Essais En Finance D’entreprise

Ecole des Hautes Etudes en Sciences Sociales
Ann´e 2007 e Num´ro attribu´ par la biblioth`que e e e

————————————

Essais en Finance d’Entreprise

THESE POUR LE DOCTORAT EN SCIENCES ECONOMIQUES DE L’ECOLE DES HAUTES ETUDES EN SCIENCES SOCIALES Pr´sent´e et soutenue publiquement par e e David SRAER

le 29 Juin 2007
Directeur de th`se e M. Jean Tirole Directeur d’Etudes ` l’E.H.E.S.S. a

JURY

M. Denis Gromb, Mme. Ulrike Malmendier , M. Brunos Biais, M. Jacques Cr´mer, e

Professeur, London Business School (Rapporteur) Assistant Professeur, Universit´ de Californie, Berkeley (Rapporteur) e Directeur de Recherche au C.N.R.S., Universit´ de Toulouse 1 e Directeur de Recherche au C.N.R.S., Universit´ de Toulouse 1 e

Remerciements
J’ai ` l’´gard de Jean Tirole un dette incommensurable : pour l’accueil chaleureux dont il m’a fait b´n´ﬁcier ` a e e e a chacune de mes venues ` Toulouse, pour l’incroyable richesse des discussions que nous avons pu avoir tout au long a du d´veloppement de cette th`se, pour le soutien qu’il m’a toujours apport´ et pour l’extraordinaire qualit´ de ses e e e e conseils scientiﬁques, je luis suis inﬁniment gr´. e David Thesmar m’a fait d´couvrir ce qu’´tait r´ellement la recherche en ´conomie et surtout, a su m’ouvrir les e e e e yeux sur ce qu’´tait la bonne recherche en ´conomie. Sans lui, et ` de tr`s nombreux ´gards, cette th`se n’existerait e e a e e e pas. Sa cr´ativit´, son enthousiasme, sa curiosit´ tout comme son amiti´ ont ´t´ un moteur formidable, aussi bien au e e e e ee quotidien que sur le long terme. Ma dette ` son ´gard est immense. J’esp`re profond´ment que notre collaboration a e e e pourra se prolonger aussi longtemps que possible ! Je r´alise pleinement la chance que j’ai pu avoir de travailler, sur les chapitres 2 et 3 de cette th`se, avec e e Augustin Landier. Augustin est certainement le chercheur le plus inventif avec lequel j’ai pu travailler jusqu’ici : avec lui, l’inspiration n’est jamais tr`s loin. . . Je tiens ´galement ` remercier chaleureusement Thomas Chaney pour e e a sa collaboration sur le chapitre 4. En plus d’ˆtre un ami ﬁd`le depuis de nombreuses ann´es maintenant, Thomas e e e est devenu ce professeur d’´conomie brillant avec lequel je suis chanceux de pouvoir travailler : j’esp`re que nous e e n’allons pas nous arrˆter en si bon chemin. . . Philippe F´vrier doit enﬁn ˆtre associ´ ` cette liste de mes co-auteurs e e e ea pr´f´r´s, mˆme si notre travail commun ne ﬁgure pas dans cette th`se. eee e e J’ai ´galement b´n´ﬁci´, tout au long de ces quatre ann´es qu’ont dur´ ma th`se, d’un environnemnt intellectuel e e e e e e e incroyablement favorable ` l’´laboration de tous ces projets de recherche. Pour les innombrables discussions, d´bats, a e e lectures partag´es, controverses, etc. . . que j’ai pu avoir avec eux, je tiens ` remercier Nicolas Coeurdacier, Emmanuel e a Farhi, Francis Kramarz, Pierre Cahuc et Guy Laroque (et plus g´n´ralement, le laboratoire de Macro´conomie du e e e CREST et ses groupes de travail), et Laurent Linnemer (les jeudi au LEI restent parmi les moments les plus agr´ables e de ces ann´es de th`se). e e Je tiens galement ` remercier Denis Gromb et Ulrike Malmendier pour avoir tous deux accept´ tr`s gentiment a e e de participer au jury de cette th`se. e Je souhaite enﬁn associer ` ce travail ma famille, que je n’ai que rarement l’occasion de remercier pour tout a ce qu’elle m’apporte et me permet de r´aliser : mes parents, ´videmment, et mon frˆre Jo¨l ainsi que ma soeur e e e e Judith. Enﬁn, comment exprimer ma gratitude ` C´cile, notamment pour m’avoir support´ pendant ces ann´es o` a e e e u les discussions ´conomiques ont certainement ´t´ plus nombreuses que de raison ? Depuis de nombreuses ann´es e ee e d´j`, son soutien est ma meilleure arme. ea

i

R´sum´ e e
Cette th`se comporte quatre essais en ﬁnance d’entreprise. Le premier chapitre porte sur le lien entre comportee ment de l’entreprise et structure de son actionnariat et de son management. L’attention est plus particuli`rement e port´e sur les diﬀ´rents styles de management qu’implique la pr´sence de la famille du fondateur de l’entreprise e e e dans l’actionnariat ou dans l’´quipe dirigeante. e Nous montrons ensuite comment des m´canismes de gouvernance interne peuvent supplanter les dispositifs e traditionnels de gouvernement de l’entreprise pour exercer une discipline eﬃcace sur les dirigeants de l’entreprise. Cette ´tude empirique, men´e sur un large panel d’entreprises am´ricaines, est soutenue par une analyse th´orique e e e e qui s’int´resse plus g´n´ralement au rˆle de l’ind´pendance des pr´f´rences au sein des organisations. e e e o e ee Le dernier chapitre de cette th`se vise ` comprendre empiriquement les liens entre valeur du collat´ral d´tenu e a e e par les entreprises et politique d’investissement. L’analyse se concentre particuli`rement sur les actifs immobiliers e que poss`dent les grandes entreprises am´ricaines. e e

Discipline : Sciences Economiques (05). Mots-cl´s : e
Finance d’Entreprise ; Structure du Capital ; Styles de Management ; Gouvernement de l’Entreprise ; Th´orie e des Organisations ; Ind´pendance des Pr´f´rences ; Contraintes de Cr´dit ; Politique d’Investissement ; Collat´ral. e ee e e

Intitul´ et adresse du laboratoire : e
Centre de Recherche en Economie et Statistiques (C.R.E.S.T.) Timbre J301, 15 Bd Gabriel P´ri e 92245, Malakoﬀ Cedex, France

Contents
Remerciements Introduction 1 Performance and Behavior of Family Firms: Evidence from the French Stock Market 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Family Business on the French Stock Market . . . . . 1.2.2 Do Family Firms Diﬀer from Other Firms? . . . . . . 1.3 Multivariate Evidence . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Empirical Strategy . . . . . . . . . . . . . . . . . . . . 1.3.2 Family Firms Outperform widely held Firms . . . . . 1.3.3 Discussion on Endogeneity Biases . . . . . . . . . . . . 1.4 Management Styles in Family Firms . . . . . . . . . . . . . . 1.4.1 Breaking down Corporate Performance . . . . . . . . . 1.4.2 Family Firms Pay Lower Wages . . . . . . . . . . . . . 1.4.3 Descendants Can Commit on Long Term Employment 1.4.4 Outside CEOs Are More Financially Literate . . . . . 1.5 Summary and Leads for Future Research . . . . . . . . . . . 2 Bottom-Up Corporate Governance 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 2.2 Data and Measurement Issues . . . . . . . . . . . . 2.2.1 Datasets . . . . . . . . . . . . . . . . . . . . 2.2.2 Constructing an Internal Governance Index 2.3 Internal Governance and Corporate Performance . 2.3.1 Basic Results . . . . . . . . . . . . . . . . . 2.3.2 Robustness Checks and Causality . . . . . . 2.4 Internal Governance and Acquisitions . . . . . . . 2.5 External Versus Internal Governance . . . . . . . . 2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . 3 Optimal Independence in Organizations 3.1 Introduction . . . . . . . . . . . . . . . . . . . . 3.2 The Model . . . . . . . . . . . . . . . . . . . . 3.2.1 Set-Up . . . . . . . . . . . . . . . . . . . 3.2.2 Equilibrium Concept . . . . . . . . . . . 3.3 Organizational Homogeneity aﬀects Reactivity 3.3.1 Implementer’s Eﬀort Choice . . . . . . 3.3.2 The Decision Maker’s Project Choice . . 3.3.3 Summary and Discussion . . . . . . . . i v

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . .

1 1 4 5 7 9 9 10 14 16 16 18 21 26 35 37 37 39 39 43 45 45 49 54 58 63 65 65 68 68 71 72 72 73 80

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . iii

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

Introduction G´n´rale e e

iv

3.4

3.5

3.6

3.3.4 Organizational Design . . . . . . . . . . . . . . . . . . . . . Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Delegation of Organizational Design to the Decision Maker 3.4.2 Allowing for Financial Incentives . . . . . . . . . . . . . . . Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Bottom-Up Corporate Governance . . . . . . . . . . . . . . 3.5.2 Public Administration and Ideological Bias . . . . . . . . . 3.5.3 Organizing for Change . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

82 85 85 88 89 89 91 92 95

4 The Corporate Wealth Eﬀect: From Real Estate Shocks to Corporate Investment 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Accounting and Governance Data . . . . . . 4.2.2 Real Estate Data . . . . . . . . . . . . . . . . 4.2.3 Loan Contracts . . . . . . . . . . . . . . . . . 4.3 Real Estate Prices and Investment . . . . . . . . . . 4.3.1 Empirical Strategy . . . . . . . . . . . . . . . 4.3.2 Main Results . . . . . . . . . . . . . . . . . . 4.3.3 City level results . . . . . . . . . . . . . . . . 4.4 Collateral and Debt . . . . . . . . . . . . . . . . . . 4.4.1 Debt Issuance . . . . . . . . . . . . . . . . . . 4.4.2 Debt Contracts . . . . . . . . . . . . . . . . . 4.5 Corporate Governance and Investment Performance 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . A Performance and Behavior of Family Firms A.1 Data . . . . . . . . . . . . . . . . . . . . . . . A.1.1 Corporate Accounts . . . . . . . . . . A.1.2 Family Ownership and Management . A.1.3 Employment Data . . . . . . . . . . . A.1.4 Stock Prices . . . . . . . . . . . . . . . A.1.5 Acquisitions . . . . . . . . . . . . . . . A.2 Additional Tables . . . . . . . . . . . . . . . . A.3 Breaking down Performance . . . . . . . . . . B Optimal Independence in Organizations B.1 Proofs . . . . . . . . . . . . . . . . . . . B.1.1 Proof of Proposition 1 . . . . . . B.1.2 Proof of Proposition 2 . . . . . . B.1.3 Proof of Proposition 3 . . . . . . B.1.4 Proof of Proposition 4 . . . . . . B.1.5 Proof of Proposition 5 . . . . . . B.1.6 Proof of Proposition 6 . . . . . . B.2 Proof of Proposition 7 . . . . . . . . . . B.3 The True Role of Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 97 100 101 107 109 109 109 112 116 122 122 124 126 129 1 1 1 2 3 4 5 6 8 9 9 9 10 12 13 14 15 16 17 18 18

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

C The Corporate Wealth Eﬀect C.1 Construction of the pt Landi Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction G´n´rale e e

v

Introduction
La r´volution de l’“´conomie de l’information”, d`s la ﬁn des ann´es 70, a permis un profond renouveau de la e e e e litt´rature de ﬁnance d’entreprise. La prise en compte des asym´tries d’information entre investisseurs et entrepree e neurs a mis ﬁn ` un monde dans lequel le ﬁnancement de l’entreprise ´tait un sujet “non pertinent” (Modigliani a e et Miller (1958)) : d´sormais, l’organisation de l’entreprise, la structure et la nature de son actionnariat, le fonce tionnement de son conseil d’administration ou la sp´ciﬁcit´ de ses actifs devenaient autant de facteurs susceptibles e e d’aﬀecter le montant des fonds disponibles pour le ﬁnancement de nouveaux projets et donc d’inﬂuer sur son comportement dans l’´conomie r´elle. Cette th`se s’inscrit dans le prolongement de cette ´volution de la ﬁnance e e e e d’entreprise. Elle s’int´resse en particulier ` trois questions fondamentales qui trouvent leur origine dans cette e a r´volution de l’“´conomie de l’information”, mais auxquelles seules des r´ponses incompl`tes ont ´t´ apport´es e e e e ee e jusqu’` pr´sent. a e

Performance et Comportement des Entreprises Familiales en France
Dans le chapitre 11 de cette th`se, je commence par explorer le lien entre la nature de l’actionnariat, l’identit´ e e des dirigeants et le comportement de l’entreprise. Alors que depuis Berle et Means, les ´conomistes ont port´ e e principalement leur attention sur les grandes entreprises cot´es ` l’actionnariat diﬀus, il se trouve que la plupart des e a entreprises autour du monde ont un actionnaire dominant, qui se trouve ˆtre tr`s souvent la famille du fondateur e e de l’entreprise (La Porta, Lopez-de-Silanez et Shleifer (1999)). En outre, les familles des fondateurs se trouvent souvent ˆtre impliqu´es dans le management quotidien des entreprises. Ainsi, dans l’´chantillon construit pour ce e e e premier chapitre (qui rassemble des donn´es sur l’int´gralit´ des entreprises cot´es ` la Bourse de Paris entre 1994 et e e e e a 2000), plus de 60% des ﬁrmes sont encores dirig´es par la famille de leur fondateur. Mˆme au sein des plus grandes e e entreprises cot´es am´ricaines, Anderson et Reeb (2003) montrent que pr`s de 16% des entreprises du S&P500 sont e e e encore aux mains de leur fondateur ou de leurs h´ritiers. Ainsi, les pr´mices du mod`le de l’entreprise selon Berle e e e
1 Le chapitre 1 est tir´ d’un article ´crit en collaboration avec David Thesmar : “Performance and Behavior of Family Firms: Evidence e e from the French Stock Market”, ` paraitre dans le Journal of the European Economic Association. a

Introduction G´n´rale e e

vi

et Means o` (1) le dirigeant ne poss`de pas l’entreprise et (2) l’actionnariat est largement dispers´ ne semble pas u e e pouvoir s’appliquer ` la plupart des entreprises. a Une vision plus pertinente du capitalisme mondial serait donc plutˆt qu’une grande entreprise cot´e est typiqueo e ment d´tenue, et fr´quemment dirig´e, par une famille. Cette nouvelle perspective appelle donc un nouvel agenda de e e e recherche sur les caract´ristiques sp´ciﬁques du management et du contrˆle dans un cadre dynastique. Le premier e e o chapitre de cette th`se cherche donc ` aborder, parmi d’autres, les questions suivantes : les familles maximisent elles e a le proﬁt de leur entreprise ? Sont-elles plus promptes ` construire des empires au d´triment des actionnaires minoria e taires ? Sont-elles trop prudentes, trop conservatrices ? Au contraire, b´n´ﬁcient-elles d’un horizon d’investissement e e de plus long terme, qui leur permet d’´viter de succomber aux ph´nom`nes de mode ? Plus g´n´ralement, se e e e e e comportent-elles de fa¸on diﬀ´rente de ces grandes ﬁrmes ` l’actionnariat diﬀus que les acad´miques connaissent si c e a e bien ? Pour r´pondre ` ces questions, notre article s’appuie sur le cas des entreprises familiales cot´es sur la Bourse de e a e Paris. L’exemple fran¸ais nous parat ˆtre un cas int´ressant ` deux titres. En premier lieu, la France est un pays c e e a d’Europe continental, si bien que ces institutions ﬁnanci`res et son histoire diﬀ`rent fortement des pays anglo-saxons, e e dans lesquels la plupart des ´tudes sur les entreprises familiales ont ´t´ conduites jusqu’` pr´sent ; en particulier, les e ee a e entreprises familiales y repr´sentent une part bien plus importante de l’´conomie dans son ensemble. En second lieu, e e contrairement ` de nombreux autres pays d’Europe continental, la France poss`de un nombre important de grandes a e entreprises ` l’actionnariat diﬀus, qui tendent ` ressembler aux entreprises manag´riales observ´es aux Etats-Unis – a a e e sans actionnaire dominant et avec un management prot´g´. Cela nous permet d’avoir acc`s ` un groupe de contrˆle e e e a o suﬃsament large auquel comparer les entreprises familiales. Nous avons ainsi collect´ des donn´es sur pr`s de 1,000 entreprises cot´es ` la Bourse de Paris sur la p´riode e e e e a e 1994-2000. Notre panel dispose d’informations d´taill´es sur les entreprises (l’emploi, les comptes sociaux, les e e acquisitions eﬀectu´es, les rendements boursiers principalement) ainsi que sur la famille fondatrice (notamment sa e participation ` l’actionnariat et au management). Dans un premier temps, nous nous int´ressons ` la corr´lation a e a e entre performance de l’entreprise et pr´sence d’une famille ` l’actionnariat et/ou au management. Ainsi, en se e a concentrant sur les performances comptables, nous mettons en ´vidence que la performance des ﬁrmes familiales est e signiﬁcativement sup´rieure ` celles des entreprises non familiales. De fa¸on coh´rente avec la litt´rature existante e a c e e sur l’“eﬀet fondateur” (Adams, Almeida et Ferreira (2005), Fahlenbrach (2005)), nous trouvons que les fondateurs sont ` la tˆte d’entreprises tr`s proﬁtables. Egalement en ligne avec les r´sultats ´tablis sur donn´es am´ricaines a e e e e e e (Anderson et Reeb (2003), Amit et Villalonga (2006)), il apparait que les entreprises familiales dirig´es par des e

Introduction G´n´rale e e

vii

managers “ind´pendants” de la famille sont bien plus performantes que les entreprises ` l’actionnariat dispers´. Il e a e semble donc qu’aussi bien en France qu’aux Etats-Unis, les familles en tant qu’actionnaires concentr´s permettent, e en moyenne, d’am´liorer les performances de leur entreprise. De fa¸on bien plus surprenante, notre ´tude montre que e c e les managers h´ritiers du fondateur de l’entreprise tendent ´galement ` sur-performer non seulement les entreprises e e a non familiales, mais ´galement les entreprises familiales dirig´es par des P.D.G. ind´pendants. Ainsi, mˆme en e e e e mettant de cˆt´ les fondateurs, les entreprises familiales (qu’elles soient dirig´es ou non par des h´ritiers) ont une oe e e performance sup´rieure aux entreprises non familiales en France. Une interpr´tation causale de ces r´sultats “en e e e coupe” est cependant probl´matique puisqu’il est possible que seules les entreprises les plus performantes soient e transmises ` des h´ritiers. Une solution possible consisterait ` s’int´resser explicitement aux transmissions (comme a e a e dans P´rez-Gonz´lez (2006)). Bien que nous observions trop peu de transitions dans notre ´chantillon pour pouvoir e a e r´aliser des tests statistiques puissants, nous voyons n´anmoins que (1) les h´ritiers n’h´ritent typiquement pas des e e e e meilleures entreprises et que (2) les h´ritiers dont les entreprises sortent de la Bourse de Paris ne sous-performent e pas systematiquement les autres cat´gories d’entreprises. Il est donc probable que les biais d’endog´n´it´ ne soient e e e e pas trop importants dans notre ´chantillon. e Nous essayons ensuite d’interpr´ter, ` l’aide des donn´es, ces diﬀ´rences de performance, en consid´rant suce a e e e cessivement les diﬀ´rentes caract´ristiques d´terminant ces performances. Une plus grande productivit´ du travail e e e e semble ˆtre l’explication la plus signiﬁcative pour la performance sup´rieure des entreprises dirig´es par leur fone e e dateur. Quant aux diﬀ´rences de performance entre les entreprises familiales dirig´es par des P.D.G. professionnels e e et celles dirig´es par des h´ritiers, nous regardons successivement les diﬀ´rences en terme de gestion de la main e e e d’oeuvre et gestion du capital. Tout d’abord, il apparait que les entreprises familiales dirig´es par des professionnels et des h´ritiers versent, en e e moyenne, des salaires plus faibles. N´anmoins, ce r´sultat peut n’ˆtre que le reﬂˆt de compositions diﬀ´rentes de la e e e e e force de travail dans ces entreprises familiales, une plus grande fraction de travailleurs non qualiﬁ´s par exemple. e Aﬁn de prendre en compte une telle possibilit´, nous apparions nos donn´es d’entreprises avec des ﬁchiers adminise e tratifs employeurs/employ´s, ce qui nous permet de contrˆler pour les eﬀets de composition en terme d’exp´rience, e o e d’anciennet´ et de qualiﬁcation. Cela nous permet de mettre en ´vidence que les P.D.G. professionnels ne payent des e e salaires plus faibles que parce qu’ils emploient des salari´s moins qualiﬁ´s. Au contraire, mˆme apr`s avoir contrˆler e e e e o pour la structure de qualiﬁcation des employ´s, nous continuons ` trouver que les h´ritiers versent des salaires plus e a e faibles. En outre, nous montrons que la demande de travail dans les entreprises familiales dirig´es par des h´ritiers e e r´pond signiﬁcativement moins aux chocs sectoriels de vente, soit, en d’autres termes, que les dirigeants h´ritiers e e

Introduction G´n´rale e e

viii

lissent les eﬀectifs de leur entreprise le long du cycle des aﬀaires. Au total, ces r´sultats peuvent s’interpr´ter comme e e la preuve que les entreprises familiales dirig´es par des h´ritiers fournissent ` leurs employ´s des contrats implicites e e a e d’assurance sur le long terme. Ces contrats implicites, cr´dibles uniquement en raison de l’horizon d’investissement e plus long des entrepreneurs familiaux, permettent aux h´ritiers de verser des salaires plus faibles pour des qualie ﬁcations plus ´lev´es. En eﬀet, en comparaison des dirigeants professionnels, les h´ritiers d´gagent bien une plus e e e e grande productivit´ du travail. e Pour conclure ce chapitre 1, nous nous int´ressons aux diﬀ´rences observables, entre les diﬀ´rentes cat´gories e e e e d’entreprises, en terme de gestion du capital. Les dirigeants professionnels semblent assujettis ` des taux d’int´rˆt a ee plus faibles sur leur dette ﬁnanci`re et ils ont tendance ` op´rer avec des ratios capital/travail plus faibles. En e a e comparaison des h´ritiers, ou mˆme des entreprises ` l’actionnariat dispers´, les grandes acquisitions r´alis´es par e e a e e e les dirigeants professionnels des entreprises familiales d´truisent moins de valeurs dans le long terme. Bien que e totalement a-structurelle et donc ` interpr´ter avec pr´caution, l’image g´n´rale qui se d´gage de ces r´sultats a e e e e e e semblent indiquer que (1) les dirigeants professionnels apportent une expertise ﬁnanci`re ` l’actionnariat familial, e a en particulier, qu’ils sont moins eﬃcaces dans l’utilisation du capital productif alors que (2) les h´ritiers ont l’horizon e managerial n´cessaire pour pouvoir s’engager ` des politiques protectrices de gestion de la main d’oeuvre, ce qui e a leur permet de b´n´ﬁcier d’une plus forte productivit´ du travail. e e e Ce chapitre repr´sente donc une contribution ` la litt´rature ´mergente sur les entreprises familiales. La majeure e a e e partie de cette litt´rature s’est, jusqu’` pr´sent, concentr´e sur les entreprises nord am´ricaines et des comparaisons e a e e e de proﬁtabilit´. L’ensemble de ces contributions (Anderson et Reeb (2003), Amit et Villalonga (2006), and P´reze e Gonz´lez (1999) pour les Etats-Unis ; Morck, Strangeland and Yeung (2002) pour le Canada), a permis d’arriver ` a a un consensus sur la plus grande performance des entreprises familiales dirig´es par des fondateurs ou des dirigeants e professionnels. Ces r´sultats sont g´n´ralement interpr´t´s comme des preuves qu’un actionnaire concentr´ et de long e e e ee e terme permet d’´viter les coˆts associ´s ` l’expropriation potentielle des actionnaires minoritaires. En revanche, la e u e a qualit´ manageriale des h´ritiers est un sujet bien plus controvers´. Deux ´tudes “en coupe” portant sur des grandes e e e e entreprises am´ricaines cot´es dans les ann´es 90 (Anderson and Reeb (2003) and Amit and Villalonga (2006)) e e e trouvent des r´sultats oppos´s. N´anmoins, l’approche en “diﬀ´rence de diﬀ´rence” adopt´e par P´rez-Gonz´lez e e e e e e e a (2006) sugg`re que les h´ritiers sont de plus mauvais dirigeants que les managers professionnels, au moins au sens e e de la proﬁtabilit´. Ce r´sultat est conﬁrm´ par Bennedsen, Nielsen, P´rez-Gonz´lez et Wolfenzon (2006) qui tirent e e e e a proﬁt de la richesse des donn´es danoises pour traiter de fa¸on pr´cautionneuse la question de la causalit´. Le premier e c e e chapitre de cette th`se permet de compl´ter ces ´tudes pour un grand pays d’Europe continentale: nos r´sultats e e e e

Introduction G´n´rale e e

ix

montrent, au contraire des pr´c´dentes ´tudes, que les entreprises familiales dirig´es par des h´ritiers pr´sentent e e e e e e une rentabilit´ sup´rieure, non seulement relativement aux entreprises ` l’actionnariat dispers´, mais ´galement e e a e e relativement aux entreprises familiales dirig´es par des managers professionnels. Cependant, il convient d’insister e qu’en plus de l’eﬀet causal du management familial (que nous cherchons ` mettre en lumi`re), des biais de s´lection a e e et de simultan´it´ pourraient ´galement expliquer cette corr´lation. e e e e De fa¸on plus int´ressante, notre papier compl´mente la litt´rature existante sur les entreprises familiales en c e e e s’int´ressant aux eﬀets d’un actionnariat et/ou d’un management familial sur d’autres dimensions du comportement e des entreprises. Notre r´sultat selon lequel les entreprises familiales versent des salaires plus faibles est, ` notre e a connaissance, une nouveaut´ pour la litt´rature. Ce r´sultat peut cependant ˆtre reli´ ` certaines ´tudes sur le lien e e e e ea e entre politiques salariales d’une part et s´paration de l’actionnariat et du management dans l’entreprise d’autre part. e Ainsi, une litt´rature vieille d’une dizaine d’ann´es, r´cemment r´sum´e et d´velopp´e par Bertrand et Mullainathan e e e e e e e (1999) montre que les probl`mes d’agence entre management et actionnariat peuvent eﬀectivement conduire ` des e a niveaux de salaire plus ´lev´s, pour tous les agents de l’organisation. L’autre nouveaut´ oﬀerte par ce chapitre 1 e e e repose sur l’analyse des diﬀ´rences de “styles de management” entre les managers professionnels et les h´ritiers : e e nos donn´es sugg`rent que les managers professionnels apportent une certaine expertise ﬁnanci`re, qui permet une e e e gestion plus eﬃcace du capital productif, alors que les h´ritiers, b´n´ﬁciant de leur horizon de long-terme, peuvent e e e mettre en place des politiques d’emploi “protectrices” qui leur permettent de n´gocier des salaires plus faibles. On e peut interpr´ter ces r´sultats ` l’aune de l’article de Bertrand et Schoar (2003) qui met en ´vidence, parmi les P.D.G. e e a e des grandes entreprises am´ricaines, de fortes diﬀ´rences en termes de politique d’investissement, d’acquisition ou e e de politique ﬁnanci`re. En particulier, Betrand et Schoar montrent que les d´tenteurs de M.B.A. tendent ` ˆtre e e a e plus agressifs en termes de levier ﬁnancier et de politique d’acquisition. Nos propres r´sultats sugg`rent que le e e management familial peut ˆtre une nouvelle dimension d’h´t´rog´n´it´ des styles de management, passionnante ` e ee e e e a explorer.

Gouvernance Interne et Performance des Entreprises
Le chapitre 22 de cette th`se plonge dans le d´bat sur le gouvernement de l’entreprise. Les acad´miques tout autant e e e que les praticiens ont admis depuis longtemps l’id´e qu’en l’absence d’une surveillance rapproch´e, les P.D.G. des e e grandes entreprises cot´es pouvaient prendre des d´cisions allant ` l’encontre de l’int´rˆt de leurs actionnaires : e e a ee en utilisant les ressources de l’entreprise pour des projets “personnels” destructeurs de valeur, en construisant des
2 Le chapitre 2 est tir´ d’un article ´crit en collaboration avec Augustin Landier et David Thesmar : “Bottom-Up Corporate Govere e nance”, Document de Travail Stern School of Business, N06-007.

