many legal systems. As an example, in Germany any major reorganization (merger, demerger, transfer of assets, or change of legal form) requires that 75% of voting capital supports the decision. 492 To ensure robustness of results, both alternatives are implemented. For the sake of brevity and readability, all tables and figures presented will be for the 100/0 case. Significance levels for results will be reported for both the 100/0 and for the 75/25 alternative. When the ownership falls to 0% 493, the annual report is searched for comments on the reasons for this reduction. In most cases, a complete sell-off to either a third party or to the public is listed as the reason. In four cases the subsidiary firm is spun off to existing shareholders; in three cases, the subsidiary firm has gone into bankruptcy, and in two cases, the subsidiary firm was merged with another firm, diluting the parent firm's stake. Both the bankruptcy and the merger cases have been classified in the 'Other' category. In all remaining cases the parent firm has continued to hold between 0% and 100% until the end of the period under consideration (fiscal year end 2004 of each respective sample firm): These cases are classified as 'No second event'. The data required for the analysis stems from various sources. Parent and subsidiary firm accounting and market value of equity data is from Datastream, with local

492 493

See Lutz (2005), p. 93. There are four cases where there is ownership between 0% and 10% after the second event: Adcapital/BE Semiconductor, where the parent firm has been holding approx. 5% since the sell-off in 2000; Adcapital/Schaltbau, where the parent firm has held approx. 8% following a reduction from 55% in 2002, with this remaining stake completely sold in 2005; Vivendi/Veolia, where the parent has been holding approx. 5% since the sell-off in 2003; and DNO/Petrolia Drilling, where the parent has been holding approx. 1% since the sell-off in 2003. The main results in this analysis are not sensitive towards including these four cases into the sell-off category.

200

currencies converted to Euro at fiscal year end exchange rates. Subsidiary firm segment level data, where required494, is handpicked from the respective subsidiary firms' annual reports, also converted to Euro at fiscal year end exchange rates where required. Subsidiary firms' SIC codes are taken from Datastream where available, and hand-filled based on personal judgement for the remaining cases. Finally, single-segment firms are identified based on Datastream, where it is assumed that if only one SIC code is listed, the company is a single-segment company in that particular SIC industry.

7. 4.2

Impact of individual parameters on the second event

7.4.2.1 Frequency of second events Assuming that ICMs are substitutes for external capital markets (ECMs), the importance of ICMs in less developed financial markets should be higher than in more developed financial markets. Differences in the development state of financial markets can be assessed either across countries (at a specific point in time), or across time (within a specific country). All existing studies analysing the second event use a US sample, and the time period for the three most recent studies roughly corresponds to this study's time period. Therefore discrepancies in frequencies can be analysed as a function of the different development states of US vs. European financial markets. US markets are characterised by a higher ratio of stock market capitalization held by minorities to GNP, a higher ratio of the number of listed domestic firms per citizen, a higher ratio of IPOs per citizen, and more developed shareholder rights, compared to many European countries. 495 Consequently US financial markets are considered to be more developed compared to European financial markets. Since the reacquisition of a partially floated subsidiary firm re-establishes the ICM which was partially closed in the initial ECO, reacquisitions should be more frequent in Europe compared to the US. Selloffs, which completely close the ICM, should be less frequent in Europe compared to the US. Looking at this study's sample, reacquisitions occur in 6% of the cases, about as frequently as in the most comprehensive US study (Vijh (2002)), and slightly less frequently than the average of the last three major US studies (15%). A complete selloff occurs in 38% of the cases, which is considerably higher than the 25% average from the last three US studies. This result thus contradicts the hypothesis. There are three potential explanations why sell-offs are a more frequent second event in Europe than they are in the US. First, the difference between Europe and US is less pronounced when the spin-off category is also considered: Spin-offs seem to occur

494

495

Subsidiary segment level data is required to calculate the imputed aggregatevalue for non-financial multi-segmentsubsidiaries (16 cases). See LaPorta/Lopez-de-Silanes/Shleifer/Vishny(1997), p. 1138 for a detailed comparison of these variables across various countries.

201

more often in the US than in Europe (9% v s . 2O//o).496 Since spin-offs also represent a way of completely separating the ties between parent and subsidiary firm, these percentages can be added to the above numbers, resulting in 34% (US) vs. 40% (Europe) of subsidiary firms being completely separated from the parent firm in the second event. Second, parent firms intending a complete disposal of the subsidiary firm may choose the intermediate step of an ECO: As posited by Kranenburg/Perotti/Rossetto (2004), an ECO may be value-maximizing even if the original intention is a complete sell-off. By partially floating the subsidiary firm, the parent firm allows the market to generate new information regarding the value of the asset. The information flow from the market to the parent firm is accompanied by an information flow from the subsidiary firm to the market (e.g., in the form of the IPO prospectus and annually published accounts). This overall increase in transparency is likely to facilitate the eventual sell-off. While this argument holds both for the US and for Europe, it may be more relevant for Europe assuming that less developed financial markets produce lower quality information before the ECO, and therefore necessitate additional means to produce information. For example, assuming that conglomerates are valued by capital markets via a sum-of-theparts analysis 497 of their business units, such a valuation may be easier in the US where more pure-play companies are available which are required to derive corresponding multipliers for each business segment. The value of a US conglomerate may thus be assessed appropriately by external investors even before a potential ECO. In contrast, in less developed markets, there may not be enough pure-play companies, so a market valuation of a conglomerate's business unit for which pure-play comparable companies are not available is more difficult before the ECO: Only once the subsidiary firm is carved out can a fair price be determined. Third, less developed ECMs in Europe could potentially prevent either an outright sale or an ECO of 100% of the subsidiary firm, thus requiring a staggered sale/ECO even when a complete sell-off is the original intention. Market practitioners regularly assess potential investor demand for a capital issue ('market capacity'), which may fall short of 100% of the subsidiary firm's value. Conversely, in more developed financial markets, companies may realise a complete sell-off in a single step, the results of which are either included in the 'No second event' category of the US papers, or not at all (in the case of a sale to a third party). The percentage of firms with no second event is slightly higher in the US studies (41% vs. 44%), consistent with this explanation. Applying the 75/25 threshold levels increases the frequency of reacquisitions (from 6% to 12%), sell-offs (from 38% to 40%) and complete initial sell-offs (from 10% to 17%), and consequently decreases the number of firms in the 'No second event' category (from 41% to 25%). Interpretations are similar to the 100/0 threshold level case. 496

497

Potentially, an explanation for this is that in the US spin-offs can be structured to be tax-free for shareholders (see Gaughan (2002), p. 408-409). The tax treatment in European countries may differ. A detailed analysis of this issue is beyond the scope of this study. In addition to DCF models, this represents the favourite valuation approach of market practitioners.

202

Overall, the differences in the frequencies of various second events between Europe and the US are likely linked to differences between less developed European markets, and better-developed US markets. These differences seem to affect the decision of parent firms regarding the second event in an ECO. However, European countries also differ among themselves in the development state of their respective financial markets. This issue is addressed in detail in section 7.4.2.8.

7. 4.2.2

Time until second event

Table 24 shows the average number of years passed until the second event for the different types of second events.

Re-acquisition Completely sold Spin-off Other All firms

Average years

N u m b e r of firms

4.5 3.4 2.2 4.2 3.6

11 61 4 6 82

p-value of ANOVA including all firms with any 2nd event

0.3885

p-value of ANOVA incl. firms with either re-acquisition or sell-of

0.2129

Table 24: Average time until second event

Time is measured beginning from the date of the initial ECO up to the last year before the second event occurs, which is assumed to happen in the middle of the following fiscal year. Reacquisitions tend to occur on average 4.5 years after the initial ECO, while complete sell-offs tend to occur 3.4 years after the initial ECO. This result is consistent with Klein/Rosenfeld/Beranek' s (1991) finding of sell-offs being more likely in the first years after the initial ECO, and reacquisitions being more frequent in latter years. They interpret their findings as support for the idea of parent firms showcasing their subsidiary firm to the markets as potential acquisition candidates, and reacquiring them when not enough interest is shown. However, an ANOVA testing whether the second event category has explanatory power for the time passed until the specific second event produces insignificant results (p-value of 0.3885 when all second event categories (i.e., including spin-offs and the 'other' category) are considered, and 0.2129 when only the reacquisition and the sell-off category are considered as independent variables). Therefore the null hypothesis of no difference in average time passed until the respective second event cannot be rejected. However, non-significance of results may be driven by the small size of the reacquisition group (n=10). For the 75/25 altemative, the average time until a reacquisition is also longer than the average time until a sell-off (4.2 vs. 3.7 years), but again the averages are not significantly different from one another.

203

7.4.2.3

Stake retained in initial ECO

Table 25 shows the average stake retained by the parent firm in the initial ECO for the different types of second events. When the subsidiary firm is subsequently reacquired by the parent firm, the average stake retained is 68%, whereas it is only 54% when the subsidiary firm is subsequently sold. This difference is statistically significant at the 5% level (p=0.0202). Again this result is consistent with similar findings for the US by Klein/Rosenfeld/Beranek (1991), who find that the probability of a reacquisition increases with the stake retained by the parent firm. This result is intuitive, and implies that parent firms may have a tendency towards either an eventual reacquisition, in which case they sell a lower stake in the initial ECO, or an eventual sell-off, in which case they sell a larger stake in the initial ECO. For spin-offs the average stake retained is 42%, supporting the above notion of a lower stake when the final outcome is a complete separation of parent and subsidiary firm. For the 75/25 alternative, the average stake retained in case of an eventual reacquisition is also larger than the average stake retained when the second event is a sell-off (67% vs. 56% years), and again the averages are significantly different from one another (p=0.0055).

Remaining stake

Re-acquisition Completely sold Spin-off Other All firms

Number of firms

68% 54% 42% 54% 55%

10 67 4 6 87

p-value of ANOVA including all firms with any 2nd event

0.0714

p-value of ANOVA including firms with 2nd event either re-acquisition or sell-off

0.0202

Table 25: Stake retained after initial ECO and second event

7.4.2.4 Over~undervaluation of subsidiary firm In Zingales' (1995) model parent firms reacquire their partially floated subsidiary firms when share prices decline, hoping to increase the value of the cash flow rights associated with ownership in the subsidiary firm, relative to its market price. Market practitioners commonly refer to this notion as undervaluation, and the reacquisition of an undervalued subsidiary firm is an intuitively appealing idea. 498 Similarly, it can be hypothesised that a sell-off will be more likely when the subsidiary firm is overvalued by the market: The parent firm believes that the value of the cash flow rights is less than what the market is willing to pay, and therefore decides to sell its remaining stake.

498

Supporters of the efficient market hypothesis will probably argue that the very notion of a company being undervalued violates basic assumptions about market efficiency, and arbitrage should immediately eliminate any undervaluation.