Introduction G´n´rale e e

x

empires non rentables, en empˆchant l’acquisition de la ﬁrme, ou mˆme, dans certains cas rares mais largement e e m´diatis´s, en s’engageant dans des op´rations frauduleuses de maquillage des comptes ou de d´tournement des e e e e actifs de l’entreprise. Pour mettre en place des contre-pouvoirs au P.D.G., le consensus d´gag´ aupr`s des praticiens e e e et des r´gulateurs consiste ` se reposer sur un conseil d’administration fort et ind´pendant du management. Dans de e a e nombreux pays, des codes informels de gouvernement d’entreprise ont ainsi recommand´, depuis plus d’une dizaine e d’ann´es, la nomination d’administrateurs ind´pendants.3 Aux Etats-Unis, la r´cente vague de scandales ﬁnanciers e e e a conduit a une importante r´ponse de la r´gulation, qui a rendu la nomination d’administrateurs ind´pendants e e e obligatoire pour les entreprises cot´es sur les principales places boursi`res. e e Il est vrai que la recherche acad´mique a d´montr´ depuis la ﬁn des ann´es 80 que les conseils d’administrations e e e e sont un outil eﬃcace de gouvernement de l’entreprise. Les conseils d’administration “ind´pendants” semblent prene dre plus en compte la performance de l’entreprise lorsqu’ils doivent d´cider des r´mun´rations ou des licenciements e e e des managers (Weisbach (1988), Dahya, Mc Connel and Travlos (2002)). Le march´ salue g´n´ralement la nomie e e nation d’administrateurs ind´pendants avec des rendements anormaux positifs (Rosenstein and Wyatt (1990)). Il e n’y a cependant aucune ´vidence que des conseils d’administration ind´pendants permettent une am´lioration de la e e e proﬁtabilit´ ou de la valeur de l’entreprise.4 Une possibilit´, au moins pour les grandes entreprises cot´es, est que e e e les conseils d’administration ind´pendants, bien qu’ils puissent ˆtre extrˆmement pr´cieux en temps de crise, sont e e e e bien trop ´loign´s de la gestion quotidienne des op´rations de l’entreprise pour cr´er beaucoup de valeur. e e e e En cons´quence, les chercheurs int´ress´s au gouvernement de l’entreprise ont d´cid´ de ne plus se concentrer e e e e e que sur les seules variables “organisationnelles”, comme la composition du conseil d’administration, et d’examiner d’autres dimensions de la gouvernance, essentiellement ` travers les statuts de l’entreprise. Les principales d´couvertes a e de cette litt´rature r´cente tendent ` prouver que des provisions favorables aux investisseurs inscrites dans les statuts e e a de la ﬁrme augmentent la valeur des actifs de l’entreprise, en les rendant plus vuln´rables aux acquisitions (Gompers, e Ishii and Metrick (2003), Cremers, Nair and John (2005),Bebchuk and Cohen (2004)). Ce chapitre de ma th`se propose d’´tudier une nouvelle mesure de la qualit´ de la gouvernance d’une entree e e prise, fond´e sur une information purement “organisationnelle”. Notre intuition est qu’il est possible de r´cup´rer e e e
3 D’ailleurs, de nombreuses entreprises ont ´t´ ravis de se soumettre ` ces recommandations. Par exemple, le Cadbury Report ´mis en e e a e 1992 en Grande Bretagne recommande que “la majorit´ des administrateurs non cadres de l’entreprise devraient ˆtre ind´pendant”. Le e e e rapport Vi´not, fait en France en 1998, propose que “les administrateurs ind´pendant devraient repr´senter au moins un tiers du conseil e e e d’administration”. Le suivi de ces recommandations n’´taient pas obligatoires, mais fut pourtant largement r´pandu. Ainsi, en 1996, e e plus de 50% des entreprise anglaises examin´s par Dahya, Mc Connel et Travlos (2002) d´clarait s’ˆtre soumis aux recommandations du e e e Cadbury Report. 4 En r´alit´, la corr´lation semble mˆme ˆtre l´g`rement n´gative. Une raison probable pour ce r´sultat est que les ﬁrmes en diﬃcult´ e e e e e e e e e e ont tendance ` engager plus d’administrateurs ind´pendant (Kaplan et Minton (1994)). En ﬁltrant ce biais d’endog´n´it´, il ne semble a e e e e pas y avoir de corr´lation syst´matique entre la proﬁtabilit´ de l’entreprise et l’ind´pendance de son conseil d’administration (Baghat e e e e et Black (2003) et Hermalin et Weisbach (2003)).

Introduction G´n´rale e e

xi

de l’information sur le bon fonctionnement d’une entreprise en examinant la composition de l’´quipe dirigeante. e Apr`s tout, les P.D.G. doivent faire face ` leurs directeurs quotidiennement, alors qu’ils ne rencontrent leur conseil e a d’administration que quelques fois par an. Plus pr´cis´ment, nous d´veloppons une mesure de “gouvernance interne” qui capture le degr´ d’“ind´pendance” e e e e e des cades dirigeants vis ` vis du P.D.G.. Utilisant un panel d’entreprises cot´es am´ricaines, nous calculons la fraction a e e de cadres dirigeants qui ont rejoint la ﬁrme avant que le P.D.G. ne soit nomm´ ` son poste. Notre hypoth`se de ea e travail est que ces cadres dirigeants sont “ind´pendants” du P.D.G. de l’entreprise. En eﬀet, comme les P.D.G. e sont tr`s souvent impliqu´s dans le recrutement de leur cadre dirigeant, il est probable que les ex´cutifs engag´s par e e e e un P.D.G. partagent les mˆmes pr´f´rences que ce P.D.G. et/ou qu’ils aient des incitations ` “retourner la faveur”. e ee a Parall`lement, il est probable que les ex´cutifs qui ont connu un autre leadership dans l’entreprise ne soient pas prˆts e e e a ` consid´rer tous les ordres comme l´gitimes, simplement car ils viennent d’un individu hi´rarchiquement sup´rieur. e e e e Notre ´tude fournit tout d’abord des preuves qu’une bonne gouvernance interne (i.e. des ex´cutifs ind´pendants) e e e pr´dit une plus forte proﬁtabilit´ future de l’entreprise, et ce en utilisant diﬀ´rentes mesures de proﬁtabilit´. Par e e e e ailleurs, de mauvaises performances ne conduisent pas ` une diminution de la gouvernance interne, ce qui sugg`re a e un eﬀet causal de la gouvernance interne sur la performance. Nos r´sultats restent similaires mˆme lorsque nous e e contrˆlons pour des mesures traditionnelles de gouvernance “externe”. Nous montrons ´galement que ces r´sultats o e e ne sont pas simplement la cons´quence de d´parts de cadres dirigeants anticipant de mauvaises performances ` venir. e e a La deuxi`me partie de ce chapitre 2 concerne l’impact de la gouvernance interne sur la qualit´ des acquisitions e e r´alis´s par les entreprises : une plus faible fraction de cadres dirigeants ind´pendants est associ´e ` des rendements e e e e a plus faibles pour les actionnaires de l’entreprise, suite ` une grande acquisition. Il est important de remarquer a n´anmoins que les indices standards de gouvernance externe ne sont pas, quant ` eux, corr´l´s aux pertes r´alis´es e a ee e e par les actionnaires d’une entreprise suite ` une acquisition. Ainsi, le conseil d’administration, la pression des a acquisitions hostiles ou le contenu des statuts de l’entreprise semblent ˆtre moins eﬃcaces pour ´viter de mauvaises e e acquisitions que la pression exerc´e par une ´quipe dirigeante ind´pendante. e e e Une contribution importante de ce chapitre de ma th`se est d’exhiber une variable “organisationnelle”, d´ﬁnie e e au niveau de la ﬁrme, poss´dant un pouvoir pr´dictif puissant sur les performances futures. Une interpr´tation e e e alternative consisterait ` dire que notre indice de gouvernance interne n’est qu’une mesure de l’´tendu du pouvoir a e d’un P.D.G. sur sa ﬁrme : les “P.D.G. puissants” pourraient ˆtre ceux les plus susceptibles de r´aliser des acquisitions e e ineﬃcaces et de remplacer les cadres dirigeants sans qu’il n’y ait un lien clair entre ces deux comportements. N´anmoins, nous prouvons que la gouvernance interne n’est que tr`s faiblement corr´l´e aux mesures traditionnelles e e ee

Introduction G´n´rale e e

xii

du “pouvoir des P.D.G.” (comme, par exemple, le fait que le P.D.G. occupe la place de pr´sident du conseil e d’administration). Une autre interpr´tation de nos r´sultats, d´velopp´e dans le chapitre 3, revient ` dire que e e e e a des cadres dirigeants ind´pendants peuvent agir comme un m´chanisme de “gouvernance par le bas”, rendant plus e e coˆteux au P.D.G. la prise de mauvaises d´cisions. u e Deux implications normatives pour les praticiens m´ritent d’ˆtre retenues de cette ´tude. Tout d’abord, notre e e e analyse statistique montre que l’intensit´ d’une telle “gouvernance interne” peut ˆtre, au moins partiellement, e e observ´e et pourrait donc ˆtre int´gr´e dans les diﬀ´rents indices de qualit´ de gouvernance d’entreprise. Cette e e e e e e remarque est ind´pendante de l’interpr´tation que l’on voudra bien faire des r´sultats: qu’elle ne soit qu’un signal e e e que les cadres dirigeants anticipent les mauvais r´sultats ou bien qu’elle mesure l’´tendu de la discipline impos´e par e e e des cadres ind´pendants, notre mesure pr´dit fortement la performance future. En revanche, la seconde implication e e que nous souhaitons mettre en lumi`re repose, elle, sur notre interpr´tation favorite, celle d’une “gouvernance par le e e bas” : en plus de surveiller le management et de le conseiller, un rˆle cl´ du conseil d’administration devrait consister o e a ` mettre en place un ´quilibre des pouvoirs au sein de l’entreprise. Dit autrement, le rˆle du conseil d’administration e o en tant que g´rant des ressources humaines ne devrait pas ˆtre limit´ au seul probl`me de succession du P.D.G.. e e e e

Ind´pendance Optimale au sein des Organisations e
Le chapitre 35 de cette th`se est inspir´ des r´sultats mis en avant dans le chapitre 2. Dans ce chapitre, nous e e e essayons de comprendre le rˆle jou´ par l’ind´pendance des pr´f´rences entre un d´cideur (par exemple le P.D.G. o e e ee e d’une entreprise) et la personne en charge de l’impl´mentation des d´cisions (par exemple un cadre dirigeant de e e l’entreprise). En eﬀet, si un rˆle cl´ des managers au sein d’une organisation consiste ` d´cider de la nature o e a e des projets ` mettre en oeuvre, l’impl´mentation des projets n’est que rarement eﬀectu´e par les managers qui a e e les ont s´lectionn´s. Cette “s´paration de l’impl´mentation et du contrˆle” n’est pas sans cons´quence sur le e e e e o e processus de prise de d´cision. Les “impl´menteurs” peuvent avoir des pr´f´rences intrins`ques sur certains projets e e ee e relativement ` d’autres ou ils peuvent simplement ne pas adh´rer ` la vision manag´riale de l’entreprise. Cette a e a e propension ` ne pas vouloir eﬀectuer les projets s´lectionn´s ne se manifeste paas n´cessairement par un conﬂit a e e e ouvert entre un “impl´menteur” et le d´cideur (ie entre le P.D.G. et le cadre dirigeant de l’entreprise): elle peut e e tout aussi bien conduire ` une sous-provision d’eﬀort au cours de l’impl´mentation du projet. Notre ´tude explore a e e th´oriquement l’existence d’une telle “contrainte d’impl´menentation” et cherche ` comprendre son impact sur e e a l’eﬃcacit´ de l’organisation. e
5 Le chapitre 3 est tir´ d’un article ´crit en collaboration avec Augustin Landier et David Thesmar : “Optimal Independence in e e Organizations”, Document de Travail Stern School of Business, N06-008.

Introduction G´n´rale e e

xiii

Le fait que les d´cideurs doivent internaliser les pr´f´rences des impl´menteurs est largement reconnu dans la e ee e litt´rature de management. C’est, par ailleurs, un des messages principaux de l’autobiographie de Sloan (1963), e “Mes Ann´es chez General Motors”. Dans le chapitre 5 de ce livre, Sloan raconte l’histoire de la “machine ` e a refroidissement ` cuivre”, un projet qui avait suscit´ l’enthousiasme des managers de GM mais n’avait pas r´ussi ` a e e a entrainer l’adh´sion des ing´nieurs en charge de l’impl´mentation du projet. Leur manque de motivation lors de la e e e mise en oeuvre mena ` un echec tr`s coˆteux pour l’entreprise. Sloan d´taille sa propre analyse de la situation en a e u e 1923, en plein coeur de la crise: “Nous comprenons bien que [...] forcer les divisions ` entreprendre un projet auquel a elle ne croit pas ne nous m`ne nulle part. Nous avons essay´ et nous avons ´chou´”. e e e e De fa¸on surprenante, le rˆle des “impl´menteurs” comme contrainte ` la prise de d´cision n’a pas encore ´t´ c o e a e ee explor´ dans la th´orie des organisations. Bien ´videmment, l’id´e que les managers et leurs subordonn´es puissent e e e e e avoir des pr´f´rences conﬂictuelles n’est certainement pas nouvelle. Un nombre important de travaux de recherche ee s’est concentr´ sur l’analyse de situations o` les impl´menteurs ont de l’information priv´ sur l’eﬀort qu’ils doivent e u e e faire pour exercer une tˆche sp´ciﬁque (Calvo et Wellisz (1979)). Une autre partie de la litt´rature s’est interess´e ` a e e e a des probl`mes de prise de d´cision dans un cadre principal-agent o` l’agent a de l’information priv´ sur la “bonne” e e u e d´cision ` prendre (Simon (1957) et plus r´cemment Aghion et Tirole (1997) ou Dessein (2002)). Enﬁn, une e a e derni`re partie de la litt´rature a cherch´ ` mettre en place des m´chanismes permettant de r´duire directement e e e a e e ces divergences dans les pr´f´rences (par exemple en d´ﬁnissant une strat´gie d’entreprise simple, comme dans ee e e Rotemberg et Saloner (1994) ou une vision manageriale claire comme dans Van den Steen (2004)). Toutefois, qu’elle s’int´resse ` des probl`mes de prises de d´cision ou d’impl´mentations, cette litt´rature partage la vue selon e a e e e e laquelle l’h´t´rog´n´it´ de pr´f´rence au sein d’une relation principal-agent est, presque par d´ﬁnition, mauvaise pour ee e e e ee e l’eﬃcacit´ de l’organisation : une organisation eﬃcace devrait toujours ˆtre constitu´e de “clones” du principal. e e e Cependant, l’h´t´rog´n´it´ dans les pr´f´rences peut s’av´rer utile d`s lors que l’on prend en compte l’existence ee e e e ee e e d’une “division du travail” entre (1) les agents qui prennent des d´cisions et (2) ceux en charge de les impl´menter. e e Nous consid´rons donc une organisation constitu´e de deux employ´es : un D´cideur en charge de s´lectionner e e e e e les projets et un Impl´menteur en charge de leur ex´cution. Les deux agents ont des pr´f´rences intrins`ques, e e ee e potentiellement conﬂictuelles, sur les diﬀ´rents projets que l’organisation peut eﬀectuer. De plus, l’organisation e doit faire face ` des probl`mes d’asym´trie d’informations : d’un cˆt´, le D´cideur poss`de de l’information sur la a e e oe e e nature du projet le plus proﬁtable ` priori (parce qu’il va, par exemple, ` des meetings, des conf´rences et a acc`s a a e e a ` des m´mos conﬁdentiels) ; de l’autre cˆt´, l’eﬀort r´alis´ par l’Impl´menteur n’est pas observable. La cl´ de notre e oe e e e e cadre d’analyse est que le D´cideur doit anticiper l’eﬀort qu’est prˆt ` consentir l’Impl´menteur sur chacun des e e a e

Introduction G´n´rale e e

xiv

projets. En moyenne, un Impl´menteur ind´pendant (ie, un impl´menteur dont les pr´f´rences intrins`ques ne sont e e e ee e pas reli´es ` celles du D´cideur) a moins d’incitations ` fournir un eﬀort d’impl´mentation important sur le projet e a e a e pr´f´r´ du D´cideur que ne pourrait avoir un impl´menteur partageant les pr´f´rences du D´cideur. Anticipant cela, eee e e ee e le D´cideur est incit´ ` utiliser plus d’informations objectives dans sa prise de d´cision et ` moins prendre en compte e ea e a ses propres pr´f´rences, ce qui augmente la proﬁtabilit´ de l’organisation. ee e Cette “contrainte d’impl´mentation” a en retour d’importantes cons´quences sur la motivation de l’Impl´menteur. e e e Parce que le succ`s du projet est important pour l’Impl´menteur, il est dispos´ ` fournir plus d’eﬀort lorsque qu’il e e ea croit que le D´cideur a choisi le “bon” projet, c’est ` dire celui ayant le plus de chances objectives de r´ussir. Un e a e Impl´menteur ind´pendant – qui sait que sa simple pr´sence dans la chaˆ de commande augmente l’information e e e ıne “objective” utilis´e dans le processus de prise de d´cision – a de fortes croyances sur la proﬁtabilit´ du projet, ce e e e qui l’incite ` faire plus d’eﬀort dans l’impl´mentation. a e Cependant, l’ind´pendance le long de la hi´rarchie vient ` un coˆt. Comme le D´cideur est plus incertain des e e a u e pr´f´rences d’un Impl´menteur ind´pendant, il peut plus souvent choisir des projets qui se trouvent, ex post, ˆtre les ee e e e projets les moins aim´s par l’Impl´menteur. Ainsi, un Impl´menteur ind´pendant se voit plus souvent confront´ ` e e e e ea la mise en place de projets pour lesquels il n’a aucune pr´f´rence intrins`que, ce qui d´truit sa motivation ` r´aliser ee e e a e le projet. L’arbitrage que nous mettons en lumi`re dans cet article oppose donc (1) des projets objectivement e proﬁtables et (2) des agents avec une plus faible motivation intrins`que ex post. Comme nous le montrons, quand e l’information priv´e du D´cideur est suﬃsamment pr´cise, l’organisation optimale met en sc`ne des Impl´menteurs e e e e e ind´pendants aﬁn d’inciter le D´cideur ` utiliser cette information priv´e. e e a e Dans notre cadre hi´rarchique, l’h´t´rog´n´it´ dans les pr´f´rences peut ˆtre b´n´ﬁque pour l’organisation mais e ee e e e ee e e e pour des raisons diﬀ´rentes de celles qui sont g´n´ralement mises en avant dans des structures “horizontales”, comme e e e des commit´s ou des parlements. Dans de telles structure, la diversit´ peut ˆtre d´sirable, dans la mesure o` elle e e e e u permet aux biais individuels de se “neutraliser”. Dans l’organisation hi´rarchique que nous ´tudions, l’ind´pendance e e e optimale ´merge comme un m´canisme qui rend la “contrainte d’impl´mentation” plus contraignante, ce qui, sous e e e certaines conditions, peut am´liorer l’eﬃcacit´ de l’organisation. Cette interaction entre la prise de d´cision et e e e l’impl´mentation est au coeur de l’arbitrage que nous mettons en lumi`re dans ce chapitre 3 et il nous permet e e ensuite de d´river des comparatives statiques int´ressantes qui ne pourraient ˆtre obtenus dans un mod`le plus e e e e “horizontal”. Ce chapitre de ma th`se est ` relier ` certains travaux r´cents sur le design des organisations, notre princie a a e pale innovation ´tant l’´tude de l’ind´pendance des pr´f´rences dans un cadre de “division du travail”. Zabojnik e e e ee

Introduction G´n´rale e e

xv

(2002) est le seul papier int´grant explicitement cette s´paration entre la prise de d´cision et l’impl´mentation, e e e e mais l’organisation qu’il consid`re est constitu´e seulement d’agents motiv´s extrins`quement et son ´tude porte e e e e e sp´ciﬁquement sur le rˆle de la d´l´gation de l’autorit´ au sein de la hi´rarchie. Dessein (2002) pr´sente un mod`le e o ee e e e e de communication entre un principal et son agent dans un pure cadre de prise de d´cision. A cause de cette “hoe mog´n´it´” dans les tˆches eﬀectu´es par l’organisation, Dessein obtient que la communication est tr`s ineﬃcace e e e a e e d`s lors que les pr´f´rences entre le principal et l’agent sont suﬃsamment ind´pendantes, ce qui contraste fortee ee e ment avec nos propres r´sultats. Dewatripont et Tirole (2005) introduise un mod`le de communication coˆteuse e e u o` l’homog´n´it´ dans les pr´f´rences peut diminuer l’eﬃcacit´ de l’organisation. Alors que Dewatripont et Tirole u e e e ee e se concentrent, comme nous le faisons, sur le lien entre congruence et qualit´ de la prise de d´cision, leur th´orie e e e repose sur les probl`mes de “free-riding” qui apparaissent entre le principal et l’agent, d`s lors que tous deux doivent e e faire des eﬀorts pour communiquer. Une autre caract´ristique de notre mod`le est qu’il permet d’endog´n´iser la e e e e cr´dibilit´ du D´cider vis ` vis de l’Impl´menteur. Dans les mod`les de “signaling” (par exemple Hermalin (1997) en e e e a e e th´orie des organisations ou Cukierman et Tomasi (1998) en ´conomie politique), un principal inform´ peut souvent e e e envoyer des messages cr´dibles en utilisant des outils du type “money burning”. Dans notre mod`le, la capacit´ ` e e ea s’engager repose sur la forme organisationnelle a l’´quilibre : un Impl´menteur ind´pendant aide le D´cideur ` aug` e e e e a menter le contenu informationnel dans son processus de prise de d´cision, ce qui, en retour, augmente la cr´dibilit´ e e e des d´cisions prises. e Ce chapitre 3 permet ﬁnalement d’´clairer le d´bat sur le degr´ d’ind´pendance optimal des agences gouvernee e e e mentales relativement au pouvoir politique. La litt´rature de management sugg`re que les agences gouvernementales e e devraient ˆtre aussi ind´pendantes que possible du pouvoir politique (voir, par exemple, Horn (1995)). De telles e e recommandations pourraient ˆtre largement soutenues par notre mod`le, dans lequel le politicien (le D´cideur) est e e e biais´ mais a de l’information priv´ sur la demande sociale. L’impl´menteur – l’agence publique – peut ˆtre soit aﬃli´ e e e e e au pouvoir politique, soit constitu´ en organisation ind´pendante. Notre th´oriue sugg`re que lorsque la demande e e e e sociale est fondamentale pour le bien-ˆtre social, une bureaucracie non aﬃli´e au pouvoir politique aide ` la mise e e a en place de r´formes moins “biais´es” de la part des politiciens. e e

Investissement des entreprises et collat´ral : le rˆle des prix de l’immobilier e o
Le dernier chapitre de cette th`se, le chapitre 46 , nous conduit au coeur d’un d´bat empirique qui traverse la ﬁnance e e d’entreprise depuis pr`s de 20 ans : la mise en ´vidence empirique de l’existence et de l’ampleur des contraintes de e e
6 Le chapitre 4 est tir´ d’un article ´crit en collaboration avec Thomas Chaney et David Thesmar : “The Corporate Wealth Eﬀect: e e From Real Estate Shocks to Corporate Investment”, Mimeo

Introduction G´n´rale e e

xvi

cr´dit. En eﬀet, en pr´sence de frictions ﬁnanci`res, la valeur du collat´ral d´tenu par une entreprise peut avoir e e e e e un impact fondamental sur le montant que cette entreprise peut emprunter et cons´quemment sur la nature des e projets dans lesquels la ﬁrme peut investir. Barro (1976), Stliglitz and Weiss (1981) et plus r´cemment Kiyotaki e and Moore (1997) ont montr´ qu’en la pr´sence d’al´a moral ou de s´lection adverse, le collat´ral pouvait am´liorer e e e e e e les capacit´s d’endettement d’une ﬁrme, et donc augmenter ses investissements. En d´pit d’une recherche th´orique e e e abondante, il n’y a que peu d’´vidences empiriques du rˆle du collat´ral sur l’investissement des entreprises. La e o e litt´rature existante s’est plutˆt concentr´e sur la corr´lation empirique entre les disponibilit´s de l’entreprise et e o e e e l’investissement. Pourtant, ces disponibilit´s ne sont pas le seul actif qu’une entreprise peut utiliser pour ﬁnancer e de nouveaux investissements. Par exemple, les d´tentions immobili`res, mais ´galement les cr´ances commerciales, e e e e les stocks ou mˆme certains ´quipements productifs servent couramment de collat´ral pour garantir de nouveaux e e e emprunts (Davydenko et Franks (2005)). Le chapitre 4 fournit une ´tude empirique d´taill´e sur les eﬀets de certains chocs exog`nes de la valeur du e e e e collat´ral disponible pour les entreprises. Nos donn´es sugg`rent que les entreprises transforment les appr´ciations de e e e e capital immobilier en investissement suppl´mentaire. En d’autres termes, les entreprises investissent plus lorsque la e valeur de leurs actifs immobiliers s’appr´cie. Au lieu de vendre ces actifs, les entreprises ﬁnancent les investissements e additionnels en ´mettant des nouvelles dettes. L’apr´ciation de valeur des actifs immobiliers diminue le risque et les e e asym´tries d’information associ´s ` ces nouveaux emprunts. Les nouveaux contrats de dette sont par cons´quent de e e a e plus longue maturit´, sont en g´n´ral plus syndiqu´s et contiennent moins de convenants prot´geant les cr´anciers. e e e e e e Une telle relaxation des contraintes de cr´dit r´duit la proﬁtabilit´ moyenne pour les ﬁrmes dont la gouvernance est e e e la plus faible, tandis qu’au contraire, elle augmente la proﬁtabilit´ des entreprises dont les actionnaires sont forts. e Ce chapitre est important pour au moins deux raisons. La premi`re est positive : nos r´sultats sugg`rent en e e e eﬀet que de forts mouvements exog`nes dans la valeur des fonds propres de l’entreprises – les actifs immobiliers e dans notre cas – ont des eﬀets majeurs sur la demande des entreprises en bien d’´quipement. Un tel “eﬀet richesse e pour les entreprises” permettrait d’expliquer comment des chocs purement ﬁnanciers g´n`rent des ﬂuctuations e e macro´conomiques persistantes, ` la fa¸on de Bernanke et Gertler (1989). Notre article d´voile ainsi les microe a c e fondations de tels mod`les macro´conomiques. La seconde implication de notre analyse est normative. Comme les e e chocs sur les prix de l’immobilier permettent d’att´nuer les contraintes de cr´dit, la d´tention d’actif immobilier peut e e e servir d’instrument de couverture eﬃcace contre le risque de cr´dit. Fond´s sur l’analyse de Holmstrom et Tirole e e (2000,2001), nos r´sultats sugg`rent que les entreprises devraient b´n´ﬁcier d’autant plus de la d´tention d’actifs e e e e e immobiliers que leurs besoins de liquidit´ sont peu corr´l´s au rendement de l’immobilier. e ee

Introduction G´n´rale e e

xvii

La premi`re raison pour laquelle nous nous concentrons sur des chocs sp´ciﬁques de la valeur de l’immobilier e e tient essentiellement ` la facilit´ avec laquelle on peut mesurer ces chocs. Par ailleurs, l’immobilier est un type de a e collateral dont l’usage est largement r´pandu, aussi bien dans les pays d´velopp´s (Davydenko et Franks (2005)) ou e e e dans les pays en voie de d´veloppement (World Bank Survey (2005)). La troisi`me vertue de l’immobilier pour notre e e ´tude est que son prix ﬂuctue largement, si bien que la valeur du collat´ral d´tenu par les entreprises et qui peut ˆtre e e e e mobilis´ pour garantir de nouveaux emprunts varie d’une ann´e sur l’autre. Une part au moins de ces ﬂuctuations e e peut ˆtre consid´r´ ` priori comme exog`ne aux variations des opportunit´s d’investissement des entreprises hors e eea e e des secteurs de la ﬁnance, de l’assurance, de la construction et de l’immobilier. Toutes ces caract´ristiques des actifs e immobiliers nous permettent d’identiﬁer proprement l’eﬀet du collat´ral sur l’investissement. e En premier lieu, nous nous int´ressons ` la sensibilit´ de l’investissement aux prix locaux de l’immobilier. Bien e a e ´videmment, une telle corr´lation devrait ˆtre en g´n´ral positive, tout simplement parce que les prix de l’immobilier e e e e e sont fortement corr´l´s aux chocs de demande locaux, et que les entreprises ont tendance ` construire des capacit´s ee a e aﬁn de pouvoir servir ces chocs de demande. Pour ´viter ce biais d’endog´n´it´, nous comparons les entreprise qui e e e e poss`dent de l’immobilier avec celles qui n’en ont pas, ce qui nous permet d”interpr´ter cette sensibilit´ diﬀ´rentielle e e e e comme l’eﬀet de la ﬂuctuation de la valeur de l’immobilier sur la politique d’investissement. Notre strat´gie empirique e repose donc sur deux sources d’identiﬁcation. La premi`re vient de la comparaison, au sein d’une mˆme r´gion, des e e e r´ponses de l’investissement entre les ﬁrmes d´tentrices d’immobiliers et les autres. La seconde source d’identiﬁcation e e consiste ` comparer, pour les entreprises poss´dant de l’immobilier l’investissement dans les r´gions o` l’inﬂation a e e u immobili`re est ´lev´e et les r´gions o` elle est faible. C’est une approche similaire ` celle utilis´e, par exemple, par e e e e u a e Case, Quigley et Shiller (2001). En d´pit de toutes ces pr´cautions, imaginons que la diﬀ´rence entre les entreprises poss´dant de l’immobilier e e e e et les entreprises n’en poss´dant pas est que les premi`res tendent ` ˆtre plus “locales” que les derni`res. Alors, et e e ae e hors de toute consid´ration li´e au collat´ral, les entreprises d´tenant de l’immobilier pourraient ˆtre plus sensibles e e e e e aux variations dans les prix de l’immobilier simplement car les prix de l’immobiliers reﬂˆtent les ﬂuctuations de e demande locale. Nous prenons au s´rieux cette possibilit´ et soumettons nos r´sultats ` diﬀ´rents tests de robustesse. e e e a e En particulier, nous identiﬁons une source de variation des prix de l’immobilier qui est orthogonale aux chocs de demande locale. Nous utilisons pour cela des restrictions r´gionales dans l’oﬀre de terrains constructibles (r´gulations e e municipales, contrˆle des loyers, etc. . . ) aﬁn de pr´dire la r´action des prix locaux de l’immobilier ` des chocs de o e e a taux d’int´rˆt. De telles restrictions vont aﬀecter la d´termination des prix de l’immobilier sans ˆtre pour autant ee e e corr´l´ ` des chocs de demande locaux. En utilisant ces restrictions comme instrument pour les prix de l’immobilier, eea

Introduction G´n´rale e e

xviii

nous trouvons des estimations tr`s proches de celles obtenues en utilisant des proc´dures standards. e e En nous appuyant sur cette strat´gie empirique, nous trouvons tout d’abord un eﬀet causal robuste de l’inﬂation e immobili`re sur les d´cisions d’investissement des entreprises. L’estimation sugg`re que, pour chaque dollar addie e e tionnel dans la valeur des actifs immobiliers, les entreprises investissent pr`s de 60 cents de plus. Au total, quand e l’immobilier s’appr´cie d’une d´viation standard, les entreprises avec des actifs immobiliers augmentent leur niveau e e de d´penses d’investissement de pr`s de 2%, relativement aux entreprises n’ayant pas d’immobilier : compte-tenu e e du fait que ces actifs immobiliers ne repr´sentent en g´n´ral qu’une tr`s faible fraction de l’actif total des entree e e e prises, cette eﬀet est qualitativement important. Ainsi, bien que l’immobilier soit un actif bien moins liquide que le cash, il n’en reste pas moins que les entreprises l’utilisent comme source de collat´ral pour ﬁnancer de nouveaux e investissements. Nous nous int´ressons ensuite aux diﬀ´rents m´canismes permettant de transformer l’appr´ciation de la valeur de e e e e l’immobilier en investissement additionnel. Nous trouvons que les entreprises poss´dant de l’immobilier dans les ´tats e e o` l’immobilier s’appr´cie modiﬁent signiﬁcativement la structure de leur capital, principalement en augmentant la u e part de la dette long terme relativement ` l’actif total. Pour am´liorer notre compr´hension du ph´nom`ne ` l’oeuvre, a e e e e a nous regardons en d´tail la nature des contrats de dettes sign´s aupr`s des institutions ﬁnanci`res. Il semble que les e e e e contrats mis en place par des entreprises d´tenant de l’immobilier suite ` une appr´ciation de l’immobilier soient e a e moins aﬀect´s par les probl`mes d’asym´tries d’information. Ces contrats sont plus syndiqu´s, ont une maturit´ e e e e e plus longue et poss`dent moins de convenants, ie de clauses sp´ciales imposant des minima de performance aﬁn de e e prot´ger les cr´anciers ` la fois de l’al´a moral et de la s´lection adverse existant dans leur relation avec l’entreprise. e e a e e Ce chapitre 4 conclue en ´tudiant la proﬁtabilit´ des investissements r´alis´s ` partir de cette augmentae e e e a tion exog`ne de la valeur du collat´ral. Comme Blanchard, Lopez de Silanes et Shleifer (1994) en discutent, e e l’investissement peut r´pondre ` de telles augmentations pour deux raisons : parce qu’il y a de la s´lection adverse e a e sur les march´s ﬁnanciers (Myers et Majluf (1984)) ou parce qu’il y a de l’al´a moral chez les managers de l’entreprise e e (Jensen et Meckling (1976)). Dans le premier cas, une relaxation des contraintes de cr´dit devrait augmenter la e valeur de la ﬁrme, alors qu’elle devrait la diminuer dans le second cas. Aﬁn de tester cette id´e, nous s´parons les e e ﬁrmes dont les actionnaires sont “forts” des ﬁrmes dont les actionnaires sont “faibles”, en utilisant des indices standards de gouvernement de l’entreprise. Nous trouvons que, pour les ﬁrmes dont la gouvernance est bonne, les chocs positifs de collat´ral ne conduisent pas ` une plus faible proﬁtabilit´. N´anmoins, les entreprises dont l’actionnariat e a e e est “faible” et qui d´tiennent des actifs immobiliers voit bien leur proﬁt diminuer quand survient une appr´ciation e e de l’immobilier. Ainsi, les chocs de collat´ral que nous mettons en lumi`re dans ce chapitre nous fournissent un e e

Introduction G´n´rale e e

xix

moyen eﬃcace d’´valuer certains eﬀets r´els du conﬂit actionnaires-manager. Alors que la plupart des th´ories exe e e istantes relie l’investissement ` la capacit´ d’endettement, et la capacit´ d’endettement au collat´ral, la litt´rature a e e e e empirique s’est concentr´e sur l’eﬀet direct des b´n´ﬁces d’exploitation (les cash ﬂows) sur l’investissement (Faze e e zari, Hubbard, and Petersen (1988)). Erickson et Whited (2000) et Hennessy et Whited (2007) ont critiqu´ cette e approche au motif que, pour des raisons aussi bien th´oriques qu’empiriques, de telles coeﬃcients ne peuvent ˆtre e e interpr´t´es facilement. Ces auteurs ont donc pr´conis´ l’usage de m´thodes d’estimation du type GMM. Une autre ee e e e branche de la litt´rature a tent´ d’isoler des chocs de cash ﬂows orthogonaux aux opportunit´s d’investissement e e e (Blanchard, Lopez-de-Silanes et Shleifer (1994), Lamont (1997), Rauh (2006)). Le chapitre 4 appartient clairement a ` cette tradition. Encore plus proche de notre article, Almeida, Campello and Weisbach (2004) se sont interess´s e au rˆle des liquidit´s au bilan sur l’investissement, plutˆt qu’aux seuls ﬂux de cash. Ces auteurs ont montr´ que les o e o e entreprises contraintes sur le march´ du cr´dit tendent ` stocker des liquidit´s sur leur bilan aﬁn d’ˆtre toujours en e e a e e mesure de ﬁnancer dans le futur des investissements rentables. Plutˆt que de regarder le cash (en ﬂux ou en stock), o notre ´tude consid`re les variations exog`nes dans la valeur du collat´ral, et ce sur un large panel d’entreprises. A e e e e notre connaissance, les seuls papiers existants sur des chocs de collat´ral sont Peek et Rosengreen (2000), Goyal et e Yamada (2001) et Gan (2006). Ces contributions se concentrent sur l’investissement dans le contexte tr`s sp´ciﬁque e e de la bulle immobili`re japonaise des ann´es 80. Le chapitre utilise une strat´gie d’identiﬁcation plus robuste – des e e e diﬀ´rences triples au lieu de simples diﬀ´rences de diﬀ´rence – et examine en d´tail les m´canismes qui permettent e e e e e de transformer ce collat´ral en investissement. e

Performance and Behavior of Family Firms

1

Chapter 1

Performance and Behavior of Family Firms: Evidence from the French Stock Market
1.1 Introduction

While, since Berle and Means, ﬁnancial economists have devoted a lot of attention to large, listed and widely held corporations, it turns out that most ﬁrms around the world have a dominant owner, in many instances the founding family.1 In addition, founding families are often involved in the actual management of the ﬁrm. In our own sample, which comprises the set of all listed French ﬁrms, more than 60% of the ﬁrms are still managed by their founding family. Even among the largest US ﬁrms, Anderson and Reeb (2003) report that some 16% of their sample of S&P500 ﬁrms are still managed by their founders or descendants. Therefore, the premises of the Berle and Means model of the corporation where (1) the CEO is not an owner and (2) ownership is dispersed, do not apply to most ﬁrms around the world. The relevant view on world capitalism is thus that the typical large listed ﬁrm is owned, and frequently managed, by a family. This new perspective calls forth a research agenda on the speciﬁc features of dynastic management and ownership. Do family ﬁrms maximize proﬁt? Are they more prone to building family empire at the expense of minority shareholders? Are they too prudent, slow reacting? On the contrary, are they more cool-headed and better at avoiding fads? More generally, do they behave any diﬀerently from the widely held corporations that academics know so well? To bridge this gap, this paper provides evidence on the performance and behavior of family ﬁrms in France. example, Laporta, Lopez-de-Silanes and Shleifer (1999) track ultimate ownership of a sample of ﬁrms listed in 27 rich countries with more than $ 500m market capitalization. They ﬁnd that 50% of all ﬁrms in their global sample are family-controlled, while only 40% of them are widely held or controlled by widely held entities. In fact, widely held corporations are prevalent in the US, the UK and Japan, while families predominate in most continental European countries. Focusing on these countries, Faccio and Lang (2000) ﬁnd that more than 60% of all listed ﬁrms in France, Italy and Germany are family ﬁrms.
1 For

Performance and Behavior of Family Firms

2

We believe the French example is of interest for two reasons. First, France is a continental European country, and as a consequence its ﬁnancial institutions and history diﬀer markedly from English-speaking countries, where most systematic studies on family ﬁrms have been conducted so far.2 In particular, family ﬁrms are much more prevalent even among the largest ﬁrms and therefore much more representative of the economy than in the US. The second reason is that, in contrast to many continental European countries, France also has a lot of widely held ﬁrms, which tend to be very much US-like managerial ﬁrms – no dominant owner and an entrenched management. This gives us access to a large enough control group to compare family ﬁrms to. We collected data on some 1,000 corporations listed on the French stock market over the 1994-2000 period. Our panel has information on the ﬁrm (employment, corporate accounts, acquisitions, stock returns) and on the founding family (ownership, management). This dataset is supplemented with information on acquisitions, stock returns and detailed information on the wage bill and skill structure. First, we provide cross-sectional evidence on the relative performance of family ﬁrms. Looking at accounting proﬁtability, we ﬁnd that family ﬁrms signiﬁcantly outperform non family ﬁrms. Consistently with the existing literature on “founder eﬀects”,3 we ﬁnd evidence that founder-managed ﬁrms are very proﬁtable. Also consistent with available evidence from the US (Anderson and Reeb (2003), Amit and Villalonga (2006)), family ﬁrms run by an outside CEO outperform widely held corporations. It thus seems that in France as in the US, families as large shareholders are, on average, good for performance. Much more surprisingly, we ﬁnd that managers who are descendants of the ﬁrm’s founder also tend to do better than non family ﬁrms, and even marginally better than professional managers in our sample. Therefore, even if we set founders aside, family ﬁrms (whether or not run by a descendant) consistently outperform non family ﬁrms in France. A causal interpretation of such crosssectional evidence is diﬃcult since only the best performing ﬁrms may be transmitted to heirs. A potential solution involves looking at transmissions of control in our data (as in P´rez-Gonz´lez (2006)). While we may have too few e a transitions in our sample to do statistically powerful tests, we nevertheless observe that (1) descendants typically do not inherit the management of the best ﬁrms and (2) descendants whose ﬁrms leave the stock market (de-list) do not systematically underperform. Endogeneity biases are thus not likely to be large in our sample. We then try to understand these diﬀerences in performance by considering the various determinants of corporate performance. Higher labor productivity seems to be the most signiﬁcant explanation for the higher performance of founder-managed corporations. Turning to the diﬀerence in performance between professionally managed and
2 For two recent exceptions, see Barontini and Caprio (2005) on Italy and Bennedsen, Nielsen, P´rez-Gonz´lez and Wolfenzon (2006) e a on Denmark. 3 See, e.g., Adams, Almeida and Ferreira (2005) and Fahlenbrach (2005).

Performance and Behavior of Family Firms

3

heir-managed family ﬁrms, we look separately at the management of labor and capital inputs. First, family ﬁrms run by both professionals and descendants are paying, on average, lower wages. Nevertheless, this result could simply stem from a diﬀerent workforce composition in these family ﬁrms, e.g. a larger fraction of unskilled workers. To account for such a possibility, we match our ﬁrm-level data with employee-level social security records, which allows us to control for the labor force composition in experience, seniority and occupations. We ﬁnd that professional CEOs in family ﬁrms pay lower wages because they indeed employ less skilled employees. On the contrary, even after controlling for their ﬁrm’s skill structure, descendants still pay low wages. We also ﬁnd that labor demand in heir-managed family ﬁrms responds signiﬁcantly less to industry sales shocks, i.e. that heir managers seem to be smoothing out employment across the business cycle. Overall, these results are consistent with a view of heir-managed family ﬁrms as providing their workers with long-term implicit insurance contracts. Such contracts allow them to pay lower wages for better skills. And indeed, heir-managed family ﬁrms exhibit, compared to professionally managed corporations, a higher level of labor productivity. We then investigate diﬀerences in the management of capital. Professional CEOs pay lower interest rates on their debt and tend to operate at lower capital to labor ratios. Compared to heir-managed family ﬁrms, or even widely held corporations, large acquisitions made by outside CEOs destroy less shareholder value in the long run. Although a-structural and thus to be interpreted with caution, the broad picture emerging from these results indicates that (1) outside CEOs bring ﬁnancial expertise to family ownership and are keener on avoiding the waste of capital, while (2) descendants have the managerial horizon necessary to commit to a protective employment policy, and are rewarded by a larger labor productivity. We indeed report that, in our data, descendants tend to survive longer as CEOs than professional managers. Our paper is thus a contribution to the emerging economics literature on family business. Most of this infant literature has, so far, focused on North-American ﬁrms and proﬁtability comparisons. Among the various contributions (Anderson and Reeb (2003), Amit and Villalonga (2006), and P´rez-Gonz´lez (1999) for the US; Morck, e a Strangeland and Yeung (2002) for Canada), the consensus is that founder-managed ﬁrms, as well as family ﬁrms run by an outside CEO outperform non family ﬁrms. This is usually interpreted as evidence that the beneﬁts of a large, long term, shareholder outweigh the costs of potential minority shareholders expropriation. The managerial quality of descendants is, however, a much more debated issue. Two cross-sectional studies on large US corporations in the 1990s (Anderson and Reeb (2003) and Amit and Villalonga (2006)) ﬁnd opposite results. The diﬀerence in diﬀerence approach taken by P´rez-Gonz´lez (2006) suggests, however, that heirs may be worse managers than e a outside CEOs. Such an insight is conﬁrmed by a recent paper by Bennedsen, Nielsen, P´rez-Gonz´lez and Wolfene a

Performance and Behavior of Family Firms

4

zon (2006), which takes advantage of the richness of Danish data to carefully study causality. Our paper has the advantage to supplement these studies for a large country of continental Europe: we ﬁnd that descendants do not do worse – even slightly better – than professional managers of family or widely held ﬁrms. As already stressed out, apart from the causal eﬀect of family management, many selection, endogeneity and simultaneity biases could yet be explaining this cross-sectional correlation. More interestingly, our paper also complements the existing family ﬁrms literature by looking at eﬀects of family ownership/management on other dimensions of ﬁrm behavior. Our robust ﬁnding that family ﬁrms pay lower wages is, to our knowledge, new to this literature. It is reminiscent of existing evidence on the relationship between wage levels and the separation of ownership and control in corporations. A decade old literature recently surveyed by Bertrand and Mullainathan (1999), along with their own ﬁndings, indeed ﬁnds that managerial slack in organizations partly takes the form of higher wages among employees. The other novelty of our paper is the analysis on the diﬀerence in management styles between hired CEOs and descendants of the founders: the data are consistent with outside CEOs bringing ﬁnancial expertise and reducing the waste of capital, while heirs being able to commit to long term employment and therefore obtaining lower wages from their employees. Such results can be related to Bertrand and Schoar (2003)’s analysis of American CEOs’ management styles: they ﬁnd strong diﬀerences between individuals in terms of investment policy, acquisition policy and ﬁnancing policy. In particular, MBA graduates tend to be more aggressive in terms of leverage and acquisition policy. Our own results suggest that family management/ownership might be yet another dimension to explain such heterogeneity in management styles.4 This paper is structured as follows. Section 2 presents the data construction and describes its content. Section 3 provides more systematic evidence on corporate performance. Section 4 looks at diﬀerences in corporate behavior between family and non family ﬁrms and between descendants and professional managers within family ﬁrms. Section 5 concludes.

1.2

Data Description

Our dataset is a panel of French listed ﬁrms over the 1994-2000 period. We restrict ourselves to non ﬁnancial, non real-estate ﬁrms. The construction of this dataset uses 5 diﬀerent sources. First, annual corporate accounts are retrieved from the DAFSA yearbooks. These books cover the set of all listed ﬁrms in France. These books also provide us with the identity of the management team, and the stakes held by the main shareholders. Second, we hand
4 In a similar vein, Bertrand et al. (2005) identify on Thai data a relationship between the nature and complexity of the family owning the ﬁrm and its performance.

Performance and Behavior of Family Firms

5

collect information on family management and ownership for most of these ﬁrms using various sources (newspapers, ﬁrm websites, annual report...). Third, we use social security records to retrieve ﬁrm level information on wages, occupational structure, age and seniority structure and gender composition. Fourth, we collect information on the major corporate acquisitions realized by these ﬁrms over the 1994-2000 period. To do this, we use the French extract of SDC platinium, a worldwide database on corporate transactions. Fifth, we obtain stock prices for these listed ﬁrms from Euronext for the 1991-2002 period.

1.2.1

Family Business on the French Stock Market

Our deﬁnition of a family ﬁrm is very close to the one used by Amit and Villalonga (2006). We report a ﬁrm as a family ﬁrm when the founder or a member of the founder’s family is a blockholder of the company. We also impose as an additional condition that this block represents more than 20% of the voting rights.5 We refer to Appendix A.1.2 for more detailed explanations on the construction of our family ﬁrm variable. Following Anderson and Reeb (2003), we then break down our sample of ﬁrms into four categories. All ﬁrms that are not family ﬁrms are called widely held ﬁrms. The listed ﬁrms that are controlled by widely held ﬁrms also belong to this category.6 When a family ﬁrm is still managed by its founder, we refer to it as a founder-managed family ﬁrm. As is detailed in A.1, this category also entails ﬁrms owned and managed by a successful raider.7 Heirmanaged family ﬁrms are family ﬁrms where the current CEO is a descendant of the company’s founder. Finally, when a family ﬁrm is run by an outside, professional CEO, we refer to it as a professionally managed family ﬁrm. To be able to compute accounting proﬁtability measures properly, we restrict our study to non ﬁnancial, non real estate companies. There are 2,973 observations in our panel (some 420 ﬁrms each year), for which we were able to retrieve the ﬁrm’s family status.8 Table 1 reports the fractions of the various types of ﬁrms in our panel. These fractions are computed without weight (line 1), weighted with book value of assets (line 2) and weighted using total employment, as reported in the accounting data (line 3). As is apparent from Table 1.1, 70% of all ﬁrms present in the sample are family ﬁrms. This is a very large number, compared to what previous studies found for English-speaking countries. Looking at US listed ﬁrms from the S&P500, Anderson and Reeb (2003) ﬁnd 35% of family-controlled companies, although they use a slightly
5 As it turns out, this additional condition is not very important as we had very few cases where family shareholders held less than 20% of the voting rights. 6 As the results in Holderness and Sheehan (1988) suggest, ﬁrms controlled by a widely held ﬁrm and widely held ﬁrms themselves are not diﬀerent in terms of both Tobin’s Q and accounting rate of returns. 7 This is where our classiﬁcation diﬀers somewhat from Anderson and Reeb (2003)’s, who focus on founding families. Hence, a successful raider would not count as a founder, and its ﬁrm would count as a widely held ﬁrm according to their categorization. Casual evidence indicates that in France, these raiders tend to have dynastic concerns, which explains our choice. 8 Out of a total of 3,522: this means that in 16% of the cases, we were unable to categorize the ﬁrm’s ownership or management.

Performance and Behavior of Family Firms

6

Table 1.1: Presence of Family Firms
All ﬁrms Widely Held Firms Founder Family Firms, managed by: Heir Professional CEO

Fraction (non weighted) Fraction (asset weighted) Fraction (empl. weighted)

1.00

0.29

0.31

0.24

0.16

1.00

0.65

0.08

0.11

0.16

1.00

0.55

0.10

0.16

0.19

Observations

2,973

864

922

721

466

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details and sources. Note: line 1 gives the unweighted fraction of the diﬀerent family status in our sample ; line 2 gives the same fraction, but weights each observation by the book value of asset ; line 3 weights the observations by total employment, as reported in the accounting data.

diﬀerent deﬁnition of family ownership. Looking at the largest 500 listed Canadian ﬁrms, Morck, Strangeland and Yeung (1998) ﬁnd a share of 50% of family ﬁrms. Our sample is more consistent with the investigations of Faccio and Lang (2002), who look at the ultimate ownership of listed ﬁrms in continental European countries: using various data sources, they ﬁnd in 1997, for France, 64% of family ﬁrms. Thus, family ownership appears much more pervasive in France than in English-speaking countries, even Canada. The surprising fact is, however, that the bulk of these family ﬁrms is still founder-controlled, since these account for 31% of the total. In contrast, only 18% of all ﬁrms investigated by Morck et al. (1998) in Canada are still managed by the initial entrepreneur. It seems that the French stock market may display more mobility than the sheer fraction of family ﬁrms might suggest. But the family status is also very persistent: heir-managed ﬁrms account for a large share of the total (24%) in the same proportion as widely held ﬁrms. Last, less than a fourth of all family ﬁrms are managed by a professional CEO: hence, even after the founder retires, the norm seems to be that an heir takes over control. Of course, the real importance of family ﬁrms is overstated by these ﬁgures. Lines 2 and 3 of Table 1.1 highlight the relative small size of family ﬁrms. In weighted terms, widely held ﬁrms account for almost two thirds of all ﬁrms. Founder-controlled corporations are especially small and only account for 10% of total employment.

Performance and Behavior of Family Firms

7

1.2.2

Do Family Firms Diﬀer from Other Firms?

Table 1.2 allows to look for systematic diﬀerences between the four types of ﬁrms we have deﬁned in the previous section. First, family ﬁrms grow, on average, much faster than non family ﬁrms, but this is mostly due to the contribution of founder-managed corporations. For these corporations, sales growth stands around 16%, instead of 9% for the average listed ﬁrm. A similar picture arises for the ratio of market to book value of assets.9 In contrast, when we look at accounting proﬁtability, all types of family ﬁrms do better than widely held ﬁrms. Founder-managed ﬁrms are the most proﬁtable ones. That founders do better in terms of proﬁts, growth and valuation is consistent with the extensive literature documenting “Founder eﬀects”(see, for a survey, Adams, Almeida and Ferreira (2005), Fahlenbrach (2005)). In a cross section, founders tend to run ﬁrms with outstanding performance, the question being whether they are inherently good managers, or whether those founders who manage to keep control are only those who perform well. Using various instruments, Adams et al. (2005) suggest that selection issues are minor, and that almost all of the founder eﬀect may be interpreted in a causal way. Using US data on listed ﬁrms, they ﬁnd a founder eﬀect on ROA of around 3 percentage points in OLS regressions and of around 2 points when using their instruments. Our cross tabulation suggests it might be even larger in the French context, although a multivariate analysis needs to be run to estimate such an eﬀect. Even when we set founders aside, family ﬁrms are still more proﬁtable than widely held corporations, although to a lesser extent. The result is particularly striking for return on equity (ROE10 ), but is also present when we look at returns on assets (ROA11 ). This is not too surprising as far as professionally managed family ﬁrms are concerned, as these companies have the advantage of having large, long term shareholders. Anderson and Reeb (2003), and Amit and Villalonga (2006) ﬁnd similar results in the cross sections of their sample of large US listed ﬁrms. For France, the concern could be that large shareholders might be using their voting rights to pursue value destroying projects that grant them private beneﬁts. Results from Table 1.2 suggest that the beneﬁts from monitoring outweigh these potential costs of expropriation. Finally, the main surprise from Table 1.2 is that family ﬁrms run by descendants also outperform widely held corporations in terms of proﬁtability. The existing literature on large US ﬁrms provides mixed evidence: while Anderson and Reeb (2003) have similar results, Amit and Villalonga (2006) and P´reze Gonz´lez (2006) exhibit opposite ones. In Canada, Morck et al. (1998) ﬁnd that heir-managed Canadian ﬁrms a underperform all other types of ﬁrms. Overall, the balance of evidence from North American studies tilts in the
9 Market to book is measured as the sum of market capitalization and (book value asset minus book value of equity) divided by book value of total assets. 10 Return on Equity is deﬁned as the ratio of earnings to book value of equity. 11 Return on Asset is deﬁned as the ratio of EBITDA to book value of total assets.

Performance and Behavior of Family Firms

8

Table 1.2: Characteristics of Family Firms
All ﬁrms Widely Held Firms Founder Family Firms, managed by: Heir Professional CEO

Total Assets (bn euros) Total Sales (bn euros) Employment

2.3 1.9 10,489

5.2 3.8 22,184

0.6 0.4 3,845

1.0 1.1 7,685

2.3 2.8 14,537

Age (years) Former SOE (fraction)

62 0.09

66 0.25

32 0.01

84 0.01

70 0.07

ROA ROE Market to Book Sales growth Dividend / proﬁt Debt / Assets

0.12 0.19 1.4 0.09 0.22 0.38

0.10 0.15 1.3 0.07 0.26 0.39

0.15 0.27 1.6 0.16 0.19 0.38

0.13 0.20 1.3 0.07 0.21 0.38

0.12 0.21 1.4 0.10 0.21 0.38

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details and sources. Note: column 1 provides summary statistics for all ﬁrms in the sample ; column 2 to 5 detail these statistics by family status. ROA is deﬁned as the ratio of EBITDA to book value of total asset. ROE is deﬁned as the ratio of earnings to book value of equity. Market to Book is deﬁned as the sum of market capitalization and (book value asset minus book value of equity) divided by book value of total assets.

Performance and Behavior of Family Firms

9

direction of underperforming heir management. The obvious problem with the univariate approach, however, is that family status in cross tabulations may be a proxy for other variables, notably age and size. That family ﬁrms are smaller than non family ﬁrms is conﬁrmed by an examination of Table 1.2, which also provides a comparison of the various types of ﬁrms in terms of size, age and capital structure. On all accounts, widely held ﬁrms are much larger than family ﬁrms, and slightly older too. This conceals, however, a great heterogeneity between family ﬁrms. For example, family ﬁrms run by an outside CEO are those who resemble the most widely held ﬁrms, both in terms of age and size. Firms run by descendants are on average smaller, but older than the average corporation in our sample. As expected, ﬁrms still run by their founders are young and very small.

1.3

Multivariate Evidence

Given that family ﬁrms tend to have a diﬀerent age and size than widely held ﬁrms, it is necessary to conduct a multivariate analysis. Our empirical strategy follows the approach taken by Anderson and Reeb (2003) in their cross-sectional analysis of US family ﬁrms.

1.3.1

Empirical Strategy

We estimate the following equation, for ﬁrm i at date t: yit = α + β1 F1i + β2 F2i + β3 F3i + γXit + εit (1.1)

where yit is a measure of corporate performance (based on accounts, market value or dividend pay-out). Fi = (Fi1 , Fi2 , Fi3 ) is our family status variable, broken down into three dummy variables representing “foundercontrolled” (F1 ), “heir-controlled” (F2 ) and “professionally managed” (F3 ), the “widely held” ﬁrm being our reference. Fi varies little with time, so we cannot identify ﬁrm ﬁxed eﬀects with this equation. As argued above, this is a major concern if we want to interpret our results in a causal way; we will therefore try to avoid it, and will postpone the discussion on endogeneity and selection. Given the absence of ﬁrm ﬁxed eﬀects, the best we could do was to allow for ﬂexible correlation across residuals εit of a given ﬁrm, using White’s (1980) method. The Xit ’s are various, possibly time varying, controls. They include year dummies, 13 industry dummies, the ﬁrm’s log assets, the ﬁrm’s log age. As further control, we also add a dummy equal to 1 when the ﬁrm has been, at some point, state-owned. As it turns out, 25% of now widely held ﬁrms used to be government enterprises (that were nationalized in 1936, in 1945 or in 1981). Privatizations started when the Right came back to power in 1986-1988, and after 1993, under both left-wing and right-wing governments. Many of these privatizations took place through

Performance and Behavior of Family Firms

10

IPOs on the stock market, in order to ensure political support from the population and to increase the size and depth of the French stock market. As it may be the case that widely held ﬁrms underperform because they face diﬃculties to adjust to privatization, we need to control for this in our regressions. We also add other controls, which can be both causes and eﬀect of corporate performance. We will thus not comment on them since they are highly endogenous; yet, we include them to replicate the regressions run by Anderson and Reeb (2003) on their US sample, and also because they could be argued to be correlated with both family status and corporate performance. To control for the eﬀect of ownership concentration – which is likely to be correlated with the presence of a family among shareholders – we add the percentage of cash ﬂow rights held by the largest shareholder (and its square, in non reported regressions). Using our data on stock prices, we also include the variance of past stock returns as a proxy for ﬁrm speciﬁc risk.12 We also include ﬁrm leverage, measured by the ratio of debt to total assets. A theoretical reason for this additional control is, for instance, Jensen (1988)’s theory of free cash ﬂows, which generates a positive correlation between leverage and performance, debt being used as a disciplining device. On the contrary, high debt could also be the result of bad performance. As it turns out, leverage comes out signiﬁcantly negative in most performance regressions, which lends more credence to the second mechanism.