204

Over/undervaluation of a subsidiary firm is assessed via the multiplier approach as suggested by Berger/Ofek (1995). A subsidiary firm's excess value is defined as499

EXVAL i = In

V/

i --~mf

i=1

where V/is the aggregate market value (defined as the market value of equity plus the book value of debt) of company i at the end of the last fiscal year prior to the year in which the second event occurs; AIi is the value of segment i's accounting item (sales, assets or EBIT) in the year prior to the second event; and Ind i - V ~ / ml is the multiple of aggregate market value to an accounting item (sales, assets or EBIT) for the median single-segment firm in segment i's industry (defined on the two digits SIC code level) in the respective year. When a segment's EBIT is negative, the segment's value is set to 0. An alternative (but related) way of assessing excess value is proposed by Lang/Stulz (1994). They define excess value for firm i as

n

EXVALi = M~i - Z q~ * BAj BAi j=l BAi where MV/is the market value of firm i (defined as market value of equity plus book value of debt), BAg is the book value of total assets of firm i, BAj is the book value of total assets of firm i's segment j, and qj is the median Tobin's Q of all single-segment firms in the same industry as business segment j. The approach is thus similar to Berger/Ofek (1995) in that the intrinsic value of the conglomerate is assessed as the sum-of-the-parts value of its individual segments. Both methods are used to ensure robustness of results.

499

See Berger/Ofek (1995), p. 60. This method is both popular with market practitioners ('sum-of-theparts valuation') and has found widespread use in the literature on the conglomerate discount (see Graham/Lemmon/Wolf (2002), Colak/Whited (2003), Burch/Nanda/Narayanan (2003), Doukas/Kan (2004)). To the author's knowledge there is no alternative methodology for measuring excess value which has found widespread acceptance. Individual exceptions include, e.g., Schoar (2002), who constructs a measure of excess overhead costs based on labour, capital and material input costs as derived from the Longitudinal Research Database.

205

Table 26 shows the average over/undervaluation of subsidiary firms for the different types of second events, as well as the number of firms in each category, which is lower compared to the total sample because firms with missing accounting or share price data are excluded. Total Number assets of firms multiplier Re-acquisition Completely sold Spin-off Other All firms p-value of ANOVA (all firms) p-value of ANOVA (only reacq & sell-off firms)

-51% 16% -44% 55% 8% 0.1209 0.0624

7 44 3 4 58

Sales Number multi- of firms plier

EBIT Number multi- of firms plier

Lang/ Number Stulz of firms

-73% 27% -72% 111% 16%

-19% 44% -22% 134% 38%

-116% -6% -230% 108% -25%

7 47 3 4 61

0.0638 0.0541

0.1564 0.1532

5 36 2 2 45

7 40 3 4 54

0.0209 0.0842

Table 26: Over/undervaluation of subsidiary firm and second event

The main result of the analysis confirms the hypothesis: Based on all three accounting items, subsidiary firms reacquired by their respective parent firms have been undervalued, and subsidiary firms completely sold off have been overvalued by the market prior to the second event. An analysis of variance shows that the categorization of the second event has statistically significant explanatory power for the level of over/undervaluation when the imputed value is based on the asset and for the sales multiplier (p-values of 0.0624 and 0.0541, respectively, when only reacquired and completely sold off firms are considered). For the EBIT multiplier, the p-value is marginally not significant when only reacquired and completely sold off firms are considered (0.1532). The levels of undervaluation are similar for the asset and the sales multipliers, and substantially lower for the EBIT multiplier. This is a result of the construction of the latter, where the value of a firm's segment with a negative EBIT is set to 0, thus causing an upward bias in value. 5~176 The levels of overvaluation for the three multipliers are a mirror image of these findings. According to the alternative Lang/Stulz (1994) methodology, reacquired subsidiary firms are considerably undervalued, while sold-off subsidiary firms are only marginally undervalued. While this result is different in that it indicates that sold-off subsidiary firms are undervalued (rather than overvalued as indicated by the Berger/Ofek (1995) methodology), the difference between the two levels of undervaluation is significant (p=0.0842), indicating that reacquired subsidiary firms are significantly more undervalued than sold-off subsidiary firms. Therefore since over- and undervaluation are relative concepts, the statement that reacquired subsidiary firms are undervalued and sold-off subsidiary firms are overvalued relative to one another still pertains. Overall, parent firms seem to take advantage of low valuations of their partially floated subsidiary firm by reacquiring them, and also of high valuations by selling their subsidiary firms. Results are robust to the 75/25 alternative, with group differences significant for all three multipliers 5~176 alternative would be to exclude all firms with at least one EBIT-negative segment, decreasing the sample size, and not alleviating the upward bias. Unreported results show that this alternative produces values very close to the actual implementationmethod.

206

(p=0.0549, p=0.0542, p=0.0847 and p=0.0758 for the assets, sales and EBIT multipliers, and for the Lang/Stulz multiplier, respectively).

7.4.2.5 Leverage and coverage of parent firm The decision of whether to sell off or reacquire could be associated with the leverage position of the parent firm. Companies with a higher debt burden could decide to sell off the subsidiary firm, while companies with a lower debt burden could decide to reacquire it. Two measures of a company's debt burden are used in the analysis: a leverage ratio, defined as the book value of debt 5~ from the last year preceding the second event, divided by the market value of equity based on the last trading day in the year preceding the second event; and a coverage ratio, defined as EBIT divided by net interest payable, both from the last year preceding the second event. The hypothesis is that parent firms reacquiring their subsidiary firms have lower leverage and higher coverage ratios compared to parent firms selling off their subsidiary firms. Table 27 shows the average leverage and coverage ratios of subsidiary firms for the different types of second events. The main result contradicts the hypothesis: Reacquiring parent firms on average seem to have higher leverage and lower coverage ratios than parent firms selling off their subsidiary firms. Group means are significantly different when all second event categories are considered (p=0.0663 and p=0.0439 for leverage and coverage ratios, respectively). There are two potential explanations for this result: First, the small size of the reacquisition group (n=7) 5~ may lead to spurious results. 5~ Second, reacquiring firms may have increased their leverage prior to the year in which the reacquisition occurred (e.g., through the issuance of debt) as a means of financing the planned reacquisition. Applying the 75/25 alternative, significance for the group differences in the leverage ratio disappears, and is only found for the coverage ratio when considering all second event categories. The same caveats regarding group size apply as in the discussion of the leverage ratio in the 0/100 alternative. Evidence on the impact of the leverage and coverage ratio on the second event decision is thus inconclusive.

501

502

503

It would have been preferable to use the market value of debt; however such data is not readily available in most commercial databases. Assuming book value of debt to equal its market value is common among market practitioners. Of the eleven reacquisition cases, three are excluded because the parent firm is a financial company, for which leverage and coverage ratios have a different interpretation relative to non-financial companies. One company is excluded because of insufficient data. A separate analysis of the individual firm values reveals that the leverage of France Telecom before reacquiring its Orange subsidiary in 2003 was 3.56, driving the average result upwards.

207

Parent Number of leverage firms

Re-acquisition Completely sold Spin-off Other All firms p-value of ANOVA (all firms) p-value of ANOVA (only reacq & sell-off firms)

1.71 0.87 0.32 0.75 0.92

7 48 4 3 62

0.0663 0.0291

Parent Number of coverage firms

3.48 6.39 10.45 -0.06 6.01

7 48 4 3 62

0.0439 0.1758

Table 27: Leverage, coverage and second event

7. 4.2.6 I C M size and efficiency The decision of whether to sell off or reacquire the subsidiary firm could be linked to the efficiency of the ICM before the initial ECO. In section 6.5.1.2 it was shown that the abnormal announcement period return for parent firms is negatively related to the efficiency of their ICMs: Parent firms with efficient ICMs are not rewarded by the markets for partially closing them, while parent firms with inefficient ICMs profit from an ECO as evidenced in positive APAR. Consequently parent firms with more efficient ICMs could be more inclined to reacquire their subsidiary firms to re-establish their partially closed ICMs, while parent firms with less efficient ICMs could be more inclined to completely sell off their subsidiary firms. Also, it was shown that parent firms with larger ICMs earn higher abnormal returns when announcing the ECO. Hence, parent firms with larger ICMs may be more likely to sell off, whereas parent firms with smaller ICMs may be more inclined to reacquire their subsidiary firms. Appendix 52 shows various size and efficiency measures of ICM for both reacquiring and selling-off parent firms. The variables are based on accounting and market data from the last year prior to the initial ECO. 5~ The difference between reacquiring and selling-off parent firms is positive (i.e., as expected and in line with the hypothesis) for six out of the eight efficiency measures, but not significantly so. The lack of statistical significance for the remaining efficiency measures may be partially due to the small sample size (n=l 6 up to n=20), caused by limited data availability in the construction of the ICM measures. Still, the null hypothesis of no explanatory power of the level of preECO ICM efficiency on the second event decision cannot be rejected. The differences of the size measures for reacquiring and selling-off parent firms show no distinct pattern, with four differences being positive and four measures being negative. Negative differences are consistent with the hypothesis of parent firms with larger ICMs selling their subsidiary firms, rather than re-establishing ICMs through a reacquisition. Positive differences (i.e., larger ICMs for reacquiring companies) may indicate that the benefits for the company from having a larger ICM outweigh the 5O4

See chapter 6 on ICMs for a more detailed discussion of these variables.

208

disadvantages for shareholders resulting from this larger ICM. However, none of the results is statistically significant. Therefore the null hypothesis of no explanatory power of the size of pre-ECO ICMs on the second event decision cannot be rejected. Applying the 75/25 alternative results in the same pattern of signs and non-significance for both the efficiency and the size measures.

7.4.2.7 Industry Two aspects are analysed: First, do same-industry ECOs tend to differ from crossindustry ECOs in their second event? Second, are certain second events more likely in certain industries? While these two questions may address the general impact of industry association on the second event decision, there is likely to be a series of casespecific aspects determining the nature of the second event. For example, parent and subsidiary firms may have strategic relationships favouring either a reacquisition or a sell-off irrespective of their industry association. A more detailed analysis on a case-bycase basis may be required to capture these effects, which is beyond the scope of this study. The analysis will therefore employ industry codes to assess the questions raised above. In Kranenburg/Perotti/Rossetto's (2004) theoretical model of ECO as real options the decision of whether to reacquire or sell off a partially floated subsidiary firm rests on the resolution of uncertainties regarding the synergies between parent and subsidiary firm. In section 3.5.3 it was found that cross-industry ECOs earn higher APAR than same-industry ECOs, and the explanation offered was that cross-industry ECOs have a higher potential of removing negative synergies between parent and subsidiary firms. Therefore it seems plausible to assume that positive synergies between parent and subsidiary firms are more likely to arise when both firms are in the same industry. In turn, this implies that parent firms with same-industry partially floated subsidiary firms should be more likely to reacquire, whereas parent firms with cross-industry subsidiary firms should be more likely to sell off their subsidiary firms. Appendix 53 shows the number of actual and expected firms across the same/crossindustry dimension and the various types of second events. Expected firms in each l"lj, b

category are calculated

as

Pa,b =

i=1

6

2

j=l

, where i indexes the nature of the

ZZni,j i=1 j=l

second event and j equals 1 for cross-industry ECOs, and 2 for same-industry ECOs. The main result lends support to the hypothesis stated above: Reacquisitions occur significantly more often in the case of same-industry ECOs, whereas sell-offs occur significantly more often in the case of cross-industry ECOs. A chi-squared test of independence produces a p-value of 0.0816, indicating that the result is significant at least at the 10% level.