1.3.2

Family Firms Outperform widely held Firms

We focus on four diﬀerent measures of corporate performance. We use three measures of accounting proﬁtability : return on assets (deﬁned as EBITDA divided by book value of total assets), return on equity (deﬁned as earnings divided by book value of equity) and the pay-out ratio (deﬁned as dividends divided by pre-tax proﬁt – computed only for ﬁrms with positive pre-tax proﬁt). We also look at a measure of market valuation, the market to book ratio (deﬁned as the sum of market capitalization and (book value of asset minus book value of equity) divided by book value of total assets13 ). Table 1.3 reports two sets of regressions. In columns 2, 4, 6 and 8, we report the regressions of corporate performance on the explanatory variables as in equation 1.1. In columns 1, 3, 5 and 7, we group all family ﬁrms dummies together into one single “family ownership”dummy. This amounts to assuming that all management arrangements in family ﬁrms have an equal eﬀect on performance, i.e. β1 = β2 = β3 . We ﬁrst turn to accounting measures of performance. A quick examination of Table 1.3 shows that family ﬁrms outperform non family ﬁrms in our sample of listed ﬁrms (columns 1 and 3). The diﬀerence in ROA is 1.7 percentage points and the diﬀerence in ROE is as high as 9.6 percentage points. These diﬀerences are both for instance, families could simply be more proﬁtable because they undertake riskier projects. to book is therefore a mesure of the market value of assets, though the lower quality of our consolidated accounts does not allow us to obtain as clean a measure as in US studies using COMPUSTAT.
13 Market 12 As,

Performance and Behavior of Family Firms

11

Table 1.3: Performance of Family Firms
Return on assets (×100) Family ﬁrm 1.7∗∗∗ (0.6) Founder CEO 1.8∗∗∗ (0.8) Heir CEO 1.9∗∗∗ (0.7) Professional CEO 1.5∗∗∗ (0.7) Log (Assets) -.3∗∗∗ (0.1) Log(Firm Age) -.6∗∗ (0.3) Former SOE -0.9 (0.7) Fraction equity of largest block Debt / Assets 0.4 (1.0) -9.2∗∗∗ (1.2) Stock return volatility -8.1∗∗∗ (1.9) -.4∗∗∗ (0.1) -.7∗∗∗ (0.3) -1.0 (0.7) 0.4 (1.0) -9.3∗∗∗ (1.2) -8.2∗∗∗ (1.9) 1.6∗∗∗ (0.5) -4.0∗∗∗ (0.8) 1.0 (2.4) 3.4 (2.8) -15.6∗∗∗ (4.4) -16.5∗∗∗ (5.7) Return on equity (×100) 9.6∗∗∗ (1.7) 10.3∗∗∗ (2.2) 9.4∗∗∗ (1.9) 9.0∗∗∗ (1.9) 1.7∗∗∗ (0.5) -3.8∗∗∗ (0.9) 1.0 (2.4) 3.4 (2.8) -15.8∗∗∗ (4.4) -16.4∗∗∗ (5.7) 0.03∗∗ (0.01) -0.13∗∗∗ (0.03) -0.19∗∗ (0.09) -0.04 (0.08) -0.55∗∗∗ (0.09) -0.75∗∗∗ (0.12) 0.08∗ (0.05) 0.15∗∗∗ (0.06) 0.04 (0.06) 0.06 (0.06) 0.03∗∗∗ (0.01) -0.11∗∗∗ (0.03) -0.18∗∗ (0.09) -0.04 (0.09) -0.56∗∗∗ (0.09) -0.73∗∗∗ (0.13) -0.9∗∗ (0.4) 1.9∗∗ (0.9) 1.7 (2.0) 1.9 (3.1) 2.0 (3.5) 1.6 (6.8) Market to book -5.5∗∗∗ (1.8) -7.6∗∗∗ (2.3) -4.8∗∗ (2.1) -4.8∗∗ (2.0) -1.0∗∗ (0.4) 1.4 (0.9) 1.5 (1.9) 2.0 (3.1) 2.3 (3.5) 1.2 (6.8) Dividend to proﬁt (×100) -

Industry FE Year FE Heir=Professional

yes yes

yes yes .65

yes yes

yes yes .79

yes yes

yes yes .84

yes yes

yes yes .98

Observations Adj.R2

2,325 .22

2,325 .22

2,329 .13

2,329 .13

2,248 .24

2,248 .23

1,138 .09

1,138 .10

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details on data construction and sources. Note: Huber-White-Sandwich estimates, allowing for correlation of all observations of a given ﬁrm. Dependent variables are ROA (ratio of EBITDA to book value of asset – column 1 and 2), ROE (ratio of earnings to book value of equity – column 3 and 4), Market to book ratio (market capitalization plus (book value of assets minus book value of equity) divided by book value of total assets – column 5 and 6) and pay-out ratio (dividends divided by pre-tax proﬁts – column 7 and 8). Family ﬁrms is a dummy indicating family ownership (column 1, 3, 5, 7). Founder CEO is a dummy indicating that the CEO is the founder of the ﬁrm. Heir CEO is a dummy indicating that the CEO is a descendant of the founder. Professional CEO is a dummy indicating that the CEO has been hired by the controlling family. Other explanatory variables are Log(Assets) (logarithm of the book value of total asset), Log(Firm Age) (logarithm of ﬁrm age measured in years plus one), Former SOE (dummy equal to 1 if the ﬁrm is a former state owned enterprise), Fraction equity of largest block (cash-ﬂow right of the largest identiﬁed shareholder), Debt/Asset (leverage ratio), Stock return volatility (standard deviation of the ﬁrm’s stock price). These regressions control for 13 industry ﬁxed eﬀects (Industry FE) and year ﬁxed eﬀects (Year FE). Line “Heir=Professional” provides the p-value of an equality test between the coeﬃcient “Heir CEO” and “Professional CEO”. Standard errors are in parenthesis. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1 % level of signiﬁcance.

Performance and Behavior of Family Firms

12

statistically signiﬁcant and economically large, since the sample standard deviation is 8 points for ROA and 23 points for ROE. Looking at columns 2 and 4, we see that the over-performance of family ﬁrms is present for all types of management. Both founders and outside CEOs working in family ﬁrms outperform widely held companies by respectively 1.8 and 1.5 points of ROA and 9.4 and 9.0 points of ROE. More surprisingly, heir-managed family ﬁrms are also more proﬁtable than widely held companies, by 1.9 percentage points of ROA. So all sub-categories of family ﬁrms outperform to a similar extent a benchmark of widely held ﬁrms. As it turns out, a test of equality is far from being rejected (F-probability = 0.81). These results are extremely robust and hold in front of various speciﬁcation checks, like removing various subsets of the control variables (not reported), running the regressions separately for each year (Table B1), and controlling for ﬁrm diversiﬁcation (Table B2). Our results are strikingly consistent with what Anderson and Reeb (2003) found for the US. Looking at ROA, they ﬁnd that founder-controlled ﬁrms outperform widely held ﬁrms by 3.5 percentage points – compared to 1.8 in our sample. Secondly, in their study, heir-controlled ﬁrms outperform widely held corporations by 2 percentage points, exactly like in ours. Last, and still in line with our results, professionally-run ﬁrms only outperform the control group by 1 point of ROA, which is not statistically signiﬁcant in their analysis. In contrast to Anderson and Reeb, we thus ﬁnd that professional managers are really similar to the rest of the family group. Last reference to the cross-sectional analysis in the literature, our results are completely at odds with Morck et al. (1998)’s evidence from Canadian ﬁrms although they adopt a similar sample construction. Indeed, Morck et al (1998) ﬁnd that heirs are the worse performers of all ﬁrms, whether family-controlled or widely held. Moreover, in their sample, even founders are outperformed by widely held corporations. This very last result in their study is surprising in light of the extensive “founder eﬀect” literature mentioned earlier. We then ask how, in the French context, the stock market prices the overperformance of family ﬁrms. As it turns out, not much (see columns 5 and 6). The diﬀerence in market to book ratios between family and non family ﬁrms is not statistically signiﬁcant, and economically small (0.08 for a sample standard deviation of 0.7). This result does, however, conceal some heterogeneity between family ﬁrms. Founder-managed ﬁrm have higher Market to Book ratio than widely held and other family ﬁrms: their MB ratio is 0.15 above widely held companies, and signiﬁcantly so. Family ﬁrms managed by a descendant of the founder or outside CEO do as well as the benchmark, neither better nor worse. This result stands in sharp contrast with our robust ﬁndings from accounting measures of performance. A potential reason for this insigniﬁcant diﬀerence may be that family ﬁrms tend to pay less dividends. One reason why this should be the case is that families seek to keep more internal funds to fund their pet projects (the

Performance and Behavior of Family Firms

13

expropriation hypothesis).14 We thus run in columns 7-8 similar regressions using the ratio of dividend to earnings as a dependent variable (deﬁned only when corporate pre-tax proﬁts are positive). The pay-out ratio is indeed signiﬁcantly lower by almost 6 points for all family ﬁrms taken together. This is economically sizeable given that the sample average pay-out ratio is 20 points. When we look at all three subcategories of family ﬁrms separately, we see that they all tend to pay a signiﬁcantly lower proportion of their proﬁts as dividends. The extent to which they do so is similar (a formal F test cannot reject equality), but it seems that founders tend to pay out less than other types of family ﬁrms. This may be due to growth opportunities, but when we include sales growth as a control, this diﬀerence remains unchanged.15 Another possible reason is the diﬀerence in how returns covary with the market return, i.e. that family ﬁrms have higher betas. It is often argued that family ﬁrms have a “long term” management policy. Such a view would state that, compared to non family ﬁrms, family companies invest less in booms, more in recessions and, for instance, commit to job preservation, such that they hoard labor in bad times, and hire less in good times. Therefore, the amount of money distributed to shareholders of family ﬁrms would be lower in downturns, and larger in upturns, implying a larger beta for family ﬁrms. Because they pay more when other assets have large returns, they are less valuable, which depresses the market to book ratio of family ﬁrms. Using our monthly stock returns data, we estimated, on the 1991-2002 period, betas for ﬁrms which do not change family status over the 1994-2000 period. We then regressed these estimated betas on family status, controlling for size, age, industry and book leverage (results available from the authors). Apart from founders, who tend to have higher betas, other family ﬁrms do not show systematic diﬀerences with widely held corporations. Diﬀerences in betas are, apparently, not the explanation to the low valuation of family ﬁrms. A last, more daring, explanation for this discrepancy between proﬁtability and stock market valuation could be that the market has been mistakenly punishing family ﬁrms over the period. This would be consistent with the stock market returns evidence by Van Der Heyden et alii (2004) on the largest listed ﬁrms: he ﬁnds excess returns for a buy-and-hold portfolio of family stocks as large as 10% over the 1994-2000 period. So either the market has misunderstood the potential held by family ﬁrms at the time, or it overestimated future returns of non family ﬁrms, many of them recently privatized by the government. Given that Van der Heyden does not use the same breakdown as we use, nor the same sample, this remains, however, a conjecture.
14 Since 1967, the French tax system is a priori neutral with respect to dividend taxation. A complex system of tax credit makes the tax rate on corporate proﬁts equal to income tax for shareholders. Also, capital gains are taxed like income. So there is no obvious ﬁscal reason for which family ﬁrms would want not to pay dividends to themselves. 15 Regressions not reported here but available from the authors.

Performance and Behavior of Family Firms

14

1.3.3

Discussion on Endogeneity Biases

The cross sectional evidence presented above, though robust and compelling, cannot be interpreted as evidence of a causal eﬀect of family ownership/management on performance. A ﬁrst, obvious, reason is that family status depends itself on performance. The performance of professionally managed ﬁrms – be they widely held or family-controlled – could be underestimated in a cross section if it were easier to transfer corporate control to a descendant when the ﬁrm performs well. This would create a simultaneity bias. To address this concern, we look at ﬁrms who are transmitted to descendants, two years before this transmission. Due to the limited time frame of our panel, we only ﬁnd 30 such events. Nevertheless, these ﬁrms do not outperform their industry prior to transmission (see Table 1.4). Thus, descendant managers do not seem to inherit the best ﬁrms in our sample. Then, we focus on family ﬁrms that were transmitted to professional managers (21 events). They tend to slightly, but not signiﬁcantly, underperform their industry benchmark prior to transmission (see table 1.4). Thus, it does not appear obvious that only the best ﬁrms remain managed by the family, although the number of transitions we base our analysis on is too small to obtain a suﬃcient statistical power. A more straightforward way to control for ﬁrm unobserved heterogeneity and its possible correlation with family status would be to look at the change in performance when the ﬁrm is transmitted to an heir and when it is transmitted to a professional manager, and to compare the diﬀerence in performance changes. This is the approach taken by P´rez-Gonz´lez (2006) in his study on US ﬁrms. In our sample, it turns out that both heirs and professional e a CEOs tend to reduce the ﬁrm’s ROA to the same extent (around -.01, as is obvious from Table 1.4). Consequently, the diﬀerence in diﬀerence estimator of the eﬀect of heir management upon ﬁrm performance is nearly zero, and statistically insigniﬁcant. Once again, the number of transitions is likely to be too small to make realistic statistical statements. A second source of upward bias is endogenous sample selection. Assume for example that heir-controlled ﬁrms who do badly have a higher tendency to go bankrupt, or to be sold out to a large group or private equity investors. In this case, the only heir-managed ﬁrms who would survive would be those who do relatively well, which would lead us to overestimate their performance. To check if this is the case, we look at the proﬁtability of all types of ﬁrms prior to de-listing. From 1994 to 1999, we observe 142 de-listings in our data: 25 founder-controlled, 26 heir-controlled and 22 professionally managed family ﬁrms de-listed over the period. Prior to de-listing, exiting ﬁrms have in general a level of proﬁtability very similar to that of remaining ﬁrms. The only sizeable diﬀerence comes from heirs: staying heirs under perform those who go private by 3 percentage of industry adjusted ROA. This

Performance and Behavior of Family Firms

15

Table 1.4: Management Transitions
Industry adjusted ROA Before transition After transition Change in adj. ROA

Firms transmitted to Heir CEO Mean Student’s t Number of observations -0.00 (.01) 30 -0.01 (.01) 30 -0.01 (0.61) 30

Firms transmitted to Professional CEO Mean Student’s t Number of observations -0.03 (.02) 21 -0.04 (.02) 21 -0.01 (0.44) 21

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details on data construction and sources. Note: this table displays the evolution of industry-adjusted performance for family ﬁrms whose control is transmitted to heir or professional CEO. “Before transition” represents two years before the transition, “After transition” stands for two years after the transition.

Performance and Behavior of Family Firms

16

is economically signiﬁcant and almost statistically so. This suggests that endogenous attrition, if anything, leads to underestimating, rather than overestimating, the performance of heir-managed ﬁrms.16

1.4

Management Styles in Family Firms

This section seeks to explain how diﬀerent types of family ﬁrms achieve better performance. We start with a simple breakdown of proﬁtability (ROA) that allows us to attribute diﬀerences in ROA to (1) productivity, (2) wage or (3) capital intensity diﬀerences. We ﬁnd that founders overperform because they are more productive, heirs because they pay lower wages. Professionals seem to operate with less capital. We then conﬁrm these ﬁndings using other sources of data. Even after controlling for skill structure, it turns out that heirs pay lower wages. This may be due to heirs being able to insure their workers against adverse industry shocks. Indeed, they have, on average, a longer horizon than professional CEOs. Moreover, we also ﬁnd that sales and employment in heir-managed ﬁrms adjust less to industry shocks. Professional managers, in contrast, are better in ﬁnance: they pay lower interest rates on their debt, and make acquisitions that are more proﬁtable in the long run.

1.4.1

Breaking down Corporate Performance

This section seeks to explain the cross-sectional diﬀerences in proﬁtability shown in Table 1.3. To shed light on the determinants of proﬁtability, we ﬁrst use the following decomposition of ROA: ROA = (L/A) × (Y /L − w) (1.2)

L/A stands for labor intensity and is measured as the ratio of the number of employees to book value of total assets. Y /L stands for labor productivity and is measured as the ratio of value added (i.e. total sales less non-labor costs of inputs) to the number of employees. Finally, w represents the average wage paid to employees and is measured as total labor costs divided by the number of employees. Unsurprisingly, ﬁrms are more proﬁtable when other things equal, (1) their production process uses less capital, (2) labor productivity is higher and (3) wages are lower. Of course, all these variables are jointly determined: capital intensity depends on the relative prices of labor and capital, labor productivity depends on organization, on the amount of capital per workers, and on the skill composition of the workforce. Finally, w is the outcome of a bargaining process involving capitalists and workers,
16 A similar bias could be that the exchange authorities require a better performance - or a more transparent governance - from family ﬁrms when they want to go public. Hence, entry in our sample would induce an upward selection bias: only the best family ﬁrms are listed. We looked at the ﬁrst-year-of-listing proﬁtability of heir-managed ﬁrms, compared to an industry benchmark. It was not any diﬀerent from the ﬁrst performance of other categories.

Performance and Behavior of Family Firms

17

both of them considering their outside options on the capital and labor markets respectively, but also corporate performance as a whole. Therefore, we are not attempting here to perform a structural estimation of the behavior of family ﬁrms, but simply taking a ﬁrst step at understanding the causes of family ﬁrms’ greater proﬁtability. We use equation (1.2) to break down the conditional diﬀerence in ROA between family and non family ﬁrms (exhibited in Table 1.3) into conditional diﬀerences in productivity, wage, and capital intensity. Simple algebra (see A.3 - equation A.2) shows that the unconditional diﬀerence in average ROA between family and non family ﬁrms can be re-written as: ∆ROA = ROAF − ROAN F = + (L/A)F ∆ (Y /L) unconditional diﬀ. in productivity

−

(L/A)F ∆w unconditional diﬀ. in wage

(1.3)

(Y /L − w)N F ∆ (L/A)

+ ∆ [cov ((L/A) , (Y /L) − w)] diﬀ. in covariance

unconditional diﬀ. in capital intensity

Thus, the unconditional diﬀerence in mean ROAs can be exactly re-written as the sum of four terms. In addition to the three obvious eﬀects (labor productivity, wage and labor intensity diﬀerences), we need to include the diﬀerence in the covariance of L/A (i.e. labor to capital ratio) and Y /L − w (i.e. value added per worker minus average wage) for family and non family ﬁrms. This last term is a direct by-product of the non linearity of equation (1.2). Equation 1.3 provides us with a simple relation between the unconditional means of the diﬀerent ratios to explain the unconditional diﬀerence in performance. However, the diﬀerences in proﬁtability observed from Table 1.3 are computed conditional on observables. Moreover, we are interested in the contribution of conditional rather unconditional diﬀerences in labor productivity, wage and labor intensity to total performance, as we do not want to capture eﬀects stemming from diﬀerences in observables. To make the link between the estimates of Table 1.3 and these conditional diﬀerences, we show in A.3 that we need to include a ﬁfth term to equation 1.3, which indeed captures the eﬀect of diﬀerences in observable characteristics across types of ﬁrms. We then obtain a simple relationship between the performance coeﬃcients estimated in Table 1.3 and the conditional diﬀerences in labor productivity, wage and labor intensity. The detail of this derivation is given in A.3. Table 1.5 reports, for each family status, the contributions to corporate performance of all ﬁve eﬀects (productivity, wage, capital intensity, covariance, eﬀect of observables). The ﬁrst line looks at the components of the diﬀerence in ROA between founder-run ﬁrms and non family ﬁrms. As it turns out, the diﬀerence (2.6 percentage points out of 2.1) can be mostly explained by a diﬀerence in productivity. The second line looks at the spread between

Performance and Behavior of Family Firms

18

Table 1.5: Contributions to Firm Performance
ROA coming from... Founder CEO Heir CEO Professional CEO Higher L/K Ratio -.003 -.002 .015 0.4 2.7 3.6 Lower wage Higher Labor Productivity 2.6 -0.6 -4.7 Diﬀ. in Observables -0.2 -0.6 0.2 Low Covariance -0.7 1.0 1.4 2.1 2.2 1.7 TOTAL

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details and sources. Note: this table breaks down the diﬀerence in performance between family ﬁrms and widely held ﬁrms, using the decomposition presented in equation A.2 in A.3. Each column presents one of the coeﬃcients of equation A.2, for each family status.

heirs and widely held ﬁrms: in this case, the bulk of the eﬀect is accounted for by diﬀerences in wage levels (2.7 percentage points out of 2.2). Finally, the third line of Table 1.5 compares professional CEOs in family ﬁrms to non family ﬁrms. Here the picture is a little more complex: professional CEOs pay lower wage, but the beneﬁt to investors is more than compensated by a lower productivity. We will make sense of this pattern in the next section. In fact, professional CEOs achieve superior performance because they run their operations with lower capital to labor ratios, not because they pay lower wages. The broad picture emerging from Table 1.5 seems to indicate that (1) founders tend to display a larger productivity of labor than non family ﬁrms (2) both hired and descendants CEOs in family ﬁrms pay wages that are sizeably lower than widely held ﬁrms (3) the productivity of labor is much lower in ﬁrms run by professionals but (4) professionals compensate somehow by running operations with higher labor to capital ratios. These results are fully conﬁrmed both quantitatively and statistically, when we run separate regressions of labor productivity, average wage and capital intensity on family status and controls. These regression results are not reported to save space. In the following sections, we seek to go deeper in our understanding of these diﬀerences, by looking more precisely at both employment and investment policies.

1.4.2

Family Firms Pay Lower Wages

The main feature of Table 1.3 is that heir and professionally managed family ﬁrms pay wages that are lower by 10% than those paid by widely held ﬁrms. A potential explanation for these lower wages could be that family ﬁrms simply hire less skilled employees, but pay the market wage. Part of this eﬀect is likely to be captured by industry eﬀects, but there may be intra-industry variations in the skill structure of ﬁrms. To check this, we matched our

Performance and Behavior of Family Firms

19

dataset of listed ﬁrms with employer tax ﬁles which report, in theory, limited information on each employee, of each company (for a thorough description of these ﬁles, see Abowd, Kramarz and Margolis (1999)). In fact, the matching is far from perfect for three reasons. First, given that the French workforce has some 25 million employees, and our limitations of computing power, we use an extract of the whole database (4%). Secondly, many employees are likely to work for a subsidiary of the listed ﬁrm present in our data. We thus need to track ownership relation between various subsidiaries of a same group in order to “consolidate”employment and wages. We do this with a survey (LIFI, INSEE), which is, by design, far from being exhaustive below a given threshold, in particular for new ﬁrms. Third, the data was available only until 1998 included. Thus, when the information is available, the employer tax ﬁles provide us, for each ﬁrm, with the average annual wage and measures of the skill structure that would normally take place in individual wage regressions. We use: the share of male employees, their mean seniority and age, and ﬁnally the fractions of managers, supervisors, skilled employees and unskilled workers. We then regress this new measure of mean wage, at the ﬁrm level, on our family variable, on the ﬁrm level controls of Table 1.3 and on these additional controls of skill structure. Table 1.6 reports the estimates of such regressions. In column 1, we simply use aggregate wage bill and employment ﬁgures from our accounting data and report an estimated wage discount in both professionally and heirmanaged family ﬁrms of around 10%. In column 2, we run the same regression, using the average wage from the tax ﬁles instead of accounting data. We obtain similar estimates for descendant and professionally managed ﬁrms (10%), but a lower estimate for founder managed ﬁrms. The reason for such a discrepancy comes from a diﬀerent size-wage relation in founder ﬁrms : in unreported regressions,17 we see that wages in large founder ﬁrms are signiﬁcantly lower than in small founder ﬁrms, whereas such a relation does not hold for other types of ﬁrms. Considering the fact that the DADS ﬁles over-represent the importance of large ﬁrms,18 this explains why the founder coeﬃcient in column 2 diﬀers from the one in column 1. In columns 3 and 4, we include the skill structure controls progressively. As is apparent from these two columns, the wage discount of professionally managed ﬁrms progressively vanishes, which suggests that family ﬁrms run by outside CEOs pay lower wages mostly because they have younger and less skilled workers. The discount remains, however, signiﬁcantly diﬀerent from zero for heir-managed ﬁrms. It thus seems that descendants manage to pay wages lower by 4%-5%, even after controlling for the ﬁrm skill structure. In column 5, we run the same regressions as in column 4, except that we now weight observations by the number of workers retrieved in the tax ﬁles. The reason regressions are available from the authors upon request. over-representation of large ﬁrms in the DADS sample comes from the LIFI ﬁles (see Appendix A.1.3), which mostly tracks ownership for large ﬁrms.
18 This 17 These

Performance and Behavior of Family Firms

20

Table 1.6: Wages in Family Firms: Accounting for Skill Structure log(wage bill / empl.) (×100) Acc. Data (1) Founder CEO -0.3 (4.8) Heir CEO -10.0∗∗ (4.5) Professional CEO -9.4∗∗ (4.7) Fraction of managers (2) -10.6∗∗∗ (4.2) -14.0∗∗∗ (3.7) -10.3∗∗∗ (4.0) (3) -6.4∗∗∗ (2.3) -5.2∗∗∗ (2.0) -4.1∗ (2.3) 110.8∗∗∗ (7.6) Fraction of supervisors 35.3∗∗∗ (5.8) Fraction of skilled empl. -7.9 (4.9) Mean age Employer Tax ﬁles (4) -2.8 (2.2) -4.3∗∗ (1.9) -2.5 (2.2) 107.7∗∗∗ (6.1) 36.2∗∗∗ (5.6) 4.0 (5.0) 1.2∗∗∗ (0.2) Mean seniority 0.5∗∗ (0.2) Fraction of male employees Log(Assets) 1.72∗ (1.00) Log(Firm Age) -1.35 (1.80) Former SOE 13.7∗∗∗ (5.64) Industry FE Year FE Observations Adj R2 yes yes 1,351 .29 1.88∗∗∗ (.77) .16 (1.32) 5.45 (4.04) yes yes 1,427 .25 2.12∗∗∗ (.43) 2.78∗∗∗ (.94) 4.64∗∗ (2.37) yes yes 1,427 .64 8.4∗∗ (4.2) 1.52∗∗∗ (.42) .93 (.94) 2.39 (2.09) yes yes 1,427 .68 (5) -7.8∗∗∗ (2.4) -3.8∗∗ (1.6) -1.9 (1.6) 111.7∗∗∗ (5.4) 47.6∗∗∗ (7.1) 14.9∗∗∗ (5.7) 0.2 (0.2) 1.9∗∗∗ (0.3) 2.0 (5.7) -.14 (.35) -.04 (.84) 4.24∗∗∗ (1.53) yes yes 1,427 .85

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details on data construction and sources. Note: Huber-White-Sandwich estimates allowing for correlation of all observations of a given ﬁrm. Column 1 regresses, on the sub-sample for which the employer tax ﬁles are available, the log of the average wage as measured with the DAFSA Yearbooks on family status and various controls (i.e. Log(Asset), Log(Firm Age) and Former SOE deﬁned in table 1.3). Columns 2 performs the same regression using the measure of wage given by the employer tax ﬁles. Column 3 adds variables controlling for the skill structure of the workforce (fraction of managers, fraction of supervisors, fraction of skilled employees). Column 4 controls additionally for mean age, mean seniority and the fraction of male employees. Finally, Column 5 weights observations by the number of workers retrieved in the tax ﬁles. All the regressions control for 13 industry ﬁxed eﬀects as well as year ﬁxed eﬀects. Standard Errors are in parenthesis. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

Performance and Behavior of Family Firms

21

why we do this is that the average wage is more precisely estimated when more workers are tracked using the tax ﬁles. As it turns out, the signiﬁcant wage discount in heir-managed ﬁrms sustains. As one can also notice, the coeﬃcient on founder managed ﬁrms becomes signiﬁcantly lower than it was in column 4. This is not surprising since, as we mentioned earlier, large founder-managed ﬁrms pay signiﬁcantly lower wages than small founder-managed ﬁrms. That outside CEOs in family ﬁrms hire lower skilled workers may explain why, in these ﬁrms, labor productivity is much lower than in widely held corporations and other family ﬁrms (as is apparent from Table 1.5). Another possibility is that professional CEOs substitute unskilled labor to capital, to make invested capital more proﬁtable. It is indeed obvious from Table 1.5 that outside CEOs appear to have a much lower capital to labor ratio than that of other types of ﬁrms. Such evidence is consistent with professional CEOs in family ﬁrms making a more parsimonious use of capital than descendants or widely held companies. This preliminary analysis suggests that family ﬁrms may be achieving higher proﬁts by two diﬀerent means: (1) descendants manage to pay lower wages for similar skill and productivity and (2) outside CEOs make a more parsimonious use of capital. We provide further evidence consistent with these two hypotheses in the following sections.