209

Industry affiliation may also play a role in the decision which second event will occur. In line with standard results in real option theory, Kranenburg/Perotti/Rossetto (2004) predict that the higher the level of uncertainty, the later it is optimal to exercise either the buy-back or the sell-off option. 5~ They also predict that the second event decision will be influenced by the parent firm's industry affiliation: According to their model, "in industries where synergies between different units are more uncertain, presumably because of rapid change in technology or regulation...buy-backs should become more common relative to sell-offs ''5~ To test these predictions, Appendix 54 shows the number of firms experiencing a given second event classified by the first digit of the parent firm's SIC code. 5~ Two interesting observations emerge: First, sell-offs occur more often than expected for parent firms with the one digit SIC code 1, 3 and 5, and less often than expected for parent firms with the one digit SIC code 6 and 7. Second, the frequencies in the 'No second event' category are a mirror image of the first finding: Parent firms with a one digit SIC code 1, 3 and 5 have more second events, while parent firms with a one digit SIC code 6 and 7 have fewer second events than expected. Looking at the industry descriptions underlying the SIC code scheme, firms with SIC code 1 are mining and construction companies, SIC code 3 classifies manufacturing companies, and SIC code 5 represents wholesale trade companies. On the other hand, SIC codes 6 are used for finance and insurance companies; and SIC codes are used for 7 for service companies. To the degree that the mining, construction, manufacturing and wholesale trade industries are more mature and characterised by less uncertainty, whereas the finance and services industries are less mature and characterised by a more rapid change in technology and thus a higher uncertainty, Kranenburg/Perotti/Rossetto's (2004) first prediction can be confirmed: Firms in less mature and more rapidly changing industries where synergies between parent and subsidiary firms are less certain seem to wait longer until a second event, which shows in a higher than expected number of firms in the 'No second event' category. The second prediction finds indirect support in that firms in more mature and less uncertain industries are more likely to sell off their subsidiary firms, whereas firms in more rapidly changing and more uncertain industries are less likely to sell off. While for SIC code 6 the actual number of reacquiring parent firms exceeds the expected number (as predicted by Kranenburg/Perotti/Rossetto (2004)), it is lower for SIC code 7. However, interpretability of this result is hampered by the low number of firms experiencing a reacquisition. Applying the 75/25 threshold levels leads to very similar results. Overall, when synergies between parent and subsidiary firms are less certain, the second event is more likely to occur later rather than earlier, and it is more likely to be a buyback, consistent with the prediction of Kranenburg/Perotti/Rossetto's (2004) real option model of ECO. 505 See Kranenburg/Perotti/Rossetto (2004), p. 24. 506 Kranenburg/Perotti/Rossetto (2004), p. 25. 507 The standard is for industry to be defined on the two digits SIC code level. However, this produces a frequency distribution with many empty or lowly populated cells. Hence for the purpose of this specific analysis industry is defined on the one digit SIC code level.

210

7.4.2.8 Region Kranenburg/Perotti/Rossetto (2004) postulate that the transparency of the financial market in which the parent firm is listed influences the stake retained after the initial ECO: Parent firms in more transparent markets will retain a higher stake. This is because floating a smaller stake implies a lower underpricing, but produces the same information content for the parent firm and for potential future investors as floating a larger stake in a less transparent capital market does. Kranenburg/Perotti/Rossetto (2004) lament that "there are no studies that compare carve-outs among different countries to confirm this implication ''5~ The present study fills this void. Financial market transparency is proxied by a country index for financial development. Prior literature laments the lack of data on the quality of intemational financial markets, and uses measures of quantity, i.e., size of the respective financial markets, as a proxy for quality. 5~ Analogously, the present study assumes that a relatively larger capital market "reflects...a more competitive financial market with better institutions ''51~ The index used has been developed by Wurgler (2000). It is a composite of two measures of the relative size of a country's equity and credit markets (stock market capitalization to GDP and credit claims to GDP). The actual index value is available for 12 out of the 13 countries in this study, and is based on average values of the two measures as of 1980, 1985 and 1990. TM The hypothesis is that a higher level of financial development as proxied by the index leads to a higher stake retained. OLS-regressing the stake retained on the index value, the regression coefficient tums out to be negative, and significantly so at the 1% level (p=0.0019). This rejects the original hypothesis and lends support to the idea of a negative relationship between financial market development and stake retained: Parent firms in less transparent markets seem to retain a higher stake in the subsidiary firm. This unexpected result merits further consideration. Earlier it was shown that a higher stake in the subsidiary firm makes a reacquisition more likely. Consistently, an additional analysis shows that reacquisitions on average occur in countries with a lower financial development index compared to sell-offs (average index of 1.0282 vs. 1.2125, difference significant with a p-value of 0.0033). Applying the 75/25 threshold levels leads to very similar results (1.0486 vs. 1.1999, p-value of 0.0023). Also, earlier it was shown that reacquisitions are generally associated with an undervaluation of the subsidiary firm in the market. Together, this offers a possible explanation for the negative relationship between financial market development and stake retained: Parent firms in less developed financial markets float their subsidiary firms and reacquire the outstanding minority stake at a discount to market value. Since it is well-known that the position of minority shareholders is weaker in countries with 508 Kranenburg/Perotti/Rossetto (2004), p. 27. 509 See Rajan/Zingales(1998). 510 Wurgler (2000), p. 197. 511 See Wurgler (2000), p. 10-12.

211

less developed financial markets (see LaPorta/Lopez-de-Silanes/Shleifer/Vishny (1998), who find that small shareholders have less rights in less developed financial markets), it may be easier for parent firms to get away with this opportunistic behaviour, compared to their counterparts in more developed financial markets. As an anecdotal case in point, the reacquisition of Wanadoo by France Telecom and the planned reacquisition of T-Online by Deutsche Telekom offer an interesting comparison. France Telecom announced its intention to reacquire the outstanding 29% of Wanadoo in February 2004. In July 2004, Wanadoo was again wholly owned by France Telecom, and by September 2004 it had been completely merged with the France Telecom group. 512 Deutsche Telekom, on the other hand, announced its intention to reacquire the outstanding 26% of T-Online in October 2004. Various shareholder groups took Deutsche Telekom to court, accusing the group of stealing value from shareholders. In November 2005 a German court ruled that Deutsche Telekom could not go ahead with the planned merger of T-Online into Deutsche Telekom for the moment, and arriving at a final decision could take several years. 513 According to Wurgler (2000), the financial development index of France is 1.06, whereas it is 1.22 for Germany. According to LaPorta/Lopez-de-Silanes/Shleifer/Vishny (1998), French civil law countries have the weakest investor protections and the least developed capital markets. France Telecom, in a country with less developed financial markets and less minority shareholder protection, has been able to reacquire its subsidiary firm relatively quickly, and potentially to the disadvantage of subsidiary firm shareholders. Deutsche Telekom, on the other hand, in a country with relatively more developed financial markets and better minority shareholder protection, is struggling to achieve a similar result. No 2nd event

Any 2nd event

Average financial development Number of firms

1.1165 71

1.1973 85

p-value of difference of group means

0.0225

Table 28: Country financial development and second event

The analysis of industry association and second event showed that a higher uncertainty in the industry is likely to lead to a lower number of second events, either reacquisitions, sell-offs, spin-offs, or other. A similar argument may apply for uncertainty caused by a parent firm's regional association: Parent firms in countries with a higher uncertainty will take longer until a potential second event. A country's uncertainty is again (inversely) proxied by its financial development index. Table 28 shows the results of a two groups difference of means test, where group 1 consists of all firms with no second event, and group 2 consists of all firms with any second event. 512 513

See France Telecom 2004 annual report. See article in Financial Times Deutschland, 4 November 2005.

212

The average financial development index for group 1 is 1.1165, whereas for group 2 the average is 1.1973. The difference is significant at the 5% level (p=0.0225). Applying the 75/25 threshold levels leads to similar results (1.1025 vs. 1.1766, p-value of 0.0648). The null-hypothesis of no difference is rejected, and the alternative hypothesis is supported: Parent firms experiencing any second event are more likely to come from countries with more developed financial markets. This is consistent with the notion that parent firms in less developed financial markets need more time to gather information about the potential resolution of uncertainty regarding the synergies between parent and subsidiary firms, because less developed financial markets are less efficient at providing the required information.

7.5

Logit regression analyses

Klein/Rosenfeld/Beranek (1991) find that for the second event announcement dates, parent firms announcing a sell-off earn positive abnormal returns, whereas reacquisition announcements lead to insignificant abnormal returns. Consequently, it is desirable for investors to assess the probability of the second event being a sell-off, as opposed to a reacquisition, which could create a profitable trading strategy. Klein/Rosenfeld/Beranek (1991) also find that abnormal returns for subsidiary firms are positive both when the parent firm announces a sell-off and when it announces a reacquisition of the subsidiary firm. Therefore, it would also be desirable to assess the probability of any second event occurring, as opposed to no second event. Two logit models to assess these questions are developed next.

7.5.1

Logit Model 1: Sell off vs. reacqubsition

Independent variables included in the model are the over/undervaluation of the subsidiary firm (VAL), a dummy variable indicating whether the ECO is classified as same-industry or cross-industry (SAME), the financial development index of the country of the ECO parent firm (FD), and a dummy variable indicating whether the subsidiary firm is an industry characterised by high uncertainty (first SIC code digit 4, 6 and 7514). The stake retained by the parent firm is not included due to its correlation with the financial development index. The measure of ICM efficiency (ICM) is not included because it dramatically reduces sample size (to n=16) due to the limited availability of segment level data required for its construction. 515 The dependent

514

515

SIC codes 4 are used for transport and telecommunication companies, SIC codes 6 for finance and insurance companies, and SIC codes 7 for service companies. Unreported results of a model including the ICM measure show a high R2, but the likelihood ratio is low, making it more unlikely that no coefficient is different from 0 (i.e., supporting the nullhypothesis). This is a direct consequence of the lower number of available sample firms when including the ICM measure.

213

variable equals 0 if the second event is a reacquisition, and 1 if the second event is a sell-off. The model therefore is 516 z =/3o + fll * VAL +/32 * IND +/33 * F D , where z (the logit) is a linear combination of the various independent variables, and the fl's are regression coefficients. The probability of a sell-off, p(sell-ofJ), is then calculated as Cz

p(sell - off )

=

-

-

l+e z

i.e., via the logistic regression function as a function of the logit z. Appendix 55 shows the main results of the logit regression. All four coefficients have the expected sign: fl(VAL) is positive, indicating that a more positive valuation relative to the imputed value of the subsidiary firm makes it more likely for the parent firm to sell off; fl(SAME) is negative, indicating that same-industry subsidiary firms are less likely to be sold off; fl(FD) is positive, indicating that a subsidiary firm is more likely to be sold off when the parent firm is listed in a financially better developed market; and fl(IND) is negative, indicating that a subsidiary firm is less likely to be sold off when it is active in a high uncertainty industry. The quantitative interpretability of the coefficients is limited due to the logistic regression used to compute p(sell-ofJ): In contrast to linear regressions, there exists no linear relationship between the independent variables and the dependent variable. Normally some interpretation is possible via the odds-ratio, where odds are defined as o d d s ( y = l)=

p ( y = l) =eZ 1 - p ( y = 1)

This allows assessing the change in the odds-ratio, given a change in the independent variables by 1 unit. Since VAL is measured using the natural logarithm of the aggregate market value divided by the aggregate imputed value, a one unit change corresponds to an increase of e 1 ~ 2.7183 = 271.8% in the market value relative to the imputed value. Increasing market value by this magnitude increases the odds-ratio for a sell-off by fl(VAL)=3.80. Similarly, increasing FD by one unit increases the odds-ratio for a selloff by approx. 1,840. However, since obviously an increase in the two measures by one unit is not comparable, interpretability of results again is limited. For the dummy variable IND, any attempt of an interpretation of the change in the odds-ratio seems non-sensical. To assess the overall quality of the model, three different types of measures are analysed:

516

See Backhaus/Erichson/Plinke/Weiber,p. 423.

214

Classification results: The quality of the classification is assessed by analysing whether the observations are placed into the correct groups by the logit model: Observations with p>0.5 are attributed to the sell-off group while all remaining observations are attributed to the reacquisition group. The percentage of correctly classified observations is 85.7%, substantially higher than the 50%-level which would be expected if observations were attributed to the two possible outcomes at random. To formally assess statistical significance, Press's Q-test is applied5~7: Press' s Q - t e s t = [K - ( K * G* a)] 2 g*(G-1) ' where K is the sample size, G is the number of groups (G=2), and a is the percentage of correctly classified observations. The test statistic is asymptotically Z 2 -distributed with one degree of freedom. The test produces a p-value very close to 0: The null-hypothesis of a random distribution of observations across the groups by the logit model can thus be rejected at the 1% significance level. Likelihood ratio test518: Likelihood (L) is a measure of the probability of observing the actual values, given the parameter estimates. Deviation is defined as -210g(L), and is a measure of the fit of the model. The likelihood-ratio test compares the deviation of the null-model (i.e., a model where all coefficients are set to 0) with the deviation of the actual model. A high difference between the two indicates that the parameters contribute to the explanation of the dependent variable. The difference is asymptotically z2-distributed with n degrees of freedom (n=number of independent variables). The difference is 16.10, which with three degrees of freedom corresponds to a critical value for a significance level of 0.0029. Therefore, the null-hypothesis of all coefficients being 0 can be rejected at the 1% significance level. Pseudo-Re-statistics: Pseudo-Re-statistics attempt to provide interpretations similar to R2-statistics in the linear regression model. Two commonly used versions are the McFaddens_R 2 519 and the Estrella-R 2 520, defined as

McFaddens - R 2 -- 1 - LLULLc and E s t r e l l a - R 2 = 1 ~,LLc ) where LL~ is the unconstrained log-likelihood of the model to be tested, LLc is the loglikelihood of the null-model, and K is the sample size. Estrella-R 2 is a refined version of McFaddens-R 2 and has the advantage of being interpretable across the whole range from 0 to 1 in a fashion very similar to the standard linear regression model's R 2. For

517

See 518 See 519 See 520 See

Backhaus/Erichson/Plinke/Weiber, p. 446. Backhaus/Erichson/Plinke/Weiber, p. 439-440. Backhaus/Erichson/Plinke/Weiber, p. 440-441. Estrella (1997), p. 7-9.

215

McFaddens-R 2, this may not be possible for all parameter values. 521 This also justifies performing an F-test based on Estrella-R 2, analogously to a standard linear regression, with F

__

Estrella- R 2

n- k - 1

1- Estrella- R 2

k

where k is the number of independent variables. McFaddens-R 2 is 0.40 and Estrella-R 2 is 0.34, indicating a fairly good model fit. 522 The F-test using the Estrella-R 2 produces a p-value of 0.0022" The hypothesis of all coefficients being 0 therefore can be rejected at the 1% significance level. Appendix 55 also shows all discussed values for the 75/25 alternative. Results are very close to those of the 0/100 case.

7.5.2

L o g # M o d e l 2: S e c o n d e v e n t vs. n o s e c o n d e v e n t

In the second model, the dependent variable is set to 1 for all sample firms with any second event, i.e., either a reacquisition, a sell-off, a spin-off or another second event, and 0 otherwise. The same dependent variables as in model 1 are used, with the exception of the over/undervaluation variable. The obvious complexity is how to assess the valuation level for an event which does not occur. Potentially, some average measure of over/undervaluation for an arbitrarily defined period of time could be constructed, requiring business segment level data for a multi-year period. This type of data is not readily available in commercial databases, and the corresponding analysis is left for future research. Appendix 56 shows the main results of the second logit regression. All three coefficients have the expected sign: fl(SAME) is negative, indicating that same-industry subsidiary firms are less likely to experience a second event; fl(FD) is positive, indicating that a subsidiary firm is more likely to experience a second event when the parent firm is listed in a financially better developed market; and fl(IND) is negative, indicating that a subsidiary firm is less likely to experience a second event when it is active in a high uncertainty industry. The overall quality of the model seems good. The percentage of correctly classified observations is 62.8%, still higher than the 50%-level expected if observations were attributed to the two possible outcomes at random. Press's Q-test produces a p-value of 0.0014, allowing the rejection the null-hypothesis of a random distribution of observations across the groups by the logit model at the 1% significance level. Similarly, the likelihood ratio test leads to a p-value of 0.0009, allowing the rejection of the null-hypothesis of all coefficients being 0. McFaddens-R 2 and Estrella-R 2 are 8% and 10%, respectively. While this is below what standard literature considers a good 521

522

See Estrella (1997), p. 10-13. While Mc-Faddens-R2 can be interpreted for 0 and 1, values between 0 and 1 are not linearly related to the odds-ratio, meaning that for some parameter values an increase in R2 is associated with a lower odds-ratio, which makes interpretations of R2 between 0 and 1 unreliable. See Backhaus/Erichson/Plinke/Weiber, p. 441, there with reference to Urban (1993), p. 62: Values between 0.2 and 0.4 already indicate a fairly good model fit.

216

fit 523, the F-test using the Estrella-R 2 produces a p-value of 0.0022" The hypothesis of all coefficients being 0 therefore can be again rejected at the 1% significance level. Appendix 56 also shows all discussed values for the 75/25 alternative. Again results are very close to those of the 0/100 alternative.

7.5.3 Interpretation of results Given Klein/Rosenfeld/Beranek's (1991) finding that parent firms announcing a sell-off earn positive abnormal returns while reacquisition announcements cause insignificant abnormal returns, a model allowing the assessment of the probability that the second event will be a sell-off, rather than a reacquisition, may produce valuable trading strategies. If a second event occurs, it is more likely to be a sell-off when the subsidiary firm is overvalued relative to its intrinsic value. This is intuitive, because the parent firm profits from a high valuation of its subsidiary firm by selling its remaining stake. Crossindustry subsidiary firms are also more likely to be sold off, implying that if uncertainty about synergies is the primary motivation for the initial partial floating, this uncertainty is more likely to be resolved by finding that synergies are not sufficient to keep the two firms together: The separation of such parent/subsidiary firm combinations removes more negative synergies than its continued co-existence would create. Parent firms listed in a financially better developed market are also more likely to sell off the subsidiary firm. This finding is another indication of a result found in previous analyses: In section 3.5.3 it was found that ECO announcements after 1998 earn higher announcement period returns than in earlier periods, and in sections 4.6.5.2 and 5.6.5 it was found that subsidiary firms carved out after 1999 perform worse operationally and in the stock market than subsidiary firms carved out earlier. The interpretation provided was that the decline in the relative value of internal capital markets caused by an increase in the efficiency of external capital market has increased the opportunity costs of an internal capital market, leading to higher gains for the announcement of a separation, but also to more separations where the quality of the subsidiary firm as a stand-alone company is questionable. Sell-offs being more likely in better-developed financial markets are consistent with this finding in that they indicate an educated market's preference for separately traded parent and subsidiary firms because of the lower value of internal capital markets. Finally, a subsidiary firm is more likely to be reacquired when it is active in either the telecommunications, finance, insurance or service industry, which are characterised by a higher degree of uncertainty about their future development prospects, but also by a higher growth potential. This confirms theoretical considerations by Kranenburg/Perotti/Rossetto (2004), and implies that more uncertainty and more growth potential increase the probability that synergies between parent and subsidiary firms will be found to be positive, making a reacquisition more attractive than a sell-off. 523

See Backhaus/Erichson/Plinke/Weiber,p. 441, there with reference to Urban (1993), p. 62: Values between 0.2 and 0.4 already indicate a fairly good model fit.

217

However, how can the investor know that a second event is likely to occur at all? For this purpose, the second model is estimated. Knowing that a second event will occur also opens up another trading possibility, since Klein/Rosenfeld/Beranek (1991) also show that subsidiary firms experience positive abnormal returns for both sell-off and reacquisition announcements. The independent variables have the same signs as in the first model: The probability of a second event occurring increases when parent and subsidiary firms are from different industries. The probability also increases when the parent firm is listed in a financially better developed market. This is consistent with the notion that parent firms in less developed financial markets need more time to gather information about the potential resolution of uncertainty regarding the synergies between parent and subsidiary firms, because less developed financial markets are less efficient at providing the required information. Finally, the probability of a second event occurring decreases when the subsidiary firm is active in a high uncertainty industry. If prospects in an industry are more uncertain, then deciding whether synergies between parent and subsidiary firms are either sufficient to justify a co-existence, or insufficient leading to a separation, takes longer. Again, this finding harmonises with Kranenburg/Perotti/Rossetto (2004), who predict that the higher the level of uncertainty, the later it is optimal to exercise either the buy-back or the sell-off option. 524 Overall, the models seem to do a good job at predicting the respective probabilities. Given the group sizes, which are small particularly for the first logit model, it would be interesting to see whether results also hold in other samples. Performing an out-ofsample test with US ECOs, which are more numerous, seems a promising objective for future research.