1.4.3

Descendants Can Commit on Long Term Employment

How do descendant CEOs succeed in paying lower wages, without recruiting low skill workers, and still obtain a high level of labor productivity? We explore here a lead inspired from Shleifer and Summers (1988): dynastic management endows the family with enough credibility to enforce implicit contracts. Under implicit labor contracts, the ﬁrm promises that most workers will keep their jobs even if total sales decrease. The ﬁrm thus provides employment insurance to its employees. In exchange for this, workers accept a lower wage, or to work harder for the same wage. Since the employee is risk averse - his labor supply is not diversiﬁed - and the ﬁrm is risk neutral - in the absence of credit constraints, the arrangement is ex ante value creating. The problem with this theory is that usually, ﬁrms are not credible when making such promises. Their incentive to renegotiate ex post is too strong, in particular when the ﬁrm can easily be taken over by a management which is bound by such a commitment (Shleifer and Summers (1988)). Families might have an advantage in enforcing this type of contract. First, they have a longer horizon than salaried managers: dynastic management can therefore create value that would be destroyed – both ex ante and ex post – by delegated management. Second, provided the family is involved in management, a culture irrationally tying top management to employees might prevent job losses in bad time, even if they were dynamically optimal. While this destroys ex post proﬁt, it creates value ex ante (Kreps (1990)). Third, since families own the ﬁrm,

Performance and Behavior of Family Firms

22

they may be able to commit without fear of being taken over ex post. Professional managers who are not owners completely lack this ability to commit. We test this by looking at the sensitivity of ﬁrm employment to industry sales shocks. A possible concern with this approach is that our sample period is short (1994-2000). It contains only one cycle, with 1994-1996 being downturn years, and 1997-2000 being expansion years. Therefore, industry level sales shocks, because they are partly determined by aggregate shocks, will capture the upward trend of the economy over the period. If family and non family ﬁrms turn out to follow diﬀerent trends of growth over this short period, we might attribute this movement to diﬀerent responses in sales shocks. To avoid this, we control for aggregate shocks, and allow ﬁrms to vary in their responses to economy wide shocks. More precisely, we estimate the following model: log Yit = αi + β (Xit ) log salesst + γ (Xit ) δt + εit (1.4)

where Yit stands for ﬁrm i’s total employment or sales at date t. αi is a ﬁrm ﬁxed eﬀect. log salesst is the log of total sales of the industry the ﬁrm i is in. δt is a year dummy indicating economy wide sales shocks.19 β(Xit ) and γ(Xit ) are elasticities to industry and economy wide shocks, which are supposed to depend on ﬁrms observables. We posit: β (Xit ) = a + b.Fi + c. log ageit + d.SOEi γ (Xit ) = a + b .Fi + c . log ageit + d .SOEi where Fi is the set of our family dummies, ageit is the ﬁrm’s age and SOE equals 1 when the ﬁrm has been state-owned. As recalled above, because log salesst partly depends on the overall state of the economy, it may well be that log salesst and δt are correlated. If we omit γ (Xit ) .δt in equation (1.4), we may capture a part of γ (Xit ) in the estimate of the sensitivity of employment to shocks (the βs). If for some other reason, family ﬁrms have grown faster over the late 1990s, and therefore have a larger γ, then the estimate of β for family ﬁrms will be upward biased. This is why we control for aggregate shocks.
19 We choose not to run directly a diﬀerence on diﬀerence equation because the ﬁxed eﬀect speciﬁcation allows us to be much more agnostic on the timing of response of employment growth to sales growth. Assume for example that our model is slightly mis-speciﬁed in the following way: employment does not react to contemporary sales, but to sales lagged by one year. In this case, the ﬁxed eﬀect estimate is going to capture most of the eﬀect by comparing the ﬁrm’s average employment before and after the sales shock. In contrast to this, the diﬀerence estimate is not going to see any correlation given that in the very year sales change, employment remains ﬁxed. Hence, while we prefer the ﬁxed eﬀect estimate of equation (1.4), it must be clear that what we have in mind is the response of employment changes to industry shocks.

Performance and Behavior of Family Firms

23

Table 1.7: Do Family Firms Smooth Employment Shocks?
Dependant variable Log(salesit ) (1) Log(salesst ) 0.20 (0.15) Log(salesst ) × Founder CEO -0.05 (0.08) Log(salesst )× Heir CEO -0.17∗∗ (0.07) Log(salesst ) × Professional CEO 0.04 (0.11) Log(salesst )× Former SOE -0.18∗∗ (0.09) Log(salesst ) × Log(Age) 0.00 (0.03) (2) 0.17 (0.16) -0.08 (0.08) -0.21∗∗∗ (0.07) 0.04 (0.12) -0.21∗∗ (0.09) 0.01 (0.04) Log(employmentit ) (3) 0.36∗ (0.20) -0.12 (0.13) -0.22∗∗ (0.11) 0.07 (0.14) -0.22∗∗ (0.11) -0.05 (0.03) (4) 0.34∗ (0.21) -0.20 (0.13) -0.27∗∗ (0.11) 0.01 (0.13) -0.16 (0.11) -0.03 (0.04)

Firm FE Year FE Year FE × Founder Year FE × Heir Year FE × Professional Year FE × SOE Year FE × Log(age)

yes yes no no no no no

yes yes yes yes yes yes yes

yes yes no no no no no

yes yes yes yes yes yes yes

Test Heir=Professional

.04∗∗

.04∗∗

.01∗∗∗

.01∗∗∗

Observations Adj. R2

1,977 .97

1,977 .98

1,898 .97

1,898 .97

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details on data construction and sources. Note: Huber-White-Sandwich estimates allowing for correlation of all observations of a given ﬁrm. Dependent variables are log of sales (column 1 and 2) and log of employment (column 3 to 4). Log(sales,t ) is the log of average sales in industry s at date t. Column 1 and 3 estimate equation (4) assuming that c = d = 0. Column 2 and 4 relax all the constraints. All regressions control for ﬁrm ﬁxed eﬀects as well as industry ﬁxed eﬀects. Line “Heir=Professional” provides the p-value of a test of equality between the coeﬃcients “Heir CEO” and “Professional CEO”. Standard errors are in parenthesis. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

Performance and Behavior of Family Firms

24

Estimation results are shown in Table 1.7. Columns 1 an 2 study the response of ﬁrms’ sales to industry shocks, while columns 3 and 4 look at employment. For industry s at date t, log salesst uses average sales for all companies in the industry, except ﬁrm i. We take average sales, instead of total sales, in order to account for attrition: if an average ﬁrm in the industry de-lists, our measure of industry sales will not be aﬀected. Finally, we restrict ourselves to industries where at least 20 ﬁrms are present, to have a precise estimate of average sales. Columns 1 and 3 assume that c = d = 0, and columns 2 and 4 relax these constraints. A look at columns 2 and 4 shows that indeed, employment reacts less to industry shocks in heir-managed ﬁrms. The result is not present unless we control for ﬁrms’ characteristics and especially the “Former SOE” dummy. This is not very surprising as there are reasons to believe that former SOEs also exhibit this pattern of labor hoarding and less volatile activity. Since most widely held ﬁrms are former SOEs, not including a control for Former SOE creates a composition eﬀect that brings the reference group (widely held ﬁrms) artiﬁcially close to the group of interest (heir managed ﬁrms). Heirs may be able to smooth out employment over the industry cycle, either (1) by choosing less risky projects or (2) by modifying their own mark-up across the cycle. Columns 1-2 suggest that the ﬁrst explanation might be true, as ﬁrms’ sales are much less sensitive - not at all, it turns out - to industry shocks in heir-managed ﬁrms. In fact, if we use ﬁrms’ proﬁtability as a dependent variable (Yit ), we ﬁnd - in non reported regressions - that ﬁrms’ proﬁtability is not more sensitive to industry shocks in family ﬁrms managed by a descendant. Results from Table 1.7 could also be explained by the fact that heir-managed ﬁrms tend to operate in “niches”that are relatively sheltered from competition. The argument is a selection eﬀect. Founders start ﬁrms in any kind of industry, but their descendants have lower than average managerial ability. If product market competition is soft, descendants can continue to run the ﬁrm. If the environment is competitive, descendants, with their lower than average managerial ability, cannot survive. They have to hire professional managers, sell their ﬁrms altogether to public or private investors. Because they self select into niche markets, descendants thus run more stable ﬁrms, both in terms of employment and sales. Because competition is softer, the ﬁrms they run are more proﬁtable. This view has two consequences: (1) heir-managed ﬁrms should be less present in competitive industries and (2) heir-managed ﬁrms should underperform in more competitive industries. We look at these two empirical implications. First, we investigate whether family, in particular heir-managed, ﬁrms tend to operate in less competitive industries. We measure competition by computing an index of sales concentration at the industry level (an Herﬁndahl index). Due to data limitations, this measure is very crude as it uses a rough industry classiﬁcation (14 categories) and sales of listed ﬁrms. Using this measure, we found that over

Performance and Behavior of Family Firms

25

Table 1.8: Performance of Family Firms and Competition
Return on assets (×100) Competition above median Competition below median

Family Firm

1.6** (1.9)

2.1** (2.4) 1.4 (1.5) 2.3* (1.8) 1.8* (1.7) 2.2** (2.1)

Founder CEO

-

Heir CEO

-

2** (2.1)

Professional CEO

-

1.2 (1.1)

Observations Adj.R2

1,235 .18

1,235 .18

1,090 .25

1,090 .25

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details on data construction and sources. Note: Huber-White-Sandwich estimates, allowing for correlation of all observations of a given ﬁrm. Dependent variable is ROA (ratio of EBIT to book value of asset). The speciﬁcation used in this table is similar to Table 1.3. Column 1 and 2 are estimated on industries with a 1994 herﬁndahl below median (competitive). Column 3 and 4 are estimated on industries with a 1994 herﬁndahl above median (non competitive). Standard Errors are in parenthesis. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

the 1994-2000 period, 32% of non family ﬁrms are in industries where competition is above median. The fraction raises to 52% for descendant-run family ﬁrms, and 51% for those run by a hired CEO. Using our crude measure of competition, there is slight evidence that descendants tend to operate more in competitive industries. We then test whether heirs do worse in competitive industries, by comparing the heir eﬀect in performance regression in competitive and in non competitive industries. In Table 1.8, we present the ROA regressions on family status, as speciﬁed in Table 1.7, columns (1) and (2). We split our sample into two parts: (1) industries whose 1994 herﬁndahl is below median (competitive) and (2) industries whose 1994 herﬁndahl is above median (non competitive). We ﬁnd no diﬀerence: whatever the degree of competition, heirs outperform their competitors by approximately the same margin. All in all, the data provide little support for the fact that descendants only survive

Performance and Behavior of Family Firms

26

in “niche” markets, and that this relative protection from competition explains their performance. Both pieces of evidence should, however, be interpreted with caution, given the noisiness of our competition measure. Although imperfect, these tests lend further credence to the ﬁrst, slightly diﬀerent, interpretation of Table 1.7: descendants “smooth out”industry shocks as part of their labor management policy. One possible reason why they could commit to long term labor contracts is that they, as managers, have a much longer horizon than professional CEOs. To see this, we look at CEO turnover, and ask if it is lower in heir-managed family ﬁrms. We measure CEO turnover as a dummy equal to 1 if the current CEO does not run the company in the coming year. We then regress it on our family status variables and on usual determinants of CEO turnover such as corporate performance (measured as ROA or annual stock return), ownership concentration, ﬁrm size, age, year and industry dummies (for a typical study of CEO turnover see for example Weisbach (1988)). We also add, as their governance is likely to be diﬀerent, a dummy for former SOEs. Linear regression results are reported in Table 1.9. Column 1 simply compares CEO turnover in family and non family ﬁrms, accounting for year and industry ﬁxed eﬀects. As it turns out, CEO turnover is much lower when the family is still in the management. In founder and heir-managed ﬁrms, the probability of changing CEO is lower by some 9 percentage points than in widely held ﬁrms. This is a huge diﬀerence, given that the mean probability of CEO turnover is equal to .10 in our sample. When an outsider runs the family business, its chances to leave the job are lower by only 3 percentage points than if it ran a widely held company. The diﬀerence is not signiﬁcant ; it is, however, signiﬁcant when we compare heirs and professional managers in family ﬁrms. In this simpliﬁed regression, we can reject with 95% conﬁdence that heirs and outside CEOs in family ﬁrms face the same probability of turnover. The diﬀerence, some 5 percentage points, corresponds to some 4 years of additional tenure. Finally, columns 2 and 3 then ask whether this signiﬁcant diﬀerence can be explained by diﬀering ﬁrm characteristics. Including ROA as a right hand side variable reduces the diﬀerence between heirs and professionals a bit, and renders its estimate noisier and insigniﬁcantly diﬀerent from zero. In this context, it is thus likely that slightly larger ROA within heir-managed ﬁrms (see Table 1.3) explains why CEO turnover is less frequent in these ﬁrms.

1.4.4

Outside CEOs Are More Financially Literate

We have seen previously that outside CEOs operate at lower ratios of capital to labor. We present here two further pieces of evidence consistent with the fact that professional managers make a more eﬃcient use of capital. The ﬁrst piece of evidence is related to the cost of debt. Using very clean data on bond issues, Anderson, Mansi and Reeb (2003) ﬁnd that, when compared to non family ﬁrms, the corporate yield spread on family ﬁrms is

Performance and Behavior of Family Firms

27

Table 1.9: CEO Turnover in Family Firms
Dependent variable (1) CEO Turnover (2) (3)

Founder CEO

-9.8∗∗∗ (1.9)

-8.7∗∗∗ (2.0) -7.2∗∗∗ (2.2) -4.4∗ (2.4) -0.5∗∗∗ (0.1)

-9.3∗∗∗ (2.5) -9.4∗∗∗ (2.5) -6.4∗∗ (2.7) -0.4∗∗∗ (0.1) 0.5 (0.5)

Heir CEO

-7.7∗∗∗ (2.1)

Professional CEO

-3.0 (2.6)

ROA

-

Log(Assets)

-

-

Log(Firm Age).

-

-

1.7 (1.1)

Former SOE

-

-

-4.1 (2.9)

Fraction equity of largest block Industry FE Year FE

-

-

5.8 (4.0)

Yes Yes

Yes Yes

Yes Yes

Test Heir = Professional (p value)

0.05∗∗

0.22

0.22

Observations

2,208

1,930

1,795

Source: panel of French listed ﬁrms over the 1994-2000 period. See A.1 for details on data construction and sources. Note: Huber-White-Sandwich estimates, allowing for correlations of all observations of a given ﬁrm. CEO Turnover, the dependent variable, is a dummy equal to 1 if the CEO loses his position in the following year. Column 1 simply controls for the family status. Column 2 adds proﬁtability (ROA) as a control. Column 3 adds Log(Assets), Log(Firm Age), Former SOE and “Fraction equity of largest block” as additional controls. All regressions control for 13 industry and year ﬁxed eﬀects. Standard errors are in parenthesis. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

Performance and Behavior of Family Firms

28

Table 1.10: Interest Rate Paid by Family Firms
Average interest on debt (×100) (1) Family Firm -0.3 (0.6) Founder CEO -0.1 (0.6) Heir CEO 0.5 (0.8) Professional CEO -1.6∗∗ (0.7) Log(Assets) -0.5∗∗∗ (0.1) Log(Firm Age) -0.1 (0.3) Former SOE 0.1 (0.1) Fraction equity of largest block Debt / Assets -0.2 (1.1) -15.2 (1.4) Stock return volatility ROA 3.0∗ (1.6) -4.8 (5.6) Industry FE Year FE “Heir=Professional” Observations Adj. R2 2,200 .22 yes yes -0.4∗∗∗ (0.1) -0.1 (0.3) 0.1 (0.9) -0.1 (1.1) -15.3∗∗∗ (1.4) 2.7 (1.6) -5.3 (5.6) yes yes .004∗∗∗ 2,200 .23 (2) -

Source : panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details on data construction and sources. Note: Huber-WhiteSandwich estimates allowing for correlation of all observations for a given ﬁrm. The dependent variable is the average interest rate paid on debt. Column 1 and 2 control for Log(Assets), Log(Firm Age), Former SOE, fraction equity of largest block, leverage, stock return volatility and ROA (deﬁned in Table 1.3). Both regressions also control for 13 industry ﬁxed eﬀects as well as year ﬁxed eﬀects. Column 1 controls for family ownership (Family ﬁrm) while column 2 controls for family management status (Founder CEO, Heir CEO, Professional CEO). Line “Heir=Professional” gives the p-value of an equality test between the coeﬃcients “Heir CEO” and “Professional CEO”. Standard errors are in parenthesis. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

Performance and Behavior of Family Firms

29

consistently 30-40 basis points lower. They argue that, as family ﬁrms are long term shareholders, they can commit more easily not to default, which reduces their risk premium. We run similar regressions to theirs, and present the results in Table 1.10. Our measure of the cost of debt is, however, much noisier: since we do not have data on the bond yield spreads, we have to content ourselves with the ratio of interest paid to ﬁnancial debt. This measure should be approximately equal to the average of all spreads on all loans and bonds, weighted by the sizes of the various issues. We then regress this average cost of debt on the same controls as Table 1.3, plus the ﬁrm’s current proﬁtability as measured by ROA. In our sample, we ﬁnd that the average interest rate paid by family ﬁrms is on average lower by 30 basis points, albeit not signiﬁcantly so. Although imprecise, the order of magnitude is consistent with ﬁndings of Anderson, Mansi and Reeb (2003)’s study.20 When we look at the various subcategories of family ﬁrms, we ﬁnd that those run by professional managers are the ones who pay signiﬁcantly lower interest rates, by a huge 160 basis points. This is consistent with both the ability to commit of long term shareholders and the eﬃcient ﬁnancial management of professional CEOs. Secondly, we look at the eﬃciency of a speciﬁc investment project: the acquisition of another ﬁrm. First, acquisitions are interesting because their proﬁtability is relatively easy to evaluate. Most of the ﬁnance literature measures the proﬁtability of a deal as the returns for long run shareholders of the bidder. These long run returns (3-5 years after deal completion) are then adjusted for risk using diﬀerent models of expected returns (multi factor models, benchmark ﬁrms, industry portfolios). In the past decade, this literature has made large strides in identifying the various biases that arise in such long run studies (for examples of recent contributions, see Lyon, Barber and Tsai (1999) and Mitchell and Staﬀord (2000)). We will thus be able to build on this literature to compare the eﬃciency of acquisition policies of family and non family ﬁrms. The second reason why acquisitions are interesting is the large variability of their long run performance. Over the long run, acquisitions are on average value destroying in the 1980s (Rau and Vermaelen (1998)) and in the 1990s (Moeller, Schlingemann and Stulz (2005)). However, these negative long run stock returns are mostly due to large acquisitions (Moeller, Schlingemann and Stulz (2005)), to friendly deals (Rau and Vermaelen) and to operations ﬁnanced with share issues (Loughran and Vijh (1997)). Hostile bids, small acquisitions ﬁnanced with cash are in general followed by positive long run returns. These results suggest that some acquisitions - the large, friendly ones, ﬁnanced with stock - are simply evidence of uncontrolled managerial hubris. Firms where corporate governance is
20 A careful reader will notice that, in Table 1.10, the sign on leverage is negative. Anderson, Mansi and Reeb (2003), who run similar regressions ﬁnd a positive correlation with leverage. Theoretically, both directions of correlation are possible: highly levered ﬁrms may have a higher cost of debt because of costs of ﬁnancial distress, but low interest rate ﬁrms may wish to take on more debt (us). As it turns out, the diﬀerence between both studies can be traced back to the diﬀerence in the measure of interest rate. Using the same measure as ours on COMPUSTAT data, we found a negative and robust correlation between leverage and interest.

Performance and Behavior of Family Firms

30

poor, shareholders passive and managers all-powerful engage in these spectacular deals to build their CEO’s empire. In contrast, ﬁrms with sound governance, large shareholders and proﬁt maximizing managers may engage in small deals or hostile bids that on average increase shareholder value. We thus ask whether family ﬁrms make better acquisitions than non family ﬁrms, by comparing the risk adjusted stock returns of bidders from 0 to 4 years after the deal completion. To do this, we match our dataset with data on monthly stock returns from SDC platinium (see Appendix A.1.4 for further description of this source). Before comparing the gains from acquisitions, we started by regressing the number of acquisitions made by each ﬁrm on our family dummies, along with year and industry controls, and ﬁnd that descendant-managed family ﬁrms indeed make signiﬁcantly less acquisitions. This eﬀect does, however, vanish once we start controlling for ﬁrm size (using log assets). Founders, who tend to run very small ﬁrms, become the only ones to make more acquisitions than the other categories of ﬁrms. One possibility is that founders make series of value enhancing, small acquisitions. As we will see below, this is not the case. We then compare the eﬃciency of acquisitions policies for each type of ﬁrm. First, we take a “naive approach”. For each date between 1 and 36 months after an acquisition, we compute the buy and hold return of holding the acquirer’s stock from deal completion until that date. We then adjust this return for risk, by subtracting the bidder’s expected return. Our model for expected returns is taken from Fama and French (1996). For each ﬁrm, we regress monthly stock returns on the time series of market return, SMB (the monthly excess return of small ﬁrms) and HML (the monthly excess return of value ﬁrms). We subtract the monthly rate on 10-year French treasury bills from stock returns and the market returns. These three factors proxy for risk dimensions that investors seem to value. The residual of this regression - the part of monthly return that is not explained by these three factors - is our measure of abnormal return. To estimate our model of expected returns, we use the pre-acquisition period as our estimation period, and then compute residuals on the post period. Given that we require at least 12 months of data before the transaction to estimate our models, we end up with 845 acquisitions for which we can compute abnormal returns in the month of acquisition.21 After that, given natural attrition and right censorship in the panel (our returns data stop in October 2003), we can follow returns for only 595 transactions after 36 months. Using our measure of abnormal returns, we then compute, for each deal, the ﬁrm’s cumulative abnormal returns from 0 month after the acquisition is completed, until 36 months after. Although long run performances of acquisition are highly heterogenous, the vast majority of US studies ﬁnd that, on average, cumulative abnormal returns are negative in the long run. The same is true for our French sample, where the cumulative abnormal return of buying
21 All types of ﬁrms are well represented: 262 acquisitions have been performed by widely held corporations, 191 by founders, 71 by heirs and 125 by professional CEOs in family ﬁrms.

Performance and Behavior of Family Firms

31

Figure 1.1: Post-Acquisition Abnormal Returns of Acquiring Firms

Non Family Firms
10 20 10 20

Family Firms: Founder CEO

-40 -30 -20 -10 0

0

6

12

18

24

30

36

-40 -30 -20 -10 0 0

6

12

18

24

30

36

Family Firms: Heir CEO
-40 -30 -20 -10 0 10 20 -40 -30 -20 -10 0 10 20

Family Firms: Outside CEO

0

6

12 18 24 30 Months Since Acquisition

36

0

6

12 18 24 30 Months Since Acquisition

36

and holding a bidder’s stock is, on average, -7%. Figure 1 plots average cumulative returns to long-term shareholders who bought the acquirer’s stock 12 months before the deal, until 36 months after the acquisition. 90% conﬁdence bands are drawn with dotted lines. We break down acquisitions into those made by non family ﬁrms, and those made by our three subcategories of family ﬁrms.22 As it turns out, long-term shareholders of family ﬁrms run by professionals do not lose, nor gain, any wealth. Acquisitions made by widely held companies are slightly value destroying, with long-run buy and hold returns averaging -5% after 2-3 years. Then come founder-managed ﬁrms, whose acquisitions seem to destroy 15% after 2 to 3 years. The worse acquisitions are performed by descendants, whose long term shareholders lose out a large statistically signiﬁcant 20% after 2 to 3 years. Such evidence suggests that, within family ﬁrms, professional CEOs are better at making acquisitions than founders or their descendants. We provide the values of mean cumulative returns and formal t-tests in Table 1.11.
22 Family status of the acquirer for each acquisition is deﬁned at the time of acquisition. If, for instance, an acquisition is disclosed when a descendant is in command, but the ﬁrm becomes widely held the year after, the acquisition still counts as “heir-managed”.

Performance and Behavior of Family Firms

32

Table 1.11: Abnormal Returns to Long Run Shareholders of Acquirers
Widely Held Firms Family ﬁrms, managed by: Founder Months since acquisition 0 0.1 (0.2) +6 -1.2 (0.7) +12 -5.0∗∗∗ (2.5) +18 -3.5 (1.6) +24 -3.6 (1.2) +30 -2.4 (0.9) +36 -3.6 (1.2) 1.3 (1.5) 1.7 (0.7) 3.4 (1.0) -2.2 (0.6) -8.5∗∗ (2.1) -13.6∗∗∗ (3.0) -12.6∗∗∗ (2.4) 0.9 (1.1) -4.0 (1.3) -3.7 (0.8) -4.6 (0.8) -12.3∗∗ (2.1) -20.0∗∗∗ (3.5) -17.0∗∗ (2.3) 0.4 (0.4) 2.6 (1.3) 3.7 (1.3) -1.6 (0.5) 4.4 (1.0) 1.6 (0.3) 6.0 (0.8) 0.04∗∗ 0.01∗∗∗ 0.02∗∗ 0.32 0.17 0.06∗ 0.71 Heir Professional CEO T-Prob of Test heir=professional

Source: panel of French listed ﬁrms, over the 1994-2000 period. See A.1 for details on data construction and sources. Note: OLS estimates. The dependent variable is cumulative abnormal returns. Explanatory variables are periods after (or before) completion of the acquisition. Column 1 gives long term abnormal returns for widely held ﬁrms, column 2 for family ﬁrms with a founder CEO, column 3 for heir-managed family ﬁrms and column 4 for professionally managed family ﬁrms. Column 5 provides an equality test for the coeﬃcient of the regression in column 3 and 4. T-statistics are in parentheses. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

Performance and Behavior of Family Firms

33

As it turns out, post acquisition, the long run returns of founders and descendants are negative and statistically signiﬁcant. This is not the case for ﬁrms run by professional CEOs, be they widely held or family-owned. As a result, within family ﬁrms, the performance of acquisitions made by heirs is signiﬁcantly lower than that of hired CEOs. This conﬁrms our contention that professional CEOs use capital more parsimoniously. In the ﬁnance literature, the computation of mean long run, buy-and-hold, abnormal returns has, however, been criticized by Mitchell and Staﬀord (2000). They suggest that acquisitions are not independent events (they tend to occur in waves), and that, therefore, innovations on subsequent stock returns are correlated. As a result, standard t-statistics on mean abnormal returns tend to overestimate their precision, and lead to over-rejection of the null that abnormal returns are zero. In addition, the distribution of cumulative returns tend to be non normal (positively skewed), which invalidates standard t-statistics. As a result, Mitchell and Staﬀord advocate the use of monthly calendar time portfolios. Such portfolios have the twin advantage of (1) aggregating the cross sectional variance of returns, which reduces the correlation problem, and (2) being normally distributed, such that standard t-statistics are reliable. Before proceeding to the results, notice that this approach is generally thought to be conservative; for instance, Lyon, Barber and Tsai (1999), argue that it lacks power to detect long run returns. We compute the (equally weighted) monthly stock returns of portfolios made of bidders whose last acquisitions took place in the past 1,2,3 or 4 years. We then compute such portfolios for each type of family acquirers. The time window is January 1994 - October 2003. Last, to test whether heirs make worse acquisitions than professional CEOs in family ﬁrms, we compute the return of a portfolio that is long in professional CEOs who are past bidders, and short in descendant who are past bidders. These returns (minus risk free rate) are then regressed on the three Fama-French factors. Table 1.12 reports the (monthly) alphas of these portfolios: the ﬁrst line is for a portfolio containing all past bidders, the second line for all widely held bidders, the third line for founder bidders, the fourth line for all heir bidders and the ﬁfth line is for all professionally managed ﬁrms. The sixth line reports the alphas for portfolios long in professionals and short in heirs. As expected, this conservative method provides results similar to cumulative returns, albeit statistically weaker. Post acquisition returns for founders and descendants are still negative, but not signiﬁcant any more (some 1%-2% a year for founders, 3%-4% for heirs). Post acquisitions returns of professionally managed family ﬁrms are weakly positive. Most importantly, within family ﬁrms, acquisitions made by heirs signiﬁcantly underperform those made by professional CEOs. The “long in professional CEOs– short in heir CEOs” portfolio generate a large and signiﬁcant 8% annual abnormal return. The last comparative result from table 1.11 is therefore robust to the calendar time portfolio approach, in spite of its conservativeness.