7.6

Conclusion

Similar to findings in the US, ECOs in Europe are temporary structures in almost two thirds of all cases. Sell-offs seem to occur more often in Europe than in the US. Differences in the development states of European vs. US financial markets serve to explain this finding: Less developed European markets may not allow a complete selloff in the initial ECO for reasons of limited market capacity, requiring parent firms to go through the process of a staggered sale. An ECO may also produce valuable information for the parent firm regarding the market's valuation of the subsidiary firm, while more developed US markets may provide this information without a prior partial floating. Despite these differences, other findings in the present study confirm previous findings from the US: Sell-offs are more likely in the first years after the initial ECO, and reacquisitions are more likely in subsequent years. Also, the probability of a reacquisition increases with the stake retained. The study systematically analyses potential determinants of the decision which second event, if any, occurs. Subsidiary firms reacquired by their parent firms are more likely 524

See Kranenburg/Perotti/Rossetto (2004), p. 24.

218

to have been undervalued by the market relative to their intrinsic value, and subsidiary firms completely sold off are more likely to have been overvalued by the market prior to the second event. This intuitive result suggests that parent firms take advantage of perceived misvaluations by either selling an overvalued stake in the subsidiary firm, or by reacquiring an undervalued stake. The finding is robust to alternative measures of relative valuation. Some evidence is found that such behaviour is more likely in countries whose financial markets are less developed, and where shareholder rights are consequently assumed to be less pronounced, allowing the parent firm to get away with what may be perceived as opportunistic behaviour. Firms in less mature and more rapidly changing industries where synergies between parent and subsidiary firms are less certain seem to wait longer until a second event, and are more likely to reacquire compared to firms in more mature and less uncertain industries. No evidence is found that either the parent firm's debt burden or the efficiency of the ICM prior to the initial ECO influence the decision whether to reacquire or sell off the subsidiary firm. The results from the univariate analyses are drawn together in two logit models. The first model assesses the importance of certain variables for the probability that a parent firm will either reacquire or sell-off its partially floated subsidiary firm. The second model assesses the importance of these variables for the decision of whether to engage in any second event at all (i.e., either reacquisition, sell-off, spin-off, or other), or whether to leave the stake of the subsidiary firm outstanding. The results in these two models largely confirm results from the univariate analyses. In particular, a second event is more likely overall, and is also more likely to be a sell-off, when parent and subsidiary firms are from different industries, when the parent firm is listed in a financially better developed market, and when the subsidiary firm is active in a more mature industry. While overall model quality seems good, generalizability of results may be limited due to the low number of cases in the reacquisition group. Since Klein/Rosenfeld/Beranek (1991) find that parent firms announcing a sell-off earn positive abnormal returns whereas reacquisition announcements lead to insignificant abnormal returns, and subsidiary firms earn positive abnormal returns for either announcement, further developing the suggested models, e.g., through their application on a US ECO sample, seems a promising (and potentially profitable) avenue for future research.

219

8

Conclusion

This final chapter summarises the key findings of the individual chapters (section 8.1), highlights recurring themes and their implications both for firms intending to engage in an ECO and for potential investors, as well as for academics interested to do research on ECOs (section 8.2), and concludes by listing future research opportunities in this field (section 8.3).

8.1

Summary of key findings

Chapter 2 defines the research object as the IPO of a subsidiary firm. ECOs are characterised by a dual nature as both measures of portfolio and financial restructuring. The sample of 178 European ECOs over the 1984 to 2004 period is derived by an extensive database and newslines search, and is distributed similarly across industries and countries, compared to all listed European firms. The frequency of ECOs is linked to stock market sentiment and overall IPO activity. Parent firms on average continue to hold a majority stake in the subsidiary firm, and display considerable size differences. Underpricing of a magnitude similar to standard IPOs tends to occur in ECOs. A firm's motivation to engage in an ECO is determined both empirically 525 and theoretically 526 The basic concepts of the efficient market hypothesis (EMH) as the underlying theory of the thesis are described. Documented anomalies are either interpreted as investor irrationality, or as reflecting unknown sources of risk. Psychological biases exist both among market practitioners and academics, supporting existing views of reality in the face of respective counter-evidence. The current consensus is that there is more to return than market risk, but the nature of the 'more' is disputed. Finance theory, initially having originated from the desire to explain investor and market behaviour as following well-defined equations similar to 19 th century physics, is now following 20 th century physics by admitting that reality is more complex than established models assume. Chapter 3 analyses a firm's share price reaction to an ECO announcement. Average abnormal returns for European ECOs are positive. However, sample firms also show abnormally negative returns in the two days after the announcement, cancelling some of

525

In practice, ECOs are found to be motivated by a firm's desire to refocus on its core operations, to exit a loss-making business, to obtain capital for investments or the repayment of debt, to diversify its investor base, to obtain an appropriate valuation of its subsidiary firm, to increase the flexibility of the subsidiary firm, or to comply with regulatory requirements. 526 In theory, an ECO can create value for parent firm shareholders by raising 'cheap' capital (i.e., if the subsidiary firm's industry is valued more highly than the parent firm's industry, or if the country in general has higher valuation levels), by increasing transparency through a higher demand for and supply of information regarding the subsidiary's business, by sending a positive signal regarding the undervaluation of the parent firm, by avoiding an underinvestment problem, by completing imperfect capital markets through the establishment of a pure-play company, by replacing internal control mechanisms subject to principal-agent issues with more efficient external control mechanisms, by enabling employee participation in the firm, and by transferring wealth from other stakeholders to parent firm shareholders.

220

the previous gains. This is mostly due to negative returns for ECOs from the 'hot market' period of 1998 to 2000, where a high level of initial enthusiasm has often been followed by disenchantment. Announcement period returns are higher when pre-event information asymmetry is high (higher transparency for investors), when parent and subsidiary firms are from different industries (removal of negative synergies), when relative transaction size is large (higher relative importance for parent firm), for ECOs occurring after 1998 and in countries with higher shareholder rights (differences in the relative value of internal to external capital markets across both time and space), when a company states refocusing of the parent firm or development of the subsidiary firm as its motivation, and when pre-event profitability is low ('why change a winning team'). Together, these findings tell firms when an ECO is likely to be welcomed by capital markets. The analysis of three additional dates of interest reveals that abnormal returns are highest on the first rumour date, close to zero on the first day of bookbuilding, and negative on the first day of trading, but increasing in the subsidiary firm's first day performance. An analysis gauging the value gains from an ECO thus cannot only focus on announcement period returns, but needs to consider additional time periods. Non-announcing companies with future ECO candidates tend to show abnormal price reactions to other firms' ECO announcements. The reaction is more pronounced for firms owning subsidiaries in the same industry as the announcing firm's subsidiary. ECO announcements thus have implications for rival companies in the same industry, and firms are well advised to watch corresponding moves by their competitors. Chapter 4 analyses the long-term operating performance of parent and subsidiary firms. There is no abnormal operating underperformance in the years prior to the ECO, refuting the hypothesis that parent firms undertake an ECO to address a critical company condition. Both parent and subsidiary firms grow stronger and are m o r e profitable than benchmark firms in the ECO year, and are less profitable in following years. Profitability and growth then return to normal levels for parent firms, whereas subsidiary firms show continued growth. This implies that positive ECO effects are permanent for some subsidiary firms, whereas parent firms do not generally profit as predicted by the divestiture gains hypothesis. Two potential explanations for this pattern in operating performance are earnings management and market timing: Evidence for the former is found in that subsidiary firm abnormal accruals are higher in the ECO year and lower in the following year. Evidence for market timing is presented in chapter 5. Two scenarios for successful ECOs are identified: In the first (a growth story), a financially non-distressed firm with a low number of business segments aims to develop the business of its subsidiary firm, in which it continues to hold a large stake. In the second (a restructuring story), a financially distressed firm addresses its critical situation by increasing its business focus through the sale of a considerable stake in its subsidiary firm, allowing management to channel its attention on the parent firm's core business.

221

Subsidiary firms grow when parent firms announce that they intend to develop the subsidiary's business, and when carved out of firms with fewer business segments. Subsidiary firms become more profitable when parent firms are not financially distressed before the ECO (implying that subsidiaries inherit features of their previous existence), when parent firms retain a larger stake (more intense monitoring, coupled with public market scrutiny), when carved out from parent firms in a different industry (elimination of negative synergies), and when carved out before 1999 (either because of lower quality ECOs in the 'hot market' period, or because of higher opportunity costs of internal capital markets resulting from more developed external capital markets, and parent firms' consequent desire to partially close their ICMs through an ECO). Chapter 5 analyses the long-term price performance of parent and subsidiary firms. Abnormal returns are negative, but results are mostly significant only when equalweighting, rather than value-weighting. BHARs produce results similar to the calendar time method. Underperformance is thus more likely for smaller firms, but heterogeneity in sample firm size limits generalizability. Method-dependent results are also a reflection of the bad-model problem, emphasizing the necessity of developing a more powerful and commonly accepted model of normal returns. Subsidiary firms outperform appropriate benchmarks in the first months following the ECO, potentially due to a self-attribution bias by investors regarding their stock picking ability, continued earnings management to prevent litigations, and supporting share purchases by the ECO firms and the underwriting investment bank to facilitate future SEOs. A market-timing hypothesis is supported by a positive price performance in the 12 months prior to the event, a negative relationship between pre-event and post-event performance, and a stronger development of valuation levels in the subsidiary firm's industry relative to other industries over the 36 months preceding the ECO. Parent firms apparently time the ECO to occur in periods of high prices, and utilise periods of high relative valuations in the subsidiary firm's industry by selling subsidiary firm equity. Parent firms which grow and become more profitable after the ECO also have a better share price performance. Post-event performance is partially a correction of the announcement period return, indicating an overreaction by investors to the initial ECO announcement. Subsidiary firms perform better when carved out from non-financially distressed parent firms (apparently inheriting features of their previous existence), from parent firms in a different industry (elimination of negative synergies), when they were profitable before the ECO (suggesting that market value is driven by fundamental values), and when carved out before 1999 (mirroring results from the LTOP analysis with similar interpretations). Chapter 6 analyses the information content of ECO announcements regarding the efficiency level of ICMs. The market reaction to the ECO announcement is more positive when the firm's ICM is large, and when its size decreases after the ECO. Market reaction is less positive when the firm's ICM functioned efficiently, and when its efficiency decreases. The results imply that investors on average disapprove of ICMs