Performance and Behavior of Family Firms

34

Table 1.12: Monthly Alphas on Calendar Time Portfolios of Acquirers
Years since last acquisition Portfolios of past bidders 0-1 year 0-2 years 0-3 years 0-4 years

Long all ﬁrms

-0.2 (0.6)

-0.1 (0.6 -0.2 (1.0) -0.2 (0.4) -0.2 (0.4) 0.1 (0.4) 0.4 (1.0)

-0.3 (1.2) -0.3 (1.4) -0.2 (0.4) -0.5 (1.4) 0.0 (0.0) 0.7∗∗ (2.0)

-0.3 (1.4) -0.3 (1.5) -0.1 (0.4) -0.6∗ (1.7) -0.1 (0.2) 0.7∗∗ (2.0)

Long non family ﬁrms

-0.4∗ (1.8)

Long Founder Firms

0.0 (0.0)

Long Heir CEOs

-0.3 (0.8)

Long Professional CEOs

0.2 (0.7)

Long Professional CEOs, Short Heir CEOs

0.9∗∗ (2.2)

Source: alphas on monthly calendar time portfolios whose returns are computed over 1994-2003. These portfolios are composed of all past bidders whose last transaction took place in the past 1, 2, 3 or 4 years. They are equal weighted and therefore, re-balanced every month. For each of these portfolios, monthly returns minus risk free rate are then regressed on the three Fama-French risk factors (market return minus risk free rate, small ﬁrm premium, the value premium) computed for France. The constants of these regressions are displayed on this table, with t-statistics in parentheses. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

Performance and Behavior of Family Firms

35

1.5

Summary and Leads for Future Research

In this paper, we have sought to understand why family ﬁrms outperform non family ﬁrms, and whether family ﬁrms have distinct “management styles”. First, founders simply have larger labor productivity. This founder eﬀect is very large, since it explains nearly all of the diﬀerence in proﬁtability between founder ﬁrms and widely held corporations. This eﬀect, though large, is very imprecisely estimated in regressions. This suggests that we may not have the most adequate model (or the adequate breakdown) to explain the performance of founder ﬁrms. Yet, our results conﬁrm those of existing papers on founder eﬀects. Fortunately, we are able to say more about the management styles of other family ﬁrms. Secondly, we have presented three pieces of evidence consistent with the fact that, thanks to their longer horizons, heirs can manage their labor force more eﬃciently. First, ﬁrms managed by a descendant of the founder pay signiﬁcantly lower wages, for a given skill structure. Second, they provide insurance across the business cycle to their workers. Third, turnover is less likely for heirs than it is for professional CEOs in family ﬁrms. These three results are consistent with an implicit insurance story : heir managers, because of their longer horizon, ﬁnd it easier than professional managers to sustain reputational contracts with their workers, providing them with more insurance in exchange of lower wages. Professional managers, whether in family ﬁrms or in widely held ﬁrms may lack the credibility necessary to implement such implicit contracts. Third, professional CEOs in family ﬁrms compensate for this lack in credibility vis ` vis their workers by mana aging capital more eﬃciently. First, they pay on average lower interest rates on their outstanding debt. Second, their external acquisitions tend to be, in the long run, more proﬁtable. One plausible explanation is that they were more likely to be trained in ﬁnance/economics at the university or in an MBA. Whether acquired by education or by experience, ﬁnancial literacy seems to be how hired CEOs are as proﬁtable as heirs in our data. We believe these results to be consistent with Bertrand and Schoar (2003)’s ﬁnding that there are heterogenous ﬁnancial management styles across companies. In particular, they show that CEOs holding an MBA degree are more ﬁnancially “agressive”than others. Here, we presented evidence that professional managers in family ﬁrms have a particular style compared to family managers, namely their ability at managing capital eﬃciently. This paper has focused on real eﬀects of family management. When we look at stock returns, it turns out that family ﬁrms, in particular those run by professional managers, have beaten the market over the 1990s, even after taking into account the risk factors that the asset pricing literature considers as important. Our data thus delivered results consistent with Van der Heyden et al (2004). This is not easy to interpret: does this mean that

Bottom-Up Corporate Governance

36

the market has suddenly understood the virtues of family management? Does this mean that non family ﬁrms have done particularly badly over the decade? Can this be interpreted as further evidence that professional managers in family ﬁrms are good at communicating to (and persuading) analysts and propping up the stock price? Another possibility would be that family ﬁrms are subject to particular risks, because they are more likely to be taken over when market conditions are good. As a result, their beta (correlation with the market return) would be time varying and would covary negatively with the market. This is a question we plan to address in future research.

Bottom-Up Corporate Governance

37

Chapter 2

Bottom-Up Corporate Governance
2.1 Introduction

Academics and practitioners have known for long that in the absence of tight monitoring, CEOs of large publicly held ﬁrms may take actions that are detrimental to their shareholders: They commit the ﬁrm’s resources to value destroying “pet” projects, build unproﬁtable empires, prevent valuable takeovers from happening, or even, in some rare yet highly publicized instances, engage in fraudulent window dressing or asset tunneling. To set up counterpowers to the CEO, the consensus among practitioners and regulators has been to rely on a strong board of directors, independent from the management. In many countries, informal codes of corporate governance have been recommending the appointment of independent directors for more than a decade.1 In the US, the recent wave of corporate scandals has triggered a stronger regulatory response, making the hiring of independent directors mandatory for ﬁrms listed on the major stock exchanges.2 Indeed, academic research has found boards to be eﬃcient tools of corporate governance. Independent boards of directors seem to pay more attention to corporate performance when it comes to CEO turnover or compensation (Weisbach (1988), Dahya, Mc Connel and Travlos (2002)). The stock market hails the appointment of independent directors with abnormal returns (Rosenstein and Wyatt (1990)). There is no evidence, however, that independent boards improve proﬁtability or even the value of corporate assets.3 One possibility, at least for large listed ﬁrms, is that independent boards, while still extremely valuable in times of crises, are too far away from day to day
1 As a matter of fact, many large ﬁrms have been eager to comply with their guidelines. For instance, the Cadbury Report issued in the UK in 1992 recommends that “the majority of non-executives on a board should be independent of the company” . The 1998 “Vi´not II” Report in France proposes that “independent directors should account for at least one-third of the Board of Directors”. e Compliance with these guidelines was not mandatory, but widespread. For instance, by 1996, more than 50% of the UK ﬁrms surveyed by Dahya, Mc Connel and Travlos (2002) claimed to comply with the Cadbury Report recommendations. 2 The NYSE and the NASDAQ require since 2003 a majority of independent directors on the board of companies listed on their exchanges. 3 In fact, the correlation might even be negative. A likely reason for this is that poorly performing ﬁrms tend to appoint more outside director (Kaplan and Minton (1994)). Filtering out this endogeneity leads to no apparent correlation between proﬁtability and board independence (Baghat and Black (2003), Hermalin and Weisbach (2003)).

Bottom-Up Corporate Governance

38

operations to add much value to a ﬁrm. As a result, corporate governance scholars have recently shifted their attention away from organizational variables such as board composition towards other dimensions of corporate governance apparent in corporate charters, bylaws or in state takeover laws. The main ﬁnding of this recent literature is that investor-friendly corporate governance provisions boost the price of ﬁrms’ assets by making them more vulnerable to takeovers (Gompers, Ishii and Metrick (2003), Cremers, Nair and John (2005),Bebchuk and Cohen (2004)). This paper studies a new measure of the quality of corporate governance based on “organizational” information. Our intuition is that there is some information to get on the functioning of a company by focusing attention on the composition of the executive suite. After all, CEOs have to face their subordinates on a daily basis, whereas boards of directors only meet a few times every year. More precisely, we develop a measure of “internal governance” that captures the degree of “independence” of top executives from the CEO. On a panel of US listed corporations, we compute the fraction of top ranking executives who joined the ﬁrm before the current CEO was appointed. We think of these executives as “independent”. As CEOs are typically involved in the recruting of their subordinates, executives hired during their tenure are more likely to share the same preferences and/or have an incentive to “return the favor”. Alternatively, executives who experienced previous leadership are less likely to take orders as legitimate simply because they come from the a superior. We ﬁrst provide evidence on corporate performance: We ﬁnd that high internal governance (independent executives) predicts high future performance, using various proﬁtability measures. Conversely, poor performance does not lead to a decrease in internal governance, suggesting a causal eﬀect of internal governance on performance. Our ﬁndings are not aﬀected when we control for traditional, “external” corporate governance measures. We also show that our results are not driven by the departure of executives ”leaving a sinking boat”, i.e. quitting due to the anticipation of the ﬁrm’s future decline. Our second piece of evidence is on the long-run value impact of large acquisitions. We show that a lower fraction of independent executives is associated with signiﬁcantly lower returns for the acquiror’s shareholders. Importantly, however, regular indices of external governance are not correlated with the long-term shareholders’ losses made after an acquisition. The board of director, takeover pressure or the design of corporate charter seem less eﬃcient at preventing bad/expansive acquisitions from happening. We believe an important contribution of our paper is to exhibit an organizational ﬁrm-level variable with strong systematic predictive power on future performance. Our internal governance variable might simply be a measure of

Bottom-Up Corporate Governance

39

the extent of CEO power over the ﬁrm: ”powerful CEOs” might be both prone to do ineﬃcient acquisitions and to replace executives with their own friends with no link between the two. However, we ﬁnd that internal governance is only very weakly correlated with traditional measures of ”CEO power” such as whether the CEO is chairman of the board. Another intepretation of our results is that independent executives may act as a “bottom-up governance” mechanism, making it costly to the CEO to take bad decisions. Our study may have two normative implications for practioneers of corporate governance. First, we learn from our statistical analysis that the intensity of such internal governance can be at least partly observed and could be included in the various indexes of the quality of a ﬁrm’s corporate governance. This implication does not depend on our interpretation of our results: be it the sign of executives “leaving the sinking boat”, of an autocratic CEOs, or of the healthy discipline of having to convince one’s subordinates, the share of independent executives as we measure it predicts performance. A second implication hinges on our “bottom-up governance” interpretation: in addition to management monitoring and advising, a key role of the board should also consist in designing the optimal balance of power within the ﬁrm. Put otherwise, the human resource role of the board is not limited to the usually emphasized CEO succession problem. The paper has ﬁve more sections. Section 2 describes the datasets we use and how we construct our index of internal governance. Section 3 looks at the relationship between internal governance and corporate performance. Section 4 looks at the costs of acquisitions. Section 5 discusses the relation between our internal governance index and usual corporate governance measures. Section 6 concludes on theoretical questions raised by our ﬁndings.

2.2

Data and Measurement Issues

We ﬁrst describe the datasets we use to complete our study. We then discuss the construction of our index of “internal governance” and outline its strengths and weaknesses.

2.2.1

Datasets

We use ﬁve datasets. EXECUCOMP provides us with the ﬁrm level organizational variables with which we proxy for internal governance. COMPUSTAT provides us with ﬁrm level accounting information. IRRC’s corporate governance and director dataset allows us to obtain standard measures of external corporate governance. Acquisitions are drawn from SDC Platinium, and stock returns from CRSP.

Bottom-Up Corporate Governance

40

2.2.1.1

Internal Governance

The ﬁrst dataset is the EXECUCOMP panel of (at least) the ﬁve best paid executives of the largest American corporations. We use this data source to measure the extent of “internal governance” in the ﬁrm. We do this by computing the fraction of executives hired after the CEO took oﬃce (i.e. the fraction of non-independent executives). Thus, internal governance is said to be poor when this fraction is high. Initially, each observation is an executive (or the CEO) in a given ﬁrm in a given year. We focus on years from 1992 to 2002; we start by removing observations for which the executive identifying number is missing. We also exclude duplicate observations. In this (nearly) raw dataset, there are 120,762 observations, which correspond to some 1,840 ﬁrms per year (20,230 ﬁrm-years) with an average of six executives each (including the CEO). As it turns out, 3,499 ﬁrm-year observations have no CEO (using the CEOANN dummy variable indicating which executive is the CEO). In some cases, it is possible to infer the CEO’s identity because, for one of the executives, the BECAMECE variable (date at which the executive became CEO) is non missing, even though the CEOANN dummy is missing (misleadingly indicating that the executive is not the CEO). By ﬁlling in these gaps, we obtain 2,472 ﬁrm year observations, and end up with 19,203 ﬁrm-years for which we know the identity of the CEO (a total of 115,933 observations in the executive-ﬁrm-year dataset). To compute the fraction of non indenpendent executives, we will need to compare the CEO’s tenure to the executives’ seniorities within the company. A ﬁrst approach - which corresponds to the results listed in the paper is to rely on the seniority (within the ﬁrm) and tenure (within the position) variables reported in EXECUCOMP. The BECAMECE variable gives us, for the current CEO, the precise date at which he(she) was appointed as CEO whether he(she) was hired from inside or outside the ﬁrm. Other executives’ seniorities can be recovered using the JOINED C variable, which reports the date at which the executive actually joined the ﬁrm. Unfortunately, these variables are often missing: we lose 2,291 ﬁrm-years (12,262 executives-ﬁrm-years) by focusing on ﬁrms where the CEO’s date of appointment is non missing. We then lose a further 6,760 ﬁrm years (39,695 executives-ﬁrm-years) by restricting ourselves to ﬁrms where we have non missing seniority for at least one executive. We end up with 11,179 ﬁrm-years, from 1992 to 2002, for which we can now compute the fraction of executives hired after the current CEO. We call this measure of executive dependence FRAC1. Overall, we lose 19,203-11,179=8,024 ﬁrm-year observations in the process of constructing our measure of internal governance, mostly because many executives do not report their seniority within the ﬁrm. In 4,307 of our remaining 11,179 ﬁrm-years, internal governance is measured by comparing the CEO’s tenure with the seniority of only one

Bottom-Up Corporate Governance

41

executive. This means that FRAC1 will be a very noisy measure of executive dependence; while this does not create an obviously spurious correlation with corporate performance or returns to acquisitions, it is going to bias our estimates of the eﬀect of internal governance downwards, as measurement error often does. A second approach would be to dispense wih the seniority and tenure variables altogether and make direct use of the fact that we can follow individuals in the EXECUCOMP panel. To remove left censorship (the panel starts in 1992), we need to restrict ourselves to ﬁrms where we observe at least one episode of CEO turnover. Once the new CEO has been appointed, we can compute the fraction of executives that were not listed in the dataset before the new CEO started (we name this alternative variable FRAC2). The main advantage of this approach is that we can dispense of both BECAMECE and JOINED C variables, which are often missing. The cost is that the need to observe CEO turnover restricts the number of ﬁrm-years to 6,617. This is less that the 11,179 observations available to compute FRAC1. Also, focusing on ﬁrms with at least one CEO turnover over the course of ten years may mechanically overweight ﬁrms facing governance problems. Finally, executives enter the panel when they either (1) are hired by the ﬁrm, (2) make it into the ﬁve best paid people list, or (3) the ﬁrm decides to report their pay in its annual report/proxy. Hence, entry in the panel is a very noisy measure of hiring. In spite of its shortcomings, the second - panel based - variable FRAC2 has a correlation coeﬃcient of 0.41 with the ﬁrst - seniority based - variable FRAC1. Both approaches led to results very similar in terms of size and signiﬁcance, so we chose to focus here on the ﬁrst measurement approach. Of course, estimates based on FRAC2 are available from the authors upon request. 2.2.1.2 Corporate Accounts

Our tests will correlate internal governance with corporate performance. Thus, for each ﬁrm-year observation from our EXECUCOMP sample, we retrieve ﬁrm level accounting information from COMPUSTAT (we lose only 161 observations, for which we cannot ﬁnd the book value of assets, in the merging process). We match by GVKEY identiﬁer. We compute proﬁtability as return on assets (ROA).4 We construct Market to Book as the ratio of the ﬁrm’s assets market value to their book value, as in Gompers, Ishii and Metrick (2003).5 We proxy ﬁrm size by log(total assets). We proxy ﬁrm age by taking the diﬀerence between the current year and the ﬁrst year of presence in the COMPUSTAT panel. We construct the 48 Fama-French industry dummies using the ﬁrm’s 4 digit SIC
4 Return on Assets is Operating Income Before Depreciation (item 13) minus Depreciation and Amortization (item 14) over Total Assets (item 6). 5 Market to Book is the ratio of market to book value of assets (item 6). The market value is computed as Total Assets (item 6) plus the number of common shares outstanding (item 25) times share price at the end of the ﬁscal year (item 199) minus Common Equity (item 60) minus Deferred Taxes (item 74).

Bottom-Up Corporate Governance

42

industry code. Finally, we windsorize some variables (ROA, Market to Book) at the 1% and 99% levels. 2.2.1.3 External Governance

We will also look at how our measure of internal governance correlates with traditionnal corporate governance measures. Thus, for each ﬁrm year observation, we gather information on corporate governance from IRRC’s corporate governance and directors dataset. This dataset provides us with commonly used proxies for corporate governance, namely, the fraction of independent directors, the number of directors sitting on the board and the fraction of former employees sitting on the board. These variables are available for the 1996-2001 period only, and mostly for large ﬁrms. Out of 11,179 ﬁrm-year observations where we can measure internal governance, only 4,531 observations have information from IRRC. We will also look at the increasingly popular Gompers, Ishii, and Metrick’s (hereafter GIM) index of corporate governance, which compiles various corporate governance provisions included in the CEO’s compensation package, in the corporate charter and the board structure. The GIM index is available for 1990, 1993, 1995, 1998 and 2001. In other years, we assume that it takes the value that it had in the most recent year when it was non missing. Again, just over a half (5,872 over 11,179) of our observations have a deﬁned GIM index. 2.2.1.4 Acquisitions

To see if top ranking executives are able to constrain major CEO decisions, we focus on the eﬀect of internal governance on the acquisition performance. We obtain the list of ﬁrms who made signiﬁcant acquisitions from SDC Platinium (deals of value larger than $ 300 million). SDC provides us with the bidder’s CUSIP and the transaction value of the deal. We focus on completed deals where the bidder bought at least 50% of the target’s shares. For each ﬁrm-year observation in our EXECUCOMP sample, we compute the number of targets acquired during that year and the overall amount spent on the deal(s). In our base sample of 11,179 ﬁrm-years where the internal governance measure FRAC1 is available, 22% of the observations correspond to ﬁrms making at least one acquisition; 1998 and 1999 are the peak years, with more than 26% of ﬁrms making at least one acquisition. Most acquirers make only one deal per year, but there are a few serial acquirers (three percent of the observations correspond to at least ﬁve deals carried out during the year). 2.2.1.5 Stock Returns

We are also interested in computing the net beneﬁt of acquisitions. To do this, we compute long run abnormal stock returns following the acquisition, for each acquirer.

Bottom-Up Corporate Governance

43

We merge the above SDC extract with our base sample from EXECUCOMP. We end up with a list of 818 deals for which we know the acquirer, the date of the acquisition, and FRAC1 (the share of executives appointed after the CEO took oﬃce). Serial acquirers are overrepresented. Out of 818 deals, 188 involve one time buyers, while 368 involve ﬁrms carrying out at least four large deals. Overall, our sample features 359 diﬀerent acquirers. We then match this deal dataset with the acquirer’s stock returns as provided by CRSP. More precisely, we retrieve monthly acquirer stock returns from a period extending 48 months prior to each acquisition to 48 months after the deal. We remove deals with less that 48 months of acquirer returns history before the acquisition. This reduces our sample size to 669 deals. We then estimate a three factor Fama-French model for each acquirer using the 48 pre-acquisition months available. We use the returns of the MKTRF, SMB, and HML portfolios from Kenneth French’s web site. We then use this model to compute abnormal returns both before and after the deal.

2.2.2

Constructing an Internal Governance Index

The assumption underlying the internal governance measure is that the CEO is directly or indirectly involved in the recruitment process of top executives. Hence, executives appointed during his tenure are more likely to be loyal to him and/or share his preferences than executives who were picked by a predecessor. However, one needs to be careful with the mechanical drivers of FRAC1. As a CEO’s seniority increases, a larger fraction of executives have (mechanically) been appointed during his tenure. Conversely, executives who have been with the ﬁrm longer are on average more likely to have been hired before the current CEO. This suggests that FRAC1 is positively correlated with CEO tenure, and negatively with executive seniority. Also, externally appointed CEOs often have the mandate to arrange an “executive shake-up.” When recruited from the outside, CEOs have enough bargaining power vis ` vis the board of directors to bring in their own teams. Hence, FRAC1 a should be mechanically larger in the presence of outsider CEOs. Finally, a new CEO’s appointment is sometimes followed by immediate waves of executive departures and arrivals that might be unrelated to internal governance (for example, top executives hoping for the top job leave the ﬁrm and need to be replaced). It might be tempting to see these mechanical sources of variation in the proportion of aligned executives as exogenous shocks to internal governance, but they might be related to ﬁrm performance for reasons orthogonal to internal governance. Ignoring these sources of variation would thus lead to biased estimates of the eﬀect of internal governance on performance. For example, CEO tenure may directly aﬀect corporate performance simply because experience on the job matters. Also, if the ﬁrm is in really bad shape, a new CEO will have to inject more “fresh blood” into the corporate suite (Hayes, Oyer, and Schaefer, 2005), which mechanically increases executive turnover.

Bottom-Up Corporate Governance

44

We therefore choose be to as conservative as possible and ﬁlter out these mechanical eﬀects when we seek to measure “internal governance.” Also, we will include them as controls in all performance regressions.6 More precisely, our internal governance (henceforth IG) index is deﬁned as the residual of the fraction of “dependent” executives regressed on its expected mechanical correlates: F RAC1it = a + b.CEOT ENit + c.EXECSENit + d.OU T SIDEit +e.KN OW Nit + f.F RAC1 1Yit + δt + εit (2.1)

where, for ﬁrm i in year t, CEOT ENit stands for CEO’s tenure (in years), EXECSENit for average executive seniority within the ﬁrm, OU T SIDEit is a dummy indicating whether the CEO comes from outside the ﬁrm, KN OW Nit is the fraction of executives for which seniority is reported in the data, and F RAC1 1Yit the fraction of executives that arrived within a year of the CEO’s nomination. We also include year dummies δt . We deﬁne our Internal Governance (hereafter IG) index as the residual εit . It is larger when more executives than expected were hired after the current CEO was appointed. Hence, high values of the IG index mean poor internal governance (consistently with the Gompers-Ishi-Metrick external governance index). The regression results are reported in Table 2.1, which has four columns. Column 1 includes the seniority variables (EXECSEN for executives and CEOT EN for the CEO). Column 2 adds the fraction of executives for which seniority is actually reported in EXECUCOMP (KN OW N , which we include to control for potential selection biases), and the fact that the CEO has been appointed from the outside (OU T SIDE). Column 3 adds the fraction of executives appointed within a year of the CEO’s nomination, to control for management ”shake-ups.” Column 4 includes ﬁrm size, age and industry as additionnal regressors. As it turns out, all these mechanical correlates of F RAC1 work as we expected them to. F RAC1 is positively and strongly correlated with CEO tenure and negatively correlated with executive tenure (Columns 1 to 4). These two variables alone explain 25% of the variance of F RAC1 (column 1). F RAC1 is positively correlated with the fraction of executives whose seniority is reported: Hence, more ”transparent” ﬁrms tend to have executives appointed after the CEO. F RAC1 is also strongly associated with the presence of outside CEOs. There are at least two possible interpretations for this. First, outside CEOs are often given a mandate to reshuﬄe the top management, and as a result the fraction of executives who joined the company with them is large. As it turns out, the coeﬃcient on OU T SIDE is somewhat reduced when we also include F RAC1 1Y in column 3. But it remains positive and signiﬁcant, which leaves room for additional explanations: the appointment of outside CEOs triggers the departure of talented executives who were hoping to get the top job.
6 By virtue of the Frisch Waugh theorem, the two approaches are equivalent. However, our residual approach will be helpful when we look at stock returns following acquisitions, since we will simply compare ﬁrms with negative and positive residuals.

Bottom-Up Corporate Governance

45

Another possibility could simply be that management shake-ups tend to happen when the ﬁrm is doing badly, which may also generate departures. In any case, the need to replace the lost executives mechanically increases our index. Finally, ﬁrm level variables (industry, age, or even size) are not strongly correlated with F RAC1, and accordingly explain little of its variance. Thus, in the following analysis, we use estimates from column (3), i.e., compute IG using CEO and executives’ characteristics (which account for 71% of the variance of F RAC1), but not the ﬁrm level variables (which account for a mere 2%). Last, one possible concern could be that our internal governance index may be correlated with intense merger activity in the past. After many mergers, top executives from the targets join the executive suite, mechanically increasing our index. If the ﬁrm still has trouble “digesting” its past acquisitions, it is likely to underperform on both accounting and stock price measures. To address this concern, we correlated our residual IG index with the number of past acquisitions for a cross section of ﬁrms in 2000. We found no evidence that high IG index ﬁrms had bought a particularly large number of ﬁrms in the 1990s. This is robust to various controls and to the year chosen. Our index is thus not a proxy for M&A “indigestion.”

2.3

Internal Governance and Corporate Performance

We start by investigating the correlation between internal governance and corporate performance. Figure 2.1 provides a ﬁrst look at the relationship between our IG index and corporate performance. In this ﬁgure, we split the sample distribution of our IG index into ﬁve quantiles. Then, for each quantile, we compute the mean industry7 adjusted performance, as well as the 95% conﬁdence band assuming normality. Performance is measured through ROA (left panel) and market to book value of assets (right panel). Figure 2.1 shows a positive and statistically signiﬁcant association between good internal governance (low values of our IG index) and corporate performance.

2.3.1

Basic Results

As discussed above, some mechanical correlates of internal governance may be correlated with corporate performance. For example, junior CEOs or executives may be on average worse performers simply because they lack experience. We thus move to a multivariate analysis that allows us to capture these ”human capital” eﬀects. We run the following regression: Yit = α + βIGit−1 + (IG controls)it + (F irm controls)it + εit
7 We

(2.2)

used the Fama-French 48 industries (Fama and French (1997)).

Bottom-Up Corporate Governance

46

Table 2.1: Mechanical Correlates of Internal Governance

Fraction of executives appointed after the CEO (×100) (1) CEO seniority 1.3∗∗∗ (0.1) Executives’ mean seniority -1.2∗∗∗ (0.0) Fraction of executives whose seniority is reported (×100) CEO from outside (2) 1.3∗∗∗ (0.1) -1.0∗∗∗ (0.0) 0.6∗∗∗ (0.0) 9.3∗∗∗ (0.8) Fraction of executives appointed in the year foll. CEO nomination (×100) Ln(Firm Age) (3) 1.4∗∗∗ (0.1) -0.9∗∗∗ (0.0) 0.5∗∗∗ (0.0) 6.9∗∗∗ (0.7) 0.5∗∗∗ (0.0) (4) 1.5∗∗∗ (0.7) -0.8∗∗∗ (0.0) 0.4∗∗∗ (0.0) 6.3∗∗∗ (0.7) 0.5∗∗∗ (0.0) -1.0∗∗ (0.5) Firm Size -0.0 (0.2) 48 industry dummies Year dummies R2 Observations No Yes 0.25 11,147 No Yes 0.62 8,728 No Yes 0.71 8,728 Yes Yes 0.73 8,166

Source: OLS estimates, allowing for heteroskdastic residuals, clustered at the ﬁrm level. The fraction of executives appointed after the CEO is regressed on various variables suspected to be mechanically correlated: column 1 controls for the fact that the CEO is an outsider, the CEO seniority, as well as for the mean seniority of executives; column 2 adds the number of executives appointed in the ﬁrst two years following the CEO nomination; column 3 adds ﬁrm speciﬁc control, namely Firm Size as measured by log of Asset, log of Firm Age and the 48 Fama French industries. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance. Our Internal Governance Index is deﬁned as the residual of column (2) regression.

where Yit measures corporate performance (ROA, market to book value of assets). IGit−1 is our measure of internal governance, lagged one period.8 We include two sets of controls. First, the mechanical correlates of our index are included since it may be argued that they directly aﬀect corporate performance (CEO tenure, mean executive
8 We seek to partially avoid obvious simultaneity biases, such as the ones we discuss below. As it turns out, our results are insensitive to the time-lag used.

Bottom-Up Corporate Governance

47

Figure 2.1: Abnormal Economic Performance by Quintile of Governance Index

seniority, share of executive hired right after the CEO, a dummy indicating if the CEO is an insider or not). However, we must bear in mind that, because of the Frisch Waugh theorem, their inclusion does not aﬀect our estimate of β (the IG index is by deﬁnition orthogonal to these variables, so they do not create any “omitted variable bias”). Secondly, we add ﬁrm level controls that are traditionnally strong correlates of performance: log(ﬁrm age+1), log(assets), year dummies, and 48 Fama-French industry dummies. Since we have several observations per ﬁrm (corresponding to diﬀerent years), and because IGit is strongly persistent, it is likely that the εit are not independent from diﬀerent observations of the same ﬁrm i. Hence, we correct for this form of heteroskedasticity by looking at Hubert-WhiteSandwich estimates. The sample correlation between performance and the IG index is strong and stable across years (results available

Bottom-Up Corporate Governance

48

Table 2.2: Accounting Performance and Internal Governance

Return on Assets (×100) Explanatory variables (1) (2) (3)

Market to Book (1) (2) (3)

Internal governance index (delayed by 1 year)

-7.1∗∗∗ (1.5) [-1.1]

-7.3∗∗∗ (1.5) [-1.1]

-3.7∗∗∗ (1.2) [-0.6]

-0.6∗∗∗ (0.2) [-0.1]

-0.6∗∗∗ (0.2) [-0.1]

-0.4∗∗ (0.2) [-0.1]

Controls:

Firm initial proﬁtability CEO, Executive characteristics Firm log(assets), log(age) Year, Industry FE

No No Yes Yes

No Yes Yes Yes

Yes Yes Yes Yes

No No Yes Yes

No Yes Yes Yes

Yes Yes Yes Yes

Source: Huber-White-Sandwich estimates, allowing for correlation of all observations of a given ﬁrm. “Internal Governance Index” is the share of EXECUCOMP executives who joined the company after the CEO was appointed, ﬁltered from mechanical eﬀects. It is deﬁned as the residual of the column (3) regression in table 2.1. Corporate performance is measured through Return on Assets (ﬁrst three columns) and through market valuation of assets (last three columns). All regressions use log(book assets), log(ﬁrm age), year dummies and the 48 Fama French industry dummies. In columns 2 and 5, we add the CEO and executives characteristics that serve as regressors in table 2.1. Column 3 (resp. column 6) further adds the ﬁrm’s ROA (resp. market to book) computed in its ﬁrst year of presence in COMPUSTAT after 1991, as a limited attempt to control for ﬁrm level unobserved heterogeneity. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance. The term reported in brackets is the marginal eﬀect of one standard deviation change in governance index on the dependent variable.