222

and consider them inefficient. Also, investors are partially able to discern efficient from inefficient ICMs, implying a considerable level of investor sophistication. There are two cases when ICMs add value. First, when a firm's business segments are related (proxied by positively correlated segment cash flows), the firm is able to crosssubsidise activities by allocating capital to segments with high investment opportunities. Since activities are related, management has the required knowledge to rank investment opportunities. Second, when a firm's business segment cash flows are negatively correlated, the firm is able to switch capital between activities, depending on the resolution of uncertainty, and thus profit from multiple states of the world. When cash flows are uncorrelated, ICMs do not add value: Their partial closure through an ECO is rewarded by markets through positive announcement period returns. Chapter 7 analyses the frequency and determinants of the second event. Sell-offs seem to occur more often inEurope than in the US, potentially because market capacity is limited in less developed European markets, requiring a staggered sale; and because partially floating the subsidiary firm produces information regarding the market's valuation views. More developed US markets provide both a higher market capacity and better information prior to the floating, enabling a complete immediate sell-off. As in the US, sell-offs are more likely in early years, whereas reacquisitions are more likely in later years. The probability of a reacquisition increases with the stake retained. The second event is more likely to be a sell-off (reacquisition) if the subsidiary firm is overvalued (undervalued) by the market relative to its intrinsic value, suggesting that parent firms take advantage of perceived misvaluations. Such behaviour is more likely in countries with less developed financial markets and shareholder rights. A sell-off is also more likely for cross-industry ECOs, where negative synergies between parent and subsidiary are more likely. Similarly, firms in less mature industries, where synergies are uncertain, wait longer until a second event, and are more likely to reacquire compared to firms in more mature industries, confirming implications from Kranenburg/Perotti/Rossetto's (2004) model. Neither the parent firm's debt burden nor pre-ECO ICM efficiency influence the second event decision. Two logit models assess the probability that a parent firm will either reacquire or sell-off, and that any second event will occur at all, respectively. The two models confirm the results from the univariate analyses, and represent the first step of a trading strategy aiming to profit from abnormal share price reactions to second event announcements.

8.2

Recurring themes and their implications

Throughout the analysis of short- and long-term performance of parent and subsidiary firms a series of returning topoi have manifested themselves:

Breaking up cross-industry combinations creates values. Cross-industry ECOs earn higher announcement period returns, lead to a higher growth for the parent firm, and both a higher profitability for and a better share price

223

development of the subsidiary firm in the years after the ECO. This result supports the idea that conglomerates whose business segments are unrelated are characterised by negative synergies. These can arise, e.g., because group management does not possess either the capacity or the knowledge to adequately manage diverse businesses, or because unrelated business segments increase the probability of capital misallocation (as in Rajan/Servaes/Zingales' (2000) model). Removing these negative synergies through an ECO creates value for all parties involved. The subsidiary firm inherits features of its previous existence. If the parent firm has been financially distressed prior to the ECO, both the subsidiary firm's profitability and its share price performance will be negatively affected. Apparently, some of the causes of the parent firm's financial distress are transferred to the subsidiary firm in the ECO. For example, if the parent firm's financial distress has been caused by inadequate management practices or inefficient processes, an ECO is unlikely to resolve these issues, and the subsidiary firm will be hampered in its operating and its price performance even when it becomes a stand-alone firm. This implies that an ECO is not a remedy for all problems: It should be accompanied by a review of internal practices to determine whether a potential sub-optimal performance prior to the ECO is due to the conglomerate nature of the parent/subsidiary firm combination, or due to more fundamental managerial and operational issues. The distressed parent firm will also grow less, but its profitability is likely to improve following the ECO. This indicates that an ECO can be used to deal with a distress situation and may help to improve profitability (presumably from a lower level). However, it cannot help to overcome all related problems, and continued growth seems to require a solid financial basis. The notion that the subsidiary firm continues to be affected by pre-ECO characteristics is further supported by the finding that subsidiary firms carved-out from parent firms with a large number of business segments tend to show a worse profitability and lower growth than other subsidiary firms. Subsidiary firms have ties with other business segments (e.g., shared operations and administrative functions), which are stronger when there are more segments in the conglomerate. Once the subsidiary is separated from the parent firm, it has to 'learn' to be independent, resulting in slower growth relative to benchmark firms. This result confirms qualitative considerations by Nick (1994), who speculates that successful ECO candidates are subsidiary firms able to conduct their business independently from the parent firm, and whose links to other business segments are limited. The optimal post-event stake is a trade-off. If the parent firm retains a higher stake in the subsidiary firm, the latter is likely to become more profitable, and the former is likely to grow stronger. Amplified public scrutiny following the partial floating of the subsidiary firm in combination with an increased monitoring by the parent firm seem to be a sound basis for improved subsidiary firm profitability, which ultimately benefits the parent firm due to its continued engagement in the carved-out firm. However, parent firm profitability tends to be worse when retaining a larger stake. Hence, if the aim of the ECO is to improve

224

margins, a complete separation of ties between parent and subsidiary firms is necessary. Potentially such a complete separation allows the parent firm to fully concentrate its attention on the required restructuring, as well as on its future core business. This result implies a trade-off regarding the optimal stake to be sold to the public in an ECO: A lower stake sold benefits the subsidiary firm (and thus also indirectly the parent firm in terms of growth) because of more intense monitoring by the parent firm, whereas a higher stake sold benefits the parent firm in terms of profitability because its management can concentrate on the core business.

Larger is better. Larger subsidiary firms both have a better profitability and a better share price development in the years following the ECO, indicating that a certain critical firm size may be required to become a viable stand-alone company. Also, larger ECOs (in relative terms) earn higher announcement period returns because of their relatively higher importance for the parent firm. In ECOs, size does matter.

Motivation counts. The motivation stated by the parent firm when announcing the ECO is linked to both short- and long-term value gains. In the short term, parent firms announcing that they will use the ECO proceeds to develop the business of the subsidiary firm, as well as parent firms announcing that they intend to refocus on their core business, earn higher returns than similar firms not announcing such a motivation. In the long term, the desire to develop the subsidiary firm's business benefits both parent and subsidiary firms in terms of growth, whereas the desire to refocus on the core business leads to a higher parent firm profitability. These results suggests two ideas: First, there are two distinct scenarios when an ECO is successful: A growth scenario (associated with the first motivation, and also with a higher stake retained and a state of no financial distress), and a restructuring scenario (associated with the second motivation, and also with a lower stake retained and a state of financial distress). Second, parent firms on average seem to be honest when stating their motivations for the ECO. If they were not and, e.g., attempted to profit from announcements which they believe are favoured by the market, firms with such positive-type announcements would not on average earn higher announcement period returns, nor grow stronger or become more profitable in the years after the announcement. An implication for researchers is that any future analysis on ECOs should bear these two markedly different scenarios in mind.

ECO performance differs by period. Subsidiary firm profitability, as well as the share price performance of both parent and subsidiary firms, is better for ECOs outside of the 'hot market' period 1998 to 2000. However, the fact that announcement period returns are not higher for this market period, but a r e higher for all ECOs carved out aider 1998, suggests an altemative interpretation: In earlier periods, only subsidiary firms with a clear potential for profitability improvement were carved out, because the value of intemal capital markets was higher relative to less developed extemal capital markets. In later periods, the relative value of intemal capital markets decreased and their opportunity costs

225

increased, and hence the value of subsidiary firms as components of internal capital markets also decreased. Subsidiary firms now are not only carved out when there is a clear potential for profitability improvement, but also for the sake of partially closing costly internal capital markets. This may have decreased the average profitability improvement. This suggests that parent firms face a trade-off when deciding whether to carve out a subsidiary firm: On the one hand, subsidiary firms are components of internal capital markets and thus valuable. On the other hand, internal capital markets can be associated with opportunity costs. The result of the trade-off is determined by two factors: The development state of the external capital market, and the potential for profitability improvement of the subsidiary as a stand-alone firm: The more the external capital market is developed, and the higher the potential for profitability improvement, the higher the incentive for the parent firm to close down the internal capital market by carving out the subsidiary firm.

Fundamental values drive market values. The price performance of parent firms is directly linked to both their growth and their profitability. Similarly, subsidiary firms who have been profitable prior to the ECO are more likely to have a positive share price performance following the ECO. However, profitability may also be a hindrance to ECOs in that parent firms profitable before the ECO earn lower abnormal returns to the announcement of an ECO than their less profitable counterparts. This supports the idea that investors are wary if an already profitable parent firm intends to change its portfolio composition through an ECO, as upside potential may be limited. 527 The analyses have also revealed a number of methodological insights:

Analysis of long-term price performance is both essential and contested. It is essential because of ample existing evidence of psychological bias-driven investor behaviour leading to price drifts and trend reversals over longer time horizons. It is contested because current methodologies lack the power to produce consistent results. Method-dependency of results prevents attributing empirical findings to either characteristics of the research object or inadequate testing procedures. This realization underlines the importance of further research aiming to formulate a more reliable and commonly accepted model of asset price returns. Such a model of normal returns is the indispensable underpinning of any test of abnormal price performance hoping to produce robust results. Existing research results on US ECOs are not generalizable without re-testing. In the analysis of short-term price performance, it was found that parent firms in countries with higher shareholder rights earn higher abnormal returns when announcing

527"If it ain't broke, don't fix it." (credited to Bert Lance, a US government representative under Jimmy Carter, apparently quoted in the May 1977 issue of "Nation's Business" as the response to the question on how to save money for the US government. For the linguistically inclined reader, the British equivalent is 'let well alone'. See http://www.highbeam.com/ref/doc0.asp?DOCID= 1090:IfitaintBROKEdontfixit& num=1&ctrllnfo= Round18% 3AMode18c%3AREFSR%3AResult).