Bottom-Up Corporate Governance

49

from the authors). Multivariate regression results are reported in Table 2.2. Columns 1 to 3 use ROA as dependent variable in equation (2.2); columns 4 to 6 use market to book value of assets.9 Columns 1 and 4 report regression results only with ﬁrm level controls, and columns 2 and 5 include the mechanical correlates of the index. As expected, the diﬀerence is negligible, and stems from the fact that the sample on which the index is estimated (Table 2.1, column 3) is slightly diﬀerent from the sample used for the performance regressions (2.2). Columns 3 and 6 further control for initial performance, as a limited attempt to control for ﬁxed eﬀects.10 Each time, our index of governance is signiﬁcantly and robustly correlated with performance: a one standard deviation increase in IG results in a decrease of about 1.5 ROA percentage point and about 10% of market value of assets. The explanatory power of this eﬀect is not very large (some 10% of standard deviation of the explained variable), but, as we will see, it is consistently signiﬁcant and easily surpasses the usual “external” corporate governance measures. Also, the small size of our coeﬃcients should not surprise us given the noise of our internal governance measure FRAC1 (see section 2.1).

2.3.2

Robustness Checks and Causality

Table 2.3 checks whether the performance-IG correlation reported in Table 2.2 is driven by any particular subperiod. In this table, we report, for both measures of corporate performance, the point estimate of β in (2.2) where internal governance is measured by IG, including both ﬁrm level and mechanical controls as in columns 2 and 5 of table 2.2, except that one regression is run for each year. As it turns out, the point estimate is fairly stable across periods and signiﬁcant in most years. As a consistency check, we veriﬁed that we get similar magnitudes and signiﬁcance levels by regressing directly on FRAC1, our unﬁltered measure of internal governance, rather than on our internal governance index. There are many stories consistent with the relation between IG and performance found in Tables 2.2 and 2.3. Our favored interpretation is that strong internal governance is a way for shareholders to “hold the CEO on a tight leash” and prevent the CEO from undertaking “crazy” projects or building an empire. One could argue, however, that the causality runs in the opposite direction: declining performance may actually trigger an increase in IG (a drop in our measure of internal governance). One plausible story could be based on management turnover. In most ﬁrms, poor performance triggers a change in the management team. In this scenario, internal governance worsens because performance declines, not the contrary. results are obtained with Return on Equity, but we did not report them because of space limitation. have also run, but not reported, regressions of corporate performance on FRAC1, the fraction of executives hired after the CEO took oﬃce, as well as various controls. FRAC1 turned out to be highly signiﬁcant in all speciﬁcations we tested. This is not surprising by virtue of Frisch Waugh theorem.
10 We 9 Similar

Bottom-Up Corporate Governance

50

Table 2.3: Accounting Performance and Internal Governance - Year by Year Results

ROA

Market To Book

1993

-9.1∗∗∗ (3.0)

-0.4 (0.4) -0.9∗∗∗ (0.3) -0.5 (0.3) -0.5∗ (0.3) -0.8∗∗∗ (0.3) -1.1∗∗∗ (0.3) -0.9∗∗∗ (0.3) -0.8∗∗ (0.3) -0.2 (0.3) -0.3 (0.3)

1994

-9.9∗∗∗ (3.5)

1995

-10.3∗∗∗ (2.7)

1996

-10.6∗∗∗ (2.6)

1997

-5.3∗∗ (2.3)

1998

-9.7∗∗∗ (2.4)

1999

-5.9∗∗∗ (2.3)

2000

-7.1∗∗∗ (2.6)

2001

-3.8 (2.6)

2002

-5.2∗∗ (2.4)

Fama-Mac Beth

-7.7∗∗∗ (0.8)

-0.6∗∗∗ (0.1)

Source: OLS estimates. Regressions of corporate performance on internal governance index and controls are run separately each year. The coeﬃcients on internal governance and their standard error are reported. Each column corresponds to the choice of one corporate performance measure (ROA or M/B). Corporate performance is then regressed on one-year-lagged internal governance index, controlling for CEO and executive seniority, fraction of executives reporting seniority, CEO’s origin, log(assets), log(ﬁrm age), sales growth and 48 industry-dummies. The speciﬁcation is identical to the regression presented in table 2.2, columns 2 and 5. The bottom row indicates the Fama-Mac Beth estimate. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance.

Bottom-Up Corporate Governance

51

While we have no “smoking gun” to assess the causal relation between internal governance and corporate performance, we can at least take two preliminary steps to reduce the likelihood of reverse causation. First, we look at the joint dynamics of internal governance and corporate performance. Do changes in corporate performance happen before or after changes in internal governance? To test these two hypotheses, we run the following two regressions: Yit IGit = α + βIGit−L + γYit−L + controlsit + εit = a + bIGit−L + cYit−L + controlsit + εit (2.3) (2.4)

where we use one and two year lags (L = 1, 2) and Yit is the ﬁrm’s corporate performance at date t. If changes in corporate performance tend to lead changes in IG, we should not be able to reject that c > 0 and γ = 0. Such a test can be thought of as the panel data version of causality tests ` la Granger in time series analysis. a Estimates of equations (2.3)-(2.4) are reported in Table 2.4. All regressions include ﬁrm level controls (age, size, year and industry dummies). Column (1) reports the estimates of β and γ of equation (2.3), while column (2) reports the estimates of b and c from (2.4). The top panel reports the regression results assuming L = 1. The bottom panel assumes L = 2. The results suggest that, in general, changes in internal governance happen before changes in corporate performance as estimates of c are never signiﬁcantly diﬀerent from zero, while estimates of β are. Another endogeneity concern, which is not ruled out by our time-series evidence is the following: executives might tend to leave companies when they anticipate poor performance (for example because they want to avoid the danger of getting ﬁred). If executives have private information on future performance, IG would rise before performance declines, but without being the cause of decline. One justiﬁcation for such anticipation eﬀects is that executives can observe the CEO’s ability, or the evolution of the ﬁrm’s markets, before they materialize in corporate accounts. As a consequence, our IG index might be simply proxying for executive turnover, which would itself be a predictor of performance decline. We thus add to equation 2.2 the fraction of executives that left the ﬁrm in the previous year as a control. This turnover control is constructed as the fraction of the ﬁrm’s year t − 1 executives who are no longer reported as working for the company at year t in the EXECUCOMP data. To be fair, they can drop out of our sample either because they are no longer employees of the company, or because they do not belong any more to the most paid employees of the company. But this is as far as EXECUCOMP allows us to go to measure executive departure. Controlling for such measure of executive turnover means that we compell the identiﬁcation

Bottom-Up Corporate Governance

52

Table 2.4: Accounting Performance and Internal Governance - Granger Causality

Internal Governance

ROA (×100)

One-year-lag speciﬁcation: Internal governance index (-1) 0.8∗∗∗ (0.0) ROA (-1) (×100) -0.5 (1.3) -1.8∗∗∗ (0.5) 0.8∗∗∗ (0.0)

Two-year-lag speciﬁcation: Internal governance index (-2) 0.7∗∗∗ (0.0) ROA (-2) (×100) -0.9 (2.4) -2.0∗∗∗ (0.8) 0.6∗∗∗ (0.0)

Controls : Firm log(assets),log(age) Year, Industry FE Yes Yes Yes Yes

Source: Hubert-White-Sandwhich estimates, allowing for residuals correlated across all observations of each ﬁrm. In the top panel, column 1 reports the estimate of a regression of internal governance on one-year lagged internal governance and oneyear lagged corporate performance. Column 2 reports the result of a regression of corporate performance on one-year lagged internal governance and one-year lagged corporate performance. Both regressions control for ﬁrm age and size, industry and year ﬁxed eﬀects. The bottom panel reports the same regression results, taking two-year-lags as explanatory variables, instead of one-year-lags. Corporate performance is measured through Return on Assets. Standard errors are between parentheses. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance.

of the coeﬃcient on our IG index to rely exclusively on the comparison between the year the CEO started his/her tenure, and the year top execs started to work for the ﬁrm. We present the new estimation results in 2.5, using the same ﬁrm level controls as in 2.2. As it turns out, executive turnover indeed has a signiﬁcant negative impact on ﬁrm performance, conﬁrming the idea that unexepectedly high executive turnover is an early sign of bad performance. Nevertheless, adding this control does not aﬀect the magnitude and signiﬁcance of the impact of our internal governance measure on performance, either measured as

Bottom-Up Corporate Governance

53

ROA or Market to Book. Overall, Table 2.4 and Table 2.5 point toward a causal link going on from high Internal Governance to bad performance.

Table 2.5: Accounting Performance and Internal Governance - Controlling for Executives Turnover

ROA Explanatory variables (1)

Market to Book (2)

Internal governance index (delayed by 1 year)

−2.2∗∗∗ (.56) [-.4]

−0.3∗∗∗ (.09) [-.05]

Controls:

Executives Turnover at year t-1

−3.1∗∗∗ (.66) [-.5]

−0.2∗∗∗ (.08) [.-03]

Firm initial proﬁtability CEO, Executive characteristics Firm log(assets), log(age) Year, Industry FE

Yes Yes Yes Yes

Yes Yes Yes Yes

Source: Huber-White-Sandwich estimates, allowing for correlation of all observations of a given ﬁrm. “Internal Governance Index” is the share of EXECUCOMP executives who joined the company after the CEO was appointed, ﬁltered from mechanical eﬀects. It is deﬁned as the residual of the column (3) regression in table 2.1. Corporate performance is measured through Return on Assets (ﬁrst three columns) and through market valuation of assets (last three columns). Executive

Turnover at year t − 1 measures the fraction of the ﬁrm’s year t − 1 executives who are no longer reported as working for the company at year t in the EXECUCOMP data. All regressions use log(book assets), log(ﬁrm age), year dummies, the 48 Fama French industry dummies, CEO and executives characteristics, and ﬁrm’s initial ROA (resp. market to book) computed in its ﬁrst year of presence in COMPUSTAT after 1991. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance. The term reported in brackets is the marginal eﬀect of one standard deviation change in governance index on the dependent variable.

Bottom-Up Corporate Governance

54

2.4

Internal Governance and Acquisitions

To test whether internal governance increases the quality of the ﬁrm’s strategy by constraining the CEO in his choices, a natural place to look is at the ﬁrm’s acquisition policy. There is substantial debate among ﬁnancial economists as to whether long-run acquisition returns are positive or negative for the acquiring ﬁrm. Loughran and Vijh (1997) ﬁnd that the returns to long-run investors in acquiring ﬁrms are on average negative, in particular when the deal is ﬁnanced with stock issues. Mitchell and Staﬀord (1999), among others, criticize their estimates, partly because post acquisition returns tend not to be independent events, as acquisitions generally cluster around stock market booms. The main problem with this literature is that there is considerable heterogeneity among types of acquisitions and their performance. Thus, ﬁnancial economists lose substantial information on their entire distribution by focusing on average returns and average proﬁtability. In attempt to reduce this heterogeneity, some recent papers have outlined the size of acquisitions as a key factor for success or failure (Moeller, Schlingemann, and Stulz (2005), Bradley and Sundaram (2004)). The evidence they present is consistent with “small” acquisitions as value-creating, and large ones as value-destroying. Following up on these papers, we look at the eﬀect of internal governance on shareholder losses (gains) in large acquisitions. But before looking at gains, we ﬁrst focus on the relation between internal governance and acquisition policy. We ﬁnd that ﬁrms with good internal governance do not make fewer acquisitions and that their acquisitions do not correspond to smaller purchases. We follow Gompers, Ishii and Metrick (2003), and use SDC to compute, for each ﬁrm-year of our EXECUCOMP extract: (1) the number of deals of more than $10 million in value and (2) the overall amount of all deals struck within the year (the sum of all transaction values if there are several deals), normalized by the acquirer’s market capitalisation. None of these measures of acquisition intensity proved to be correlated with our IG-index. Moreover, we ﬁnd that the IG index is not correlated with the number of past acquisitions, which means that selecting ﬁrms with poor internal governance does not select ”serial acquirers.” We then turned to the impact of internal governance on acquisition quality. As argued above, we focus on large acquisitions (whose value exceeds $300 million $). To measure the performance of acquisitions, we ﬁrst follow Loughran and Vijh (1997) and focus here on the acquirer’s long term abnormal stock returns, which we compute using a four factor pricing model (the Fama-French (1996) three factors plus Carhart’s (1997) momentum factor) estimated at the ﬁrm level in the 48 months preceeding the acquisition. We restrict ourselves to the 1993-2002 period, in order to be able to use EXECUCOMP information. We split the sample of transactions into two parts (each comprising some 400 deals): deals where the acquirer

Bottom-Up Corporate Governance

55

Table 2.6: Long Run Abnormal Returns Following a Major Acquisition

Internal Governance Poor Months since acquisition -12 -6 0 +6 +12 +18 +24 +30 +36 +42 +48 2.7 0.6 -5.3 -11.4 -16.0 -18.0 -21.2 -21.7 -25.0 -25.4 1.9 1.6 2.5 -0.4 -6.4 -8.4 -11.4 -12.3 -12.9 -9.4 -0.8 1.0 7.8∗∗∗ 10.9∗∗∗ 9.6∗∗∗ 9.6∗∗ 9.8∗∗ 9.4∗ 12.0∗∗ 16.2∗∗∗ Good Diﬀerence

External Governance Poor Good Diﬀerence

1.7 1.0 -3.3 -5.5 -11.8 -15.1 -14.0 -14.3 -14.2 -11.8

1.6 1.2 0.6 -2.8 -4.5 -5.3 -9.0 -11.3 -14.8 -15.3

-0.1 0.3 3.9 2.7 7.3∗∗ 9.7∗∗ 4.8 3.0 -0.6 -3.5

Source: 818 acquisitions from SDC Database. Abnormal returns are computed after estimating, for each acquirer, a Fama French 3 factor model + momentum on the 48 months preceeding the acquisition. Cumulative abnormal returns, starting 12 months before the deal, are computed for each ﬁrm. Column 1 reports, every 6 months, the average cumulative abnormal returns of acquirers with internal governance lower than median. Column 2 does the same for above median internal governance acquirers, while column 3 reports the diﬀerence. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance, using a standard test of equality, assuming away the equality of variances.

has above-median IG index (poor internal governance), and deals where the acquirer has below-median IG index (good internal governance) in the year preceding the acquisition. Columns 1 and 2 of Table 2.6 report, separately for good and poor internal governance acquirers, the average cumulative abnormal returns, starting 12 months before the deal up to 48 months after the deal. Column 3 reports the diﬀerence in cumulative returns, and tests for the equality of average returns using a standard t-test, without assuming equality of variances. Figure 2.2 plots cumulative abnormal returns for each month, separately for poor (left panel) and good (right panel) internal governance acquirers. We ﬁnd that ﬁrms with poor internal governance make largely underperforming acquisitions. Four years after the acquisition, ﬁrms with good internal governance have on average lost some 15% of shareholder value, which is signiﬁcantly diﬀerent from zero. However, ﬁrms with poor internal governance have lost almost 30%, which is both signiﬁcantly diﬀerent from zero and from the wealth lost by long term shareholders of well governed ﬁrms.

Bottom-Up Corporate Governance

56

Figure 2.2: Long-Run Returns From Acquisitions: Good vs Poor Internal Governance

This diﬀerence is robust to (1) the way we split the sample, on condition that each contains enough observations in each category (good/poor governance) and (2) to the pricing model (results are almost similar when we omit the momentum factor; they are somewhat noisier, but still point in the right direction if we use the CAPM or if we merely substract the market return from stock returns). As a robustness check, we then look at the signiﬁcance of our results using a calendar time portfolio method as recommended e.g., by Mitchell and Staﬀord (1997). This method addresses the critique that, due to their time overlap, post-event returns are not independent. Another problem is that measurement errors inherent in the computation of individual abnormal returns are compounded by calculating cumulative returns. For our sample of acquisitions, we therefore construct two equally weighted portfolios of ﬁrms that completed

Bottom-Up Corporate Governance

57

at least one acquisition within the last n months. The ﬁrst portfolio is long in acquirers whose internal governance index one year prior to the acquisition is below-median (the “good internal governance” portfolio). The second portfolio is long in acquiring ﬁrms with above median IG index (the “poor internal governance” portfolio). Both portfolios are therefore rebalanced each month as acquirers whose deal occurred more than n months ago leave
P G and new acquirers enter. Let Rn,t (resp. Rn,t ) be the monthly return of the poor (resp. good) internal governance

portfolios. We then estimate the abnormal returns of the two portfolios with a four-factor asset pricing model (the three Fama French factors plus the momentum factor, all available from the Kenneth French website). We also estimate the abnormal returns of a portfolio that is long in good internal governance, and short in poor internal governance acquirers, as in equation (2.7):

G E Rn,t − Rfn,t

G G = αn + βn (Rmt − Rft ) + sG . SM Bt n

(2.5)

+hG . HM Lt + uG .U M Dt n n
P E Rn,t − Rfn,t P P = αn + βn (Rmt − Rft ) + sP . SM Bt n

(2.6)

+hP . HM Lt + uP .U M Dt n n
G P E Rn,t − Rn,t G−P G−P = αn + βn (Rmt − Rft ) + sG−P . SM Bt n

(2.7)

+hG−P . HM Lt + uG .U M Dt n n
G−P P G represent the average monthly abnormal returns, given the The intercepts of these regressions αn , αn , and αn

model. These ”alphas” are reported in Table 2.7, for n = 12, 24, 36 and 48 months. First, notice that the long-run abnormal returns of all acquisitions (which we report as a benchmark in the ﬁrst line of Table 2.7) are slightly positive and marginally signiﬁcant, in contrast with the results of long run stock returns, which are negative and signiﬁcant. This discrepancy is at the heart of the methodological controversy on long-run stock return studies.11 When we sort by internal governance value, results conﬁrm our cumulative abnormal returns analysis: Abnormal returns to good internal governance ﬁrms after major acquisitions are positive and signiﬁcant (some 0.5% per month) within 1, 2, 3 or 4 years following the deal announcement. They are small and insigniﬁcant for poor internal governance ﬁrms. Our long-short portfolio’s alphas are positive and signiﬁcant when the selection window looking at three year returns on acquiring ﬁrms, Mitchell and Staﬀord (2000) ﬁnd an equal weighted monthly alpha of -0.2%, which is statistically signiﬁcant. The diﬀerence between our result and theirs may stem from the time period chosen (we look at 19932002, while their time frame is 1961-1993). Another possibility is that ﬁrms are selected on the basis of their belonging to EXECUCOMP (we will return to this issue below). This, however, should not aﬀect the comparison between poor and good governance ﬁrms.
11 When

Bottom-Up Corporate Governance

58

Table 2.7: Post Acquisition Alphas of Acquiring Firms: Sorted by Internal Governance

Equal weights 0 - 12 0 - 24 0 - 36 0 - 48

All acquisitions

0.3 (1.3)

0.2 (1.4) 0.5 (2.6) -0.0 (0.1) 0.5 (1.9)

0.3 (1.6) 0.6 (3.1) -0.0 (0.2) 0.6 (2.5)

0.3 (1.9) 0.6 (3.3) -0.0 (0.0) 0.6 (2.5)

Longs Good IG

0.5 (2.3)

Longs Poor IG

0.1 (0.4)

Good IG - Poor IG

0.5 (1.4)

Source: CRSP, Compustat and Execucomp over 1993-2002. This table reports the monthly alphas, in percentage points, of various portfolios, estimated using the Fama French 3 factor model, augmented with the momentum factor (UMD). The ﬁrst line presents the monthly alphas of equal-weighted portfolios of ﬁrms who made a signiﬁcant acquisition less than 12, 24, 36 and 48 months ago. The second line looks at the portfolio of past acquirers whose level of internal governance was above median, before acquisition. The third line looks at the portfolio of past acquirers whose level of internal governance was below median, before acquisition. The fourth line looks at the portfolio long in high governance acquirers and short in low internal governance acquirers. T-statistics are between brackets.

is suﬃciently large (last 24, 36 or 48 months), and less so in the short run (last 12 months). This is to some extent consistent with evidence from Table 2.6, where the diﬀerence in value destroyed widens over time.

2.5

External Versus Internal Governance

We have shown that our measure of “internal governance” correlates well with (1) overall corporate performance and (2) the eﬃciency of some crucial strategy choices (acquisitions). However, one possible story consistent with such evidence is that we are proxying for corporate governance in the “traditional” sense: ﬁrms with weak shareholders, weak boards and imperial CEOs could also be the ones where the CEO has all the power to appoint faithful executives. Hence, a well-entrenched CEO is more likely to replace executives who do not show suﬃcient loyalty, which makes our IG index rise. At the same time, weak boards do not have the means to oppose large, wasteful acquisitions. This alternative story puts external governance back to the fore: when “external” governance is poor, the ﬁrm performs less well, and most executives have had less time on the job than the CEO. If this were true, however,

Bottom-Up Corporate Governance

59

Figure 2.3: Cumulative Returns on Portfolios Sorted by Internal Governance Index

the existing literature on “external” governance would have had no trouble in ﬁnding a positive statistical relation between corporate performance and measures of governance quality. Existing contributions have repeatedly failed to ﬁnd a positive correlation between the share of outsiders in the board and proﬁtability (see Baghat and Black (2003), and also Hermalin and Weisbach (2003) for a survey). Using corporate charter-based governance measures, Gompers, Ishii, and Metrick (2003) do not ﬁnd a consistent correlation between investor-friendly ﬁrm-level institutions and operating performance. Thus, the available evidence casts doubts on internal governance as a proxy of external governance in our regressions. We look directly at the correlations between our measure of internal governance and some measures of “external” governance that are used in the literature. To do this, we regress our internal governance index on (1) the Gompers-

Bottom-Up Corporate Governance

60

Ishii-Metrick index of governance, which takes large values for management-friendly corporate charters, (2) the fact that the CEO is also the chairman of the board, which measures the CEO’s degree of power on the board (see, for example, Adams, Almeida and Ferreira (2004)), (3) the size of the board (Yermack (1996) shows that ﬁrms with large boards are less eﬃcient), (4) the share of current employees, and (5) of past employees as corporate directors. The ﬁrst measure is available for a subset of our main sample - the largest ﬁrms. The second measure is available for our whole sample as it is extracted from EXECUCOMP. The third, fourth, and ﬁfth measures are extracted from IRRC’s boards and directors database and so available only for a subsample of our main dataset. Overall, the evidence is not consistent with internal governance being a proxy of external governance. Regression results, controlling for both ﬁrm level variables and mechanical correlates of IG, are reported in Table 2.8. Columns 1-3 include the external governance indexes separately, while column 4 combines all of them. Some results point slightly toward a correlation between the two governance measures. Our index is correlated with the charter based GIM index (the coeﬃcient is small and signiﬁcant at 5%). Also, internal governance is worse when the CEO is chairman, suggesting that CEOs who are powerful inside the ﬁrm are also powerful in the boardroom. The only other signiﬁcant relation is more surprising: internal governance turns out to be better when there are more employees sitting on the board of directors. The literature on independent directors reports this correlation: it is usually interpreted as evidence that bad performing ﬁrms tend to appoint outsiders on the board (Hermalin and Weisbach (2003), Kaplan and Minton (1994)). One other, more daring, interpretation of this negative correlation between internal and external governance is the following. The particularity of these board members (employees) is their intimate knowledge of human capital and the power struggles within the ﬁrm. Insiders sitting on the board therefore have enough information about the competence of executives to eﬃciently interfere with the CEO in the nominating process. By preventing the CEO from appointing new subordinates, they enforce a good level of internal governance. This interpretation does, however, reverse the conventional wisdom on employee-directors. Table 2.8 suggests there might be some weak correlation between internal and external governance. We thus provide new estimates of equation (2.2) in Table 2.9 including an external governance measure as further control. Columns 1-3 focus on ROA as a measure of performance, while columns 4-6 look at the eﬀects on the market valuation of assets. Columns 1 and 4 include the GIM index only, and the ﬁrm controls and mechanical correlates of (2.2). Columns 2 and 5 add our internal governance IG index. Columns 3 and 6 include the other external governance indexes. Consistent with Gompers, Ishii and Metrick (2003), the GIM index is negatively correlated with market to book,

Bottom-Up Corporate Governance

61

Table 2.8: Are Internal and External Governance Related ?

Internal Governance Index (×100) (1) (2) (3) (4)

GIM Governance index

0.3∗∗ (0.2)

-

-

0.1 (0.2)

CEO is Chairman

2.0∗∗ (0.9)

-

0.4 (1.2)

Board size

-

-

-0.2 (0.2)

-0.1 (0.2) -8.5∗ (4.4) 1.5 (5.6)

Frac directors who are current employees Frac indep. directors who are former employees

-

-

-11.5∗∗∗ (3.7)

-

-

0.2 (4.9)

CEO/Firm controls 48 industry dummies Year dummies

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Observations

4,734

7,432

3,447

2,610

Source: OLS estimates, allowing for heteroskdastic residuals. Internal governance index (see table 2.1) is regressed on various corporate governance indicators, controlling for log(assets), log(ﬁrm age), sales growth, 48 industry-dummies, year ﬁxed eﬀects, CEO tenure and executive seniority. Columns 1 to 4 add various corporate governance controls. Column 1 uses the (mostly) corporate charter-based corporate governance index from Gompers, Ishii and Metrick (2003). Column 2 uses the number of directors on the board as a measure of board eﬀectiveness. Columns 3 uses two classical measures of board dependence to the CEO: the share of currently employed directors and the share of past employees. Column 4 uses all four measures simultaneously. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance.

Bottom-Up Corporate Governance

62

Table 2.9: Internal Versus External Governance

Return on Assets (×100) (1) (2) (3) (1)

Market To Book (2) (3)

Internal governance (delayed by 1 year) GIM governance index

-

-6.9∗∗∗ (1.9)

-5.7∗∗∗ (2.3) -0.0 (0.1) 0.2 (0.8)

-

-0.87∗∗∗ (0.26)

-0.88∗∗∗ (0.33) -0.00 (0.02) -0.21∗ (0.11)

-0.1 (0.1)

-0.1 (0.1) -

-0.02∗ (0.01) -

-0.02 (0.01) -

CEO = Chairman

-

Board size (# directors)

-

-

-0.1 (0.1)

-

-

-0.02 (0.02)

% Directors currently employed % Directors previously employed

-

-

7.4∗∗∗ (2.7)

-

-

-0.49 (0.41)

-

-

3.3 (3.7)

-

-

-0.51 (0.50)

Firm/CEO controls Year eﬀects 48 Industry eﬀects

yes yes yes

yes yes yes

yes yes yes

yes yes yes

yes yes yes

yes yes yes

Observations

4,634

3,933

2,274

4,416

3,495

1,977

Source: Huber-White-Sandwich estimates, allowing for correlation of all observations of a given ﬁrm. The measure of internal governance is the share of EXECUCOMP executives who joined the company after the CEO was appointed. Corporate performance is measured through Return on Assets (ﬁrst three columns) and through Return on Equity (last three columns). All regressions use as controls: CEO and executive seniorities, sales growth, log(book assets), log(Firm age), year dummies and the 48 Fama French industry dummies. Columns 1 and 4 use the corporate charter based corporate governance index from Gompers, Ishii and Metrick (2003). Columns 2 and 5 use the number of directors on the board as a measure of board eﬀectiveness. Columns 3 and 6 use two classical measures of board dependence on the CEO: the share of current and past employees serving as directors. The limited availability of corporate governance data is responsible for the drop in observation number. ∗ , ∗∗ and ∗∗∗ means statistically diﬀerent from zero at 10,5 and 1% level of signiﬁcance.

but not with operating performance. However, this correlation with market to book disappears once we include the index of internal governance. To be fair, the coeﬃcient estimate becomes noisier, but not smaller, partly because the number of observations where our index and the GIM index overlap is less than 5,000. Notice that in columns

Bottom-Up Corporate Governance

63

2 and 5, the coeﬃcient on internal governance is identical to some results in Table 2.2. Finally, the inclusion of the other external governance indexes shows that (1) most are not really correlated with corporate performance, which is consistent with the existing literature, (2) the share of inside directors is positively correlated with performance and (3) the eﬀect of our index remains unaﬀected by the inclusion of these controls, even though they considerably reduce the sample size. Before concluding, we run a similar horse race between external and internal governance for our acquisition returns results. The simplest way to do this is to ask whether after an acquisition, the long-run stock returns of acquirers with poor external governance underperform those of acquirers with good external governance. To do this, we repeat the exercise in section 2.4 by computing long-run cumulative abnormal returns for acquirers with an above- and below-median GIM index. Average long-run returns are reported in columns 4-6 of Table 2.6, as well as in ﬁgure 2.4. Column 4 displays returns to long-term shareholders of management-friendly companies, column 5 does the same for shareholder-friendly companies. Column 6 computes the diﬀerence, and tests for equality. As is apparent from both the Table and the ﬁgure, the two subgroups display strong negative returns for large acquisitions (10-15% after four years); the diﬀerence between them, however, is small and insigniﬁcant statistically.