226

an ECO than parent firms in countries with lower shareholder rights. In the second event analysis, both inter-continental differences (between US and Europe) as well as intra-continental differences (between European countries with differently developed financial markets) were identified. Hence, future research must control for the country factor if the legal setting or the development state of the financial market are expected to drive results. Announcement dates must be 'clean'. The analysis of short-term price performance revealed that abnormal returns on contaminated dates are significantly higher than on clean dates. Researchers are therefore required to control for announcement date contamination, even if it reduces sample size. Additional periods need to be considered when assessing value creation. Researchers analysing the value implications of certain corporate finance events may have to consider other time periods than merely the announcement period of the event. The analysis of short-term price performance revealed that in addition to significantly positive returns on the announcement date, returns are significantly positive on the first rumour date and significantly negative on the IPO date. The analysis of long-term price performance suggests the possibility of a long-term underperformance (potentially due to market timing). Together these findings imply that focussing solely on the announcement of an event may not capture the whole story. Naturally, considering additional time periods and longer time horizons increases the probability of result contamination by other events. The implied trade-off is unlikely to have a general solution but should be borne in mind when determining the test design. Dual nature of ECOs is reflected in short- and long-term performance. ECOs have a dual nature as both a restructuring and a financing mechanism. This dimorphism is reflected in the (positive) short term market reaction to an ECO announcement being similar to restructuring activities like spin-offs, and the (slightly negative) long term market reaction being similar to financing activities like IPOs and SEOs. Researchers analysing value creation in ECOs will need to control for this in future studies. Results support the notion of capital market efficiency. The thesis modestly contributes to the discussion about market efficiency. The analysis of internal capital markets revealed that investors view ICM negatively and are able to discern efficient from inefficient ICMs. This signals a level of investor sophistication previously doubted by some. 52s To the degree that sophistication proxies for rationality, this provides support for the existence of one of the requirements for market efficiency. Similarly, the finding of a method-dependency of results regarding the long-term price

528

The literature abounds with examples of apparent investor irrationality. See section 2.3.4 for a list of such pricing anomalies, and footnote 479 for an example of how investors fail at seemingly straightforward analyticaltasks.

227

performance, together with the low absolute levels of negative BHARs, support the view of fairly efficient capital markets.

8.3

Avenues for future research

An empirical researcher may feel like Hercules fighting Hydra: In the process of answering one question, (at least) two new questions are likely to rear their heads. Unlike Hercules, the researcher should not feel daunted by this: As pointed out by Thorstein Veblen, "The outcome of any serious research can only be to make two questions grow where only one grew before ''529. The analysis of abnormal long-term price performance has made it clear that the limiting factor, both for this and for future studies, is the lack of a model able to produce normal returns. The review of existing literature on market efficiency and potential anomalies in section 2.3.4 has also made it clear that building such a model is conceptually challenging because it is unclear whether violations of the current form of the EMH are due to investor irrationality, or whether they are manifestations of hitherto unexplored sources of risk. Progress in this understanding is highly desirable and required to formulate a sound basis for future studies of long-term price performance. The present study has implemented a series of tests requiring a considerable amount of data for each sample firm. This data has not always been available due to limitations both in the availability of historic IPO prospectus and in the level of detail available in commercial databases. Such limitations may be less severe in the US, compared to Europe, for two reasons. First, the Compustat Business Segment Information database provides basic accounting data for US firms reporting multiple lines of business. Such a database is regularly used in US empirical research 53~ and allows, e.g., the construction of the ICM measures used in this study without having to manually collect segmentlevel data for all sample firms as was done in this study. Second, the number of US ECOs greatly exceeds the number of European ECOs. TM Hence, an analysis using a US sample may afford to lose more firms due to limited data availability before reaching a statistically critical smallness. This is particularly required when constructing variables which are as data-intensive as the measures of ICM size and efficiency analysed in chapter 6, where the present study has reached (and may have fallen below) critical size thresholds. Similarly, the analysis of the determinants of the nature of the second event in chapter 7 is hampered by the fact that the total number of reacquisitions in Europe is

529 Quotation found on http://www.chemistrycoach.com/question.htm, there quoted from Veblen's work "The Place of Science in Modem Civilization and Other Essays". 530 See Rajan/Servaes/Zingales (2000), p. 54, Chevalier (2000), p. 7 and Wulf (2002), p. 39. 531 Vijh (2002) finds 336 ECOs in the 1980-1997 period. Schipper/Smith (1986) find 48 ECO announcements in the 1965-1979 period, together totalling 384 ECOs in the US in the 1965-1997 period.

228

limited532, potentially raising concerns about the significance and hence generalizability of results. Therefore applying the tests developed in this study on a sample of US ECOs will help to increase confidence in (or refute) the results presented in this study. In chapter 6, it was found that investors seem able to judge the efficiency of ICMs. Doubt was raised as to whether investors are likely to go through the lengthy process of constructing the specific ICM measures used in the present study. Rather, it seems more likely that simpler proxies for the efficiency of ICMs exist. Finding such proxies may benefit future research in being able to control for the level of ICM efficiency, which has been found to drive results, without having to spend considerable time and resources on constructing such proxies. In chapter 7, two logit models were developed to assess the probability of whether any (and if so, which) second event is likely to occur. While these models may serve as a first step in the creation of a profitable trading strategy, additional work is required. In particular, the models do not specify when the second event is likely to occur. Conceivably, the level of relative over/undervaluation of the subsidiary firm could drive the timing of the second event. Assessing this level requires the construction of a time series of relative valuation measures. Given that the calculation of the intrinsic value requires business segment level data 533, the issue again is one of limited data availability in a European database context: Manually collecting business segment level data for a sample of 178 ECOs for a multi-year period seems prohibitively expensive. Companies continually announce plans for future ECOs. In Germany, changes in the corporate tax law in 2002 have enabled conglomerate firms to conduct ECOs without incurring taxable revaluation gains. Infineon has announced it is considering a carve-out of its memory chip business. TM RWE has announced its intention to list a minority stake in its Thames Water business unit. 535 Altana aims to list its pharmaceutical business unit. 536 German conglomerates continue to own various businesses where synergies are unlikely, and which therefore represent potential candidates for an ECO. Practitioners regularly mention Linde 537 and MAN 538 as candidates for conducting an ECO of one of their respective subsidiaries. While TUI has decided against a floatation of its Hapag Lloyd container shipping business unit in August 2005, the unit still remains a candidate for a divestiture. 539 In other European countries, potential ECO candidates include

532

See Appendix 51: Out of the 178 sample firms, 21 were reacquired within the sample period (21 is based on the 75/25 alternative. When using the 100/0 alternative, the number of reacquired firms is even smaller, i.e., 11). 533 See section 7.4.2.4 for more details. 534 See Financial Times Deutschland, 17.11.2005, www.ftd.de. 535 See Financial Times Deutschland, 24.10.2005, www.ftd.de. 536 See Financial Times Deutschland, 30.10.2005, www.ftd.de. 537 Linde has three business units: Industrial gases, plant engineering and material handling. 538 MAN has five business units: Commercial vehicles, industrial services, printing systems, diesel engines and turbo machines. See also manager-magazin, 14.11.2005. 539 See Financial Times Deutschland, 22.8.2005,www. ftd. de.

229

Winterthur, the insurance subsidiary owned by Credit Suisse in Switzerland; 54~ Experian, the retail intelligence and analysis business owned by GUS (who already has ECO experience with its floatation of Burberry) in the UK; Arkema, the petrochemicals business owned by Total in France; and Huskvamer, the outdoor business owned by Electrolux in Sweden. TM ECOs thus will continue to be a popular corporate restructuring tool and will feature prominently in acaderfiic research.

540

541

See press release by Credit Suisse from December 7, 2004, available on http://www.creditsuisse.com/de/news/pa_mr_browserde.jsp?ns=37064. For the last three cases see UBS InvestmentResearch (2005), p. 4.

231

Appendix

~

~.-'=

~~

ell

r ~

.-~

..= 0 0 t"q

-8

~~

r~

~

co

< Z

~0

==

~.~

oj~

t~

I n~

~.-

"0

oo= z ~

r~

o~

r~

.~.

r

d~

0

~

er

[-~ . ~

r

o

o r~

e

0

..~

~

.= o 9,,~

r~

r~

r 0 0 r

ell

0 oo

l'--

=

~

~

9, ~ r~

o~

"N

9~ ?~ 0.=

o

~g

o~

r.~

om

0 0

=

~0 0 0

0

0

233

Appendix 2" Derivation of sample This table indicates how the sample in this thesis was derived. See section 2.1.5 for more details. See section 2.1.2 for the criteria of the specific ECO definition used in the derivation of the sample.

Companies

Companies left after deduction

All European capital issues in SDC in last 20 years

20,067

IPO flag N (=non-IPOs)

12,794

7,273

1,487

5,786

902

4,884

Transaction size below US$10Mio Missing transaction size Transaction type: Follow-on offering

86

4,798

Parent business: Venture capital

54

4,744

4

4,740

10

4,730

Parent business: Trust Parent business: Treasurer Office Parent business: State

8

4,722

13

4,709

162

4,547

6

4,541

Parent business: Leverage and investment firms

50

4,491

Parent business: Central government and city

10

4,481

Parent business: Government

115

4,366

Spin-off flag P (=privatization)

219

4,147

3,015

1,132

954

178

Parent business: Private equity Parent business: National government Parent business: National bank

Identical issuer and parent firm name Firms not meeting criteria of ECO definition

234

Appendix 3: Sample by year and country This table shows the composition o f the sample by year o f announcement and by country o f the announcing parent firm (BE=Belgium, CH=Switzerland, D=Germany, FN=Finland, FR=France, GR=Greece, IT=Italy, NL=Netherlands, N O = N o r w a y , PT=Portugal, SP=Spain, UK=United Kingdom). Announcement years could not be identified for two sample firms.

Country/ BE CH

D

FN

2

0

FR GR

IT

NL NO

PT

SP SW UK

Total

Year 1984

0

0

0

0

0

0

0

0

0

0

0

2

1985

0

0

0

0

0

0

0

0

0

0

0 . 0

0

0

1986

0

0

0

0

0

0

0

1

0

0

0

0

0

1

1987

0

0

1

0

0

0

0

0

0

0

0

0

0

1

1988

0

0

0

0

0

0

0

0

0

0

0

0

1

1

1989

0

0

3

0

0

0

1

1

0

0

0

0

0

5

1990

0

0

3

0

0

0

0

0

0

0

0

0

0

3

1991

0

0

3

0

0

0

0

0

0

0

0

0

0

3

1992

0

0

4

0

0

0

0

0

0

0

0

0

1

5

1993

0

0

5

0

1

0

0

0

1

0

0

2

4

13

1994

0

0

6

0

0

0

0

0

2

0

1

3

1

13

1995

0

1

2

1

1

0

0

1

0

0

0

1

1

8

1996

0

3

3

0

2

0

1

0

0

0

0

0

1

10

1997

1

1

1

1

3

0

1

1

2

0

0

3

3

17

1998

1

0

11

0

1

0

2

0

0

0

0

1

2

18

1999

0

1

12

2

2

0

0

1

0

1

2

0

3

24

2000

1

2

9

1

6

1

0

0

1

1

4

2

7

35

2001

0

1

1

0

0

0

1

1

0

0

0

0

1

5

2002

0

0

0

0

0

0

1

0

0

0

1

0

1

3

2003

0

0

2

0

1

0

0

0

1

0

0

0

1

5

2004

0

0

0

1

1

0

1

0

0

0

1

0

0

4

NaN

0

0

1

0

0

0

0

0

0

0

0

0

1

2

Total

3

9

69

6

18

1

8

6

7

2

9

12

28

178

235

Appendix 4: Sample by year (relative to IPOs) This table lists the number of ECOs in all years of the sample period, and compares it to the number of all European IPOs in each year.

Year

ECOs

IPOs

% ECOs of IPOs

1984

2

9

22.2%

1985

0

21

0.0%

1986

1

29

3.4%

1987

1

34

2.9%

1988

1

22

4.5%

1989

5

39

12.8%

1990

3

30

10.0%

1991

3

52

5.8%

1992

5

52

9.6%

1993

13

102

12.7%

1994

13

322

4.0%

1995

8

251

3.2%

1996

10

384

2.6%

1997

917

488

3.5%

1998

18

530

3.4%

1999

24

780

3.1%

2000

35

1058

3.3%

2001

5

374

1.3%

2002

3

230

1.3%

2003

5

142

3.5%

2004

4

355

1.1%

NA

2

NA

NA

4 0

0 k

00 0 ,me

0

0 ,--i ;>

X

r~ 0

r~

r~ ~J

= C~ r~

r~

r~

r~

m r~

~176 ~ 9

=

=

r~

=

_= .=.

L~

r~

r.~

;>

~. =

~'~ o== !

i

!

!

!

i

!

C'q

O0

.,..~

,0

--

0 C",l

~ ~~

~-=-e e

;I

~~ ~

~

_~'~

.~

r,t)

~ "~ ~

~~.~

'~

~

~=

~~~~~~o.~

~

.~

.. -s

~o

m t~

Z ~Q

~

.o

c/~

.,_~

.~.~

~

m

0

~_~ ~, ~,~

N

~ o I

i

i

i

r

i

i

I

i

0

en .,..~ i

i

i,,,,,i

0 i

..~

~

~

,~

i

i

i

i

0 i

i

i

0 i

i

2

"i

~

i

241

Appendix 9: Sample by country (relative to IPOs) This table lists the number of ECOs in all countries, and compares it to the number of all European IPOs in each country.

Country

ECOs

IPOs

% ECOs of IPOs

BE

3

103

2.9%

CH

9

146

6.2%

69

844

8.2%

FN

6

119

5.0%

FR

18

874

2.1%

GR

1

213

0.5%

IT

8

371

2.2%

NL

6

195

3.1%

NO

7

155

4.5%

PT

2

62

3.2%

SP

9

125

7.2%

SW

12

208

5.8%

D

UK Total

28

1,889

1.5%

178

5,304

3.4%

242

Appendix 10: Existing studies on E C O announcement period returns - US This table summarises key existing studies on announcement period returns in ECOs in the US. ***, ** and * indicates significance at the 10%, 5% and 1% level, respectively.

Author

Year Country Period

N Method

Event window

C A R Significance

76 M R A R

[-44;-5]

3.1% no

52 M R A R

[-4;0] [-40;-5]

1.8% ** 2.1% no 2.8% ***

US

Schipper/Smith

1986 US

1963-1983

Klein/Rosenfeld/Berane

1991 US

1966-1983

Michaely/Shaw

9 1995 US

1981-1988

28 M R A R

[-4;0] [-1;0] [-1;0]

Slovin/Sushka/Ferraro

1995 US

1980-1991

32 M R A R

[-2;+2] [-10;-1]

0.4% no 1.3% no

Allen/MeConnell

1998 US

1978-1993

188 M R A R

[0;+1] [-1;+1]

1.2% ** 1.9% ***

54 MRAR

[-1;+1]

6.6% ***

1981-1995

60 M R A R 265 M R A R

[- 1 ;+ 1] [-1;+1]

0.0% no 2.3% ***

[-1;0]

2.2% ***

Hand/Skantz

1998 US

Powers Chemmanur, Paeglis

2001 US

1981-1989

181 M R A R

2001 US

1984-1999

19 M R A R

Haushalter/Mikkelson Hulburt/Miles/Woolridg

2001 US 2002 US

1994-1996 1981-1994

31 185 M A R

1.1% * 1.2% ***

[-30;-6]

1.8% no

[-1;+1]

2.0% **

[-2;+2]

3.4% ***

[-1;+1]

2.2% **

[-1;+1]

1.9% *** 2.1% *** -0.4% no

Hulburt

2002 US

1981-1994

172 M R A R

[-1;0]

1.6% *** 1.3% *** 3.6% **

V~h

2002 US

1980-1997

336 M A R

[-252;-2]

14.9% ***

M A R / M R t [-1;+1]

1.9% ***

MRAR

[-1;+1]

1.2% ***

MRAR

[-1;+1]

4.9% ***

Mulherin/Boone

2002 US

1990-1999

125 Market ind [-1;+1]

Glatzel

2003 US

1993-1997

30 MRAR/Cl~[-20;-1]

2.3% *** '-0.10% no

[-1;0]

'-0.29% no

[-1;+1]

,/1.13%no

[0;+1]

,/1.35% no

243

Appendix

11" E x i s t i n g s t u d i e s o n E C O a n n o u n c e m e n t

This table summarises r e t u r n s in E C O s

period returns - Non-US

key existing studies on announcement

in G e r m a n y

and in Europe.

period

* * * , ** a n d * i n d i c a t e s

s i g n i f i c a n c e at t h e 1 0 % , 5 % a n d 1 % l e v e l , r e s p e c t i v e l y .

Author

Year Country Period

N Method

Event window

Pellens

1993 D

1984-1991

! 1 MRAR

Hasselmann

1997 D

1992-1994

11 MRAR

Kaserer/Ahlers

2000 D

1984-1997

23 MRAR

Langenbach

2001 D

1984-1999

L6ffler

2001 D

1984-1996

32 MRAR 18 MRAR 19 MRAR

Stienemann

2003 D

1989-2002

49 MRAR

Elsas/L6ffier

2003 D

1984-2000

MRAR

Wagner

2004 D

1980-2002

58 MRAR

[-30;0] [-5;0] [-25;0] [-5;0] [-250;0] [-50;0] [-1;0] [- 1;+ 1] [-1;0] [- 1;0] [-30;+10] [-3;+1] [0] [-10;0] [-1;0] [-1;+1] [-10;+10] [-5;+5] [-1;0] [-30;+10] [-1;+1] [0;+1]

Europe Kaserer Btihner

2002 Be, S, N, ? 2004 Be, S, N, 1991-2001

CAR Significance

Germany

? 66 MAR 66 66

[-1;0] [-20;+1] [-1;+1] [0;+1]

-0.3% -0.5% 4.0% -1.3% 14.7% 5.0% -0.3%

NA NA NA NA ** ** no -0.4% no 1.4% ** 0.7% no 7.1% *** 1.8% no O.9% * 2.9% * 1.6% *** 1.2% * 4.1%* 3.9% *** 1.1% *** 3.6% ** 1.7% *** 2.8% *** 2.4% 4.5% 2.4% 2.0%

? ** ** **

!

o

o

c~

o

op.~ ,,#

! o c~

.c:

~P

o.

~

~

o

.Jl

~

~~

~.

.

245

Appendix 13: Abnormal return on individual days around event This table shows the mean and median abnormal return on individual days in a [-10;+10] window around the event. Significance is assessed by a standard t-test for the mean, and sign and rank tests for the median. P-values below O. 10 are printed in bold.

Days relative to announcement

Mean AR

-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10

0.48% 0.43% -0.25% -0.06% 0.47% 0.07% 0.22% -0.47% -0.20% -0.18% 1.34% -0.17% -0.55% -0.46% -0.17% -0.22% 0.51% 0.16% -0.07% -0.16% 0.21%

p-value Median AR p-value sign

p-value rank

0.1106 0.0837 0.0470 0.8521 0.2929 0.3657 0.3150 0.3667 0.8194 0.6659 0.0005 0.4011 0.0275 0.1597 0.4056 O. 1819 0.2445 0.6918 0.5445 0.8203 0.6114

0.3207 0.3210 0.0571 0.9504 0.4183 0.1942 0.0607 0.7765 0.5019 0.4891 0.0000 0.6346 0.0912 0.0945 0.9753 0.2627 0.1813 0.9686 0.4403 0.8107 0.2404

0.05% 0.02% -0.22% -0.03% -0.10% 0.13% 0.25% -0.04% -0.01% -0.19% 0.57% -0.07% -0.26% -0.40% -0.01% -0.15% 0.11% -0.06% -0.15% -0.04% 0.09%

0.6171 0.9203 0.0357 0.9203 0.7642 0.4839 O. 1936 0.6171 0.9203 0.2713 0.0010 0.3681 0.0214 0.4839 1.0000 O. 1936 0.6171 0.4839 0.3681 0.7642 O. 1936

246

Appendix 14: CARs around event date - [-50;+20] day window; This figure shows the CARs around the event date in a [-50;+20] day window around the event.

247

Appendix 15: A b n o r m a l r e t u r n s per year This table shows mean abnormal returns in each year of the sample period. P-values below 0.10 are printed in bold.

Year 1984 1986 1987 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Total/Average

No. of companies 2 1 1 4 3 2 1 7 6 4 6 l0 7 16 21 2 2 2 3 100

Mean abnormal return 1.40% 0.30% -1.12% 0.69% 1.66% -0.34% 1.16% -0.23% -0.47% 0.31% 1.32% 3.01% 1.37% 0.83% 1.97% 1.32% 5.36% 3.70% 1.21% 1.34%

p-value 0.190 NaN NaN 0.218 0.026 0.760 NaN 0.647 0.808 0.215 0.025 0.061 0.175 0.169 0.034 0.249 0.250 0.280 0.096

248

Appendix 16: Abnormal returns per period This table shows mean abnormal returns in two consecutive periods. The first period begins in 1984 and ends on December 31 of the year indicated in column 1. The second period begins on January 1 of the following year. P-values below 0.10 are printed in bold.

Period" 1984- Mean Period Mean Period 1 2 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

1.40% 1.40% 1.03% 0.50% 0.50% 0.59% 0.88% 0.70% 0.73% 0.41% 0.21% 0.23% 0.40% 0.96% 1.01% 0.97% 1.20% 1.20% 1.29% 1.34%

1.33% 1.33% 1.34% 1.37% 1.37% 1.40% 1.39% 1.43% 1.43% 1.58% 1.75% 1.83% 1.88% 1.67% 1.72% 2.19% 2.71% 3.11% 2.21% 1.21%

Co. in period 1

Co. in period 2

p-value

2 2 3 4 4 8 11 13 14 21 27 31 37 47 54 70 91 93 95 97

98 98 97 96 96 92 89 87 86 79 73 69 63 53 46 30 9 7 5 3

0.511 0.511 0.441 0.316 0.316 0.270 0.328 0.244 0.247 0.090 0.027 0.018 0.022 0.159 0.162 0.058 0.112 0.086 0.288 0.525

.,-~

l

!

r

o

o

o

o

o r

,