2.6

Conclusion

This paper argues that the careful design of the chain of command within the ﬁrm aﬀects the eﬃciency of the decisions that are taken. Our informal argument is that independently-minded executives always impose more constraints on the CEO than executives who owe him their jobs. These constraints may prevent controversial decisions from being taken, and have in general the useful eﬀect of de-biasing the CEO. To do this, top executives do not have to formally disobey, or enter in open conﬂict with their boss: they may simply choose to be less enthusiastic in their work. Such an argument is explored theoretically in a companion paper (Landier, Sraer and Thesmar (2007)). We believe an important contribution of this paper is to exhibit an organizational ﬁrm-level variable with strong systematic predictive power on future performance. Accordingly, we also believe that our study has important normative implications for practioneers of corporate governance and ﬁrm organization. First, our statistical analysis proves that the intensity of internal governance can be at least partly observed in existing datasets and could be included in the various indexes of the quality of a ﬁrm’s corporate governance. This implication does not depend on our interpretation of our results: be it the sign of executives “leaving the sinking boat”, of an autocratic CEOs, or of the healthy discipline of having to convince one’s subordinates, the share of independent executives as we measure it predicts performance. Furthermore, our paper suggests that, in addition to management monitoring and advising,

Optimal Independence in Organizations

64

Figure 2.4: Long-Run Returns From Acquisitions: Good vs Poor External Governance

a key role of the board should consist in designing the optimal balance of power within the ﬁrm. Put otherwise, the human resource role of the board is not limited to the usually emphasized CEO succession problem.

Chapter 3

Optimal Independence in Organizations
“Workers do, and managers ﬁgure out what to do.” F. Knight (1921)

3.1

Introduction

A key role of managers in organizations is decision making. Yet, as pointed out by Knight (1921), a project is rarely implemented by the manager who has selected it. This “separation of implementation and control” is not innocuous for decision making. “Implementers” may have intrinsic distastes over selected projects or may simply not adhere to the manager’s vision for the ﬁrm. Such “natural” reluctance to carry out selected projects may not manifest as an open conﬂict, but rather as an under-provision of implementation eﬀort. This paper explores theoretically the existence of such “implementation constraints” and relates them to organizational eﬃciency. The insight that decision makers need to internalize Implementers preferences is well recognized in the practitioner management literature. Arguably, it is one of the key messages of Alfred Sloan’s (1963) autobiography, “My Years with General Motors”. In chapter 5, Sloan relates the story of the “copper-cooled engine” a project that raised the enthusiasm of GM’s managers but failed to raise the support of the line-engineers in charge of implementing it. Their lack of motivation in implementing the innovation resulted in failure, at a very large cost for the company. Sloan quotes his own analysis of the situation in 1923, at the core of the crisis: “We feel that [...] forcing the divisions to take something they do not believe in [...] is not getting us anywhere. We have tried that and we have failed.” Surprisingly, this role of “implementers” as a constraint to decision making has not been explored in the theory of organizations. Of course, the idea that managers and their subordinates may have conﬂicting preferences is certainly not new to the economic literature. An extensive body of research has focused on the role of moral hazard

65

Optimal Independence in Organizations

66

in organizations, analyzing situations where “implementers” have private information about the eﬀort they provide to exert a speciﬁc task (Calvo and Wellisz (1979)). Another strand of the literature has dealt with decision making problems in a principal-agent setting where the agent has private information about the “right” decision to make (see Simon (1957) and more recently Aghion and Tirole (1997) or Dessein (2002)). Finally, a last part of the literature has been trying to design mechanisms aiming at directly reducing the divergence in preferences in such decision making situations (by, e.g. deﬁning a narrow strategy, as in Rotemberg and Saloner (1994), or a clear managerial vision, as in Van Den Steen (2004)). Whether studying decision making or task implementing problems, this entire literature shares the view that preference heterogeneity within the principal-agent relationship is, almost by deﬁnition, harmful to organizational eﬃciency: an eﬃcient organization should always be made of “clones” of the principal. However, preference heterogeneity may prove useful once one starts to acknowledge the “division of labor” among (1) those who make decisions and (2) those who have to implement them. We thus consider an organization of two employees with diﬀerent functions: a Decision Maker (she) in charge of selecting a project, and an Implementer (he) in charge of its execution. Both individuals have intrinsic and possibly diﬀering preferences over projects. Furthermore, there is an informational divide within the organization: on the one hand, the Decision Maker has superior information on the most likely successful project (because, e.g., she goes to meetings, conferences, and has access to conﬁdential memos) ; on the other hand, the Implementer’s provision of eﬀort is not observable. The key feature of this set-up is that the Decision Maker has to anticipate the eﬀort the Implementer is willing to provide on each particular project. An independent Implementer (i.e. an implementer with intrinsic preferences unrelated to those of the decision maker) is more likely to be reluctant to work on the Decision Maker’s preferred project. Anticipating this, the Decision Maker has to use more “objective” information in her decision process and to take less account of her own preferences, which raises the organization’s proﬁtability. This “implementation constraint” has in turn an important consequence on the Implementer’s motivation. Because project’s success matters for the Implementer, he is willing to provide more eﬀort when he believes the Decision Maker is taking an informed decision and not a self-serving one. An independent Implementer – who knows that his mere presence increases the “objective” information used in the decision making process – holds stronger beliefs on the project’s proﬁtability, which makes him spend more eﬀort on the project implementation. Preferences’ independence along the chain of command comes, however, at a cost. As the Decision Maker is more uncertain, by deﬁnition, about the nature of the independent Implementer’s preferred project, she more often

Optimal Independence in Organizations

67

selects projects that are ex post disliked by these Implementers. Therefore, an independent Implementer is more often confronted with projects he does not have intrinsic preferences for, harming his motivation to execute the project. The trade-oﬀ we exhibit in this paper is therefore one between (1) more proﬁtable, “objective”, projects selected and (2) less intrinsically motivated agents ex post. As we show, when the Decision Maker’s private information is suﬃciently precise, the optimal organization features independent Implementers to provide her with incentives to use this information. In our hierarchical setting, heterogeneity in preferences may therefore be beneﬁcial to the organization, but for diﬀerent reasons than in “horizontal” structures like committees or parliaments. In such structures, diversity might be desirable, as it allows individual biases to “cancel each others out”. In the hierarchical organization we study, optimal independence emerges as a mechanism to make the “implementation constraint” more binding, which, under certain condition, is beneﬁcial to organizational eﬃciency. This interaction between decision making and implementation is at the heart of the trade-oﬀ we highlight and allows us to derive interesting comparative static properties that could not be obtained in a more “horizontal” model. Our work is related to some recent literature on organizational design, our main innovation being the study of independence in a “division of labor” framework. Zabojnik (2002) is the only paper to acknowledge the separation between decision making and implementation, but the organization he considers is composed with only extrinsically motivated agents and his focus is on the role of delegation of authority within the hierarchy. Dessein (2002) presents a model of communication between a principal and her agent in a pure decision making situation. As a result of this “task homogeneity”, he obtains that communication is very ineﬃcient, which stands in sharp contrast with our own results. Dewatripont and Tirole (2005) introduce a model of costly communication where homogeneity in preferences may be detrimental to organizational eﬃciency. While Dewatripont and Tirole focus, as we do, on the link between congruence and decision making, their theory relies on the potential free-riding issues that may appear between the sender and the receiver along the communication process. Another feature of our model is that it endogenizes the Decision Maker’s credibility with respect to the Implementer. In signaling models (e.g. Hermalin (1997) in the organization literature and Cukierman and Tomasi (1998) in political economy), an informed principal often manages to send credible messages using some “money burning” devices. In our model, the ability to engage in credible message sending relies on the equilibrium organizational form: independent Implementers helps the Decision Maker to increase the informational content of her selection process, enhancing in turn her decision’s credibility. We ﬁnally discuss at length three applications of our theory. We ﬁrst explore the current debate on ”corporate

Optimal Independence in Organizations

68

governance”. This debate has, so far, mainly focused on the role of boards of directors in improving corporate decision making. Yet, ﬁnancial economists have found very little empirical evidence that directors aﬀect ﬁrm behavior. Our analysis suggests that preferences heterogeneity within the chain of command – for instance between ﬁrm’s CEO and its top executives – could be an eﬀective device to improve performance, at least when private information held by the CEO matters for ﬁrm proﬁtability. Our paper also sheds light on the optimal degree of political independence of government agencies. The management literature suggests that government agencies should be as independent as possible from political power (see, for instance, Horn (1995)). Such a recommendation could be easily derived from our model, where the politicians (the Decision Maker of our model) is biased but has privileged information about social demand. The Implementer (the agency) can be either aﬃliated to the political power or be constituted as an independent organization. Our theory suggests that when social demand is critical to social welfare (e.g. ﬁeld knowledge about the acceptance of reform), a neutral bureaucracy helps eliciting less biased reforms from the politicians. We conclude our analysis by investigating the role of uncertainty in the model. This comparative static is motivated by the large managerial literature insisting on the vital need to organize ﬁrms for change. We investigate how a ﬁrm’s optimal strategy relates to its environmental turbulence and whether change should come from the top or the bottom of the hierarchy. We do so by deriving an extension of the model where one of the project (the “status quo”) is a priori more likely to be proﬁtable than the other ( the “change”). In a low-uncertainty environment, we ﬁnd that ﬁrms’ optimal organization should be “monolithic”, i.e. composed of both pro- “status quo” Implementer and pro- “status quo” Decision Maker. The remainder of the paper is organized as follows. Section 3.2 exposes the set-up of the model and discusses its diﬀerent assumptions. Section 3.3 ﬁrst explores the various equilibria of the decision making game and then turns to the question of optimal organization design. Section 3.4 highlights our key assumptions by doing various robustness checks on our hypotheses. Section 3.5 explores three applications of our theory in corporate governance, bureaucratic organization and product market turbulences. Section 3.6 concludes with leads for further research.

3.2
3.2.1

The Model
Set-Up

The organization belongs to an Owner who seeks to maximize expected proﬁts. It has two employees: a Decision Maker (she) and an Implementer (he). The Decision Maker selects a project and the Implementer implements it.

Optimal Independence in Organizations

69

Project Structure
There are two projects, labeled 1 and 2.1 There are also two equally likely states of nature (θ), also labeled for convenience 1 and 2. Projects can either fail, in which case they deliver 0 to the ﬁrm’s Owner, or succeed and deliver a proﬁt R. The ﬁrst condition for a project to be successful is that it is ﬁts the state of nature: to be successful, project i ∈ 1, 2 must be selected when state of nature is i. Once selected, a project must be implemented by the Implementer. There is moral hazard at the implementation stage: the Implementer has to choose an implementation eﬀort e, which is assumed to be unobservable and discrete (e ∈ {0, 1}). Exerting high eﬀort (i.e. e = 1) entails a private, non-transferable, cost c to the Implementer. c is ˜ ˜ random and is a priori distributed according to a c.d.f. F (.). F is deﬁned on R+ and is supposed to be strictly increasing and concave common knowledge. We make the extreme assumption that project selection and Implementer’s eﬀort are perfect complements: to be successful, the eﬀort level of the Implementer must be high (e = 1) and the good project must be selected (i.e., project i in state of nature i). This is done only to clarify exposition and without loss of generality so long as project selection and implementation eﬀort are weak complements in the production function (an assumption similar to that made in Dewatripont and Tirole (2005)). The Decision Maker has superior information on the state of nature. More precisely, we assume that she receives a binary private signal σ ∈ {1, 2} on the state of nature, such that: P(σ = “i |θ = i) = α > 1 , for all i = 1, 2 2
23

. Moreover, as F is a c.d.f. function, we have that F (0) = 0. Finally, note that F (.) is

While the signal σ is private information to the Decision Maker, its precision α is common knowledge.

Agents’ Utilities
The Owner is risk-neutral and maximizes expected proﬁt. To simplify exposition, we ﬁrst assume that monetary incentives cannot be oﬀered, because, for instance, agents are inﬁnitely risk averse on the monetary part of their
1 Our qualitative results do not depend on the number of possible projects. Anticipating our results below, we remark that, if there were more than 2 possible projects, it would still be the case that the Implementer and the Decision Maker may have conﬂicting intrinsic preferences on the project’s set. This discrepancy would still compel the Decision Maker to assign a smaller weight to her own preferences and more weight to objective, private information. Thus, dissent would still foster organizational reactivity. 2 An alternative modeling choice would consist in assuming that the Implementer exerts a continuous level of eﬀort e ∈ [0, 1), which yields, when the appropriate project has been selected (i.e. project i in state i), a probability of success e at a cost C(e), where C() is a convex, strictly increasing function deﬁned over [0, 1). Both modeling choice are equivalent. In particular, the assumption that F is concave is exactly equivalent to the assumption that C () > 0. 3 The technical assumption that F is concave is not necessary for most of the analysis, but it helps us compute most of the comparative statics of the model (see below).

Optimal Independence in Organizations

70

utility (as in Aghion and Tirole (1997)). Thus, the Decision Maker and the Implementer derive utility only from private beneﬁts attached to the successful completion of a project. Discussion on monetary incentives is deferred to Section 3.4.2. The Decision Maker obtains private beneﬁt B > 0 (resp. B > 0) when her most (resp. least) preferred project is ¯ chosen and succeeds (with B > B > 0). When the project fails, she receives no private beneﬁt at all. In order to ﬁx ideas, but without loss of generality, we will assume throughout the paper that the preferred project of the Decision Maker is project 1. We also assume that it is public information that the Decision Maker prefers project 1. The Implementer also obtains a private beneﬁt ¯ (resp. b) when his most (resp. least) preferred project is chosen b and succeeds. However, in case of project failure, he ends up with no private beneﬁt. Finally, we deﬁne a congruence parameter β as the ex ante probability that both Decision Maker and Implementer share the same preferred project (i.e. project 1). We interpret β as a measure of organizational homogeneity (or congruence of tastes). Organizational design by the organization’s Owner boils down to the choice of β (except in sections 3.4.1 and 3.4.2 where we look at questions of delegation and wage contracting). Throughout the paper, we assume that, at the project selection stage, the Decision Maker ignores the Implementer’s preferred project and only know the congruence parameter β.

Sequence of Events and Information Structure
The sequence of events has four stages: 1. Organizational design: The Owner of the ﬁrm chooses the level of congruence β, deﬁned as the probability that the Implementer will, as the Decision Maker, prefer project 1. β then becomes public information. 2. Decision making: The Decision Maker receives her private signal σ, with precision α, about the state of nature. She then selects a project. This choice is assumed to be irreversible.4 At the time of project selection, the Decision Maker knows β but ignores the Implementer’s preferred project. 3. Implementation: The Implementer is hired and his preferred project, as well as his implementation cost c, ˜ are revealed. He has to implement the project selected in period 2, but can decide whether or not to exert eﬀort. of time pressure, for instance, of because a large project-speciﬁc investment needs to be made in order to launch the project. The purpose of this assumption is to prevent the Implementer from communicating his preferences to the Decision Maker. It can be shown that such communication weakens, but does not destroy, the role of dissent. We return to this issue at the end of Section 3.3.3.
4 Because

Optimal Independence in Organizations

71

4. Outcome: The project either succeeds (yielding proﬁt R to the organization and private beneﬁts to the agents) or fails (proﬁt and private beneﬁts are then equal to 0). The corresponding time line is drawn in Figure 3.1.

1
Principal chooses the congruence parameter

2
Decision maker: • receives private signal • orders the project to implement

3
• Cost of effort and type of the implementer are revealed •Implementer chooses level of effort e

4
Payoffs are realized

β

Figure 3.1: Timing of the model

3.2.2

Equilibrium Concept

We look for standard Perfect Bayesian Equilibria of our game. Formally, an equilibrium is deﬁned by two equilibrium strategies (O, µ). O() is the Decision maker’s selection process, i.e. a function that maps the Decision Maker’s private signal σ ∈ {1, 2} into the project space {1, 2}: O : σ ∈ {1; 2} → {1; 2} µ() is the Implementer’s posterior belief function, i.e. a function that maps the project selected by the Decision Maker into a probability that state of nature is 1: µ : O ∈ {1; 2} → P(θ = 1|O) ∈ [0, 1) A perfect Bayesian equilibrium of the game is a couple (O, µ) that veriﬁes: 1. Individual rationality: Given the Implementer’s posterior belief function µ(.), O (σ) maximizes the Decision Maker’s utility for all σ ∈ {1, 2}. 2. Bayesian updating: The posterior P(θ = O|O) is obtained using the order function O(.), the Implementer’s prior about state θ and Bayes’ law.

Optimal Independence in Organizations

72

Our model is therefore similar to a standard signaling game (see, e.g., Spence (1973)), with an informed principal (the Decision Maker who knows the true value of the signal) and an uninformed agent (the Implementer who does not observe this signal). As with general signaling games, there are many equilibria in our model if we do not impose any restrictions on the Implementer’s ex post beliefs when an “unexpected” project is selected. The standard reﬁnement of beliefs in the signaling literature is the notion of strategic stability, introduced by Kohlberg and Mertens (1986). We will use in this paper a weaker reﬁnement, known as D1 (see Cho and Sobel (1990]), which is suﬃcient to identify a unique equilibrium in our basic model. Intuitively, the D1 reﬁnement requires that if the Implementer observes an out-of-equilibrium project selection, he should believe this choice came from the type “most eager” to make the deviation. Therefore, our solution concept is Perfect Bayesian Equilibria reﬁned by the D1 criterion.

3.3

Organizational Homogeneity aﬀects Reactivity

This section derives conditions on the congruence parameter β for the existence of the various potential equilibria of our model. As usual in standard signaling games, we ﬁrst assume equilibrium strategies and then ﬁnd the conditions under which these strategies deﬁne an equilibrium as deﬁned in section 3.2.2 Before turning to equilibrium strategies, we ﬁrst look at the Implementer’s eﬀort choice (section 3.3.1), conditional on a given posterior belief and then derive the diﬀerent equilibria of the model. Finally, we move to organizational design (section 3.3.4), i.e. we determine the level of congruence β that maximizes the Owner’s proﬁt.

3.3.1

Implementer’s Eﬀort Choice

In this section, we take as given the Decision Maker’s selection process and the Implementer’s posterior belief and focus on the Implementer’s choice of eﬀort. Therefore, let µ1 (resp. µ2 ) be the Implementer’s posterior belief that the state of nature is 1 when the Decision Maker has selected project 1 (resp. 2). Consider the case where the Decision Maker selects project 1. If the Implementer’s preferred project is project 1, he will put in high eﬀort (i.e. e = 1) whenever:

µ1¯ − c ≥ 0 b However, if the Implementer’s preferred project is project 2, he only exerts the high level of eﬀort on project 1 when: µ1 b − c ≥ 0

Optimal Independence in Organizations

73

From an ex ante point of view, the Implementer should prefer project 1 with probability β. Thus, in period 2, at the decision making stage, the Decision Maker knows that high eﬀort will be exerted on project 1 with probability:

P(e = 1|O = 1) = βF (µ1¯ + (1 − β)F (µ1 b) b) High eﬀort on project 1 is more likely to be exerted when (1) the Implementer believes 1 is the state of nature and (2) the Implementer is more likely to prefer project 1. Similarly, selecting project 2 should result in the following probability of eﬀort by the Implementer:5 P(e = 1|O = 2) = βF ((1 − µ2 )b) + (1 − β)F ((1 − µ2 )¯ b)

3.3.2

The Decision Maker’s Project Choice

We now derive the diﬀerent equilibria of our model. As usual in standard signaling games, there are three types of equilibria: (1) a separating equilibrium, where the Decision Maker selects diﬀerent projects for diﬀerent signals received (2) a pooling equilibrium, where the Decision Maker selects the same project whatever the nature of her private information and (3) semi-pooling equilibrium, where the project selected only partially reveals the nature of the Decision Maker’s private information. This section determines conditions on the congruence parameter β for the existence of these equilibria. Reactive Equilibria We start by looking for separating equilibria. In such an equilibrium, the Decision Maker will select a diﬀerent project for each signal she receives. The natural candidate for such an equilibrium consists in the Decision Maker always selecting the project indicated by the signal, i.e. O(σ) = σ. Conditional on this strategy, the Implementer’s equilibrium posterior beliefs are then uniquely determined by Bayes’ rule:

µ1 = 1 − µ2 = α For this to be an equilibrium, there are two incentive constraints that need to be satisﬁed (i.e. the individual rationality constraints in state 1 and in state 2): when receiving signal i, i ∈ {1, 2}, selecting project i should maximize the Decision Maker’s utility. Consider the ﬁrst such incentive constraint, imposing that when facing a signal 2, the Decision Maker should prefer to select project 2:
5 Remember that µ is the posterior belief that 1 is the state of nature when project 2 has been selected, so that 1 − µ is the posterior 2 2 belief that 2 is the state of nature in this same situation.

Optimal Independence in Organizations

74

α
Proba. of state 2

. βF (αb) + (1 − β)F (α¯ . b)
Proba. of eﬀort with order 2

B
Low Decision Maker beneﬁt

≥

(1 − α)
Proba. of state 1

. βF (α¯ + (1 − β)F (αb) . b)
Proba. of eﬀort with order 1

¯ B
High Decision Maker beneﬁt

(3.1)

This incentive constraint involves the following trade-oﬀs. A priori, the Decision Maker’s own preferences induce ¯ her to select project 1 instead of project 2 (as B > B). However, when the private signal indicates state of nature 2, project 2 becomes more likely to succeed, which makes project 2 more attractive for the Decision Maker (as α > 1 − α). Finally, a more congruent Implementer will have higher incentives to exert eﬀort on project 1, so that a higher congruence β will make the Decision Maker more likely to choose project 1. We stress that the Decision Maker’s incentives to select project 2, conditional on having received signal 2, are always lower than that of the (unbiased) Owner. Indeed, the owner’s incentive constraint would be exactly similar to that of an unbiased ¯ Decision Maker, i.e. similar to equation (3.1) with B = B = R. Such a constraint would clearly be less binding that constraint (3.1). Overall, the above condition (3.1) simply states that the Implementer and the Decision Maker should not be too congruent if this “reactive” equilibrium is to hold. This condition is equivalent to: β≤ ¯ αF (α¯ − (1 − α)F (αb)B b)B ¯ − F (αb) (1 − α)B + αB = β2 ¯ F (αb) (3.2)

When congruence β is high, the Decision Maker ﬁnds it costly not to select project 1, as she anticipates that the Implementer is very likely to prefer this particular project. Such incentive to “pander” to the Implementer’s preferences may be too strong to allow the reactive equilibrium to emerge (i.e. when β ≥ β2 ). Note that in some cases, β2 as deﬁned by equation (3.2) can be either superior to 1 or negative. For instance, when the signal is very informative (i.e. α close to 1), the Decision Maker is never willing to select a project diﬀerent from the signal she received, as this would lead the organization to a certain failure so that β2 > 1; similarly, when the low private beneﬁt B tends to 0, the Decision Maker has no incentives to ever select project 2, so that β2 < 0. With a slight abuse of notation, we deﬁne:

∗ ∗ β2 ≡ min{1, max{0, β2 }}

(3.3)

The second incentive constraint is the symmetric of constraint (3.1): facing a signal 1, the Decision Maker must choose project 1 . From this condition emerges a diﬀerent trade-oﬀ. When β is very low, it becomes very likely that the Implementer dislikes project 1 so that he has less incentives to provide eﬀort on this project. Therefore,

Optimal Independence in Organizations

75

the Decision Maker may be tempted to ignore both her private signal and her own intrinsic preferences, that both provide her with incentives to select project 1, in order to conform to the Implementer’s own inclination to project 2. Formally, the second incentive constraint is written as a lower bound on β:

β≥

¯ (1 − α)F (α¯ − αF (αb)B b)B ¯ − F (αb) αB + (1 − α)B = β1 ¯ F (αb)

(3.4)

∗ It may be the case that β1 , as deﬁned in equation (3.4), is negative, when, for instance, the signal is very

b informative (i.e. α close to 1) or when the Implementer’s bias is moderate (i.e. b/¯ close to 1). In such cases, this second incentive constraint is never binding as conforming to the Implementer’s preferences either (1) leads the organization to a certain failure (when α is close to 1) or (2) increases only marginally the Implementer’s provision b of eﬀort (when b/¯ close to 1). Note, however, that β2 is always inferior to 1: with a perfectly aligned Implementer, the Decision Maker has no reason to ever select project 2 when her private signal is 1, as project 1 is more likely to succeed and is both the Implementer’s and the Decision Maker’s preferred project. With a slight abuse of notation, we deﬁne:

∗ ∗ β1 ≡ max{0, β1 }

(3.5)

Proposition 1 summarizes the results of this section: Proposition 1 There is only one type of separating equilibrium in our model. In this equilibrium, the Decision Maker always select the project indicated by her private signal: this is a reactive equilibrium. This equilibrium exists whenever : ¯ B α F (α¯ b) ≤ , (3.6) B 1 − α F (αb) i.e. when (1) the signal is suﬃciently precise or (2) the Implementer’s bias is high enough or (3) the Decision Maker’s bias is low enough. When condition (3.6) is veriﬁed, the reactive equilibrium exists for all values of the congruence parameter β lying in [β1 , β2 ], where β1 and β2 are two thresholds, deﬁned by equation (3.3) and (3.5), and such that: 0 ≤ β1 ≤ β2 ≤ 1

Intuitively, the scope for reactivity (i.e. the interval [β1 , β2 ]) is (1) increasing with the signal precision α (2) ¯ decreasing with the Decision Maker’s bias (i.e. B/B) and (3) decreasing with the Implementer’s bias (¯ b/b). Proof See Appendix B.1.1. Non-Reactive Equilibria We now turn to pooling equilibria where the Decision Maker selects the same project, whatever the signal she receives. Our model features two diﬀerent types of such equilibria: one where the Decision Maker always selects

Optimal Independence in Organizations

76

her preferred project, project 1 (i.e. ∀σ ∈ {1, 2}, O(σ) = 1); the other where she always selects project 2 (i.e. ∀σ ∈ {1, 2}, O(σ) = 2). Both these equilibria lead to an organization that does not react to the Decision Maker’s private signal. We ﬁrst consider in detail the case where the Decision Maker’s decision process simply selects project 1 for each type of signal received (i.e. ∀σ ∈ {1, 2}, O(σ) = 1). In this case, on the equilibrium path, the order has no informational content, so that the Implementer’s posterior belief on the state of nature is, conditional on project 1
1 being selected: µ1 = 2 .

Out of equilibrium, any belief µ2 ∈ [1 − α, α] is a priori admissible. Indeed, as project 2 being selected is an out-of-equilibrium event, Bayes’ law does not constrain these beliefs µ2 . Without further restrictions on these out-of-equilibrium beliefs µ2 , our model will naturally admit a lot of pooling equilibria, supported by diﬀerent out-ofequilibrium beliefs. To limit the scope for multiple equilibria, we thus use the D1 reﬁnement (Cho and Sobel (1990]) – see section 3.2.2. Intuitively, this reﬁnement requires that if the Implementer observes an out-of-equilibrium order, he infers that this order was triggered by the signal that would have made the type of decision maker the “most eager” to make the deviation.6 We do show in proposition 2 that Decision Makers receiving a signal 2 are always “more eager” to deviate from the pooling equilibrium we are considering. This is quite intuitive: deviating from the equilibrium implies selecting project 2. Such a deviation is naturally more proﬁtable to a Decision Maker who has received signal 1 rather than signal 2. The only out-of-equilibrium belief satisfying the D1 reﬁnement should thus lead the Implementer to consider that only a Decision Maker having received signal 2 may have deviated from the equilibrium. Formally, this implies that:7 µ2 = 1 − α (3.7)

As is the case with separating equilibrium, there are a priori two diﬀerent incentive constraints that need to be satisﬁed: the Decision Maker must prefer to select project 1 whether her private signal is 1 or 2. As we show in proposition 2, there is only one relevant incentive constraint: the Decision Maker should select project 1 even when receiving private signal 2. It is easy to show that this constraint is satisﬁed when β is large enough. For low levels of congruence β, a Decision Maker facing a signal 2 is indeed tempted to deviate from the equilibrium as (1) project 2 is the Implementer’s preferred project (2) project 2 is the project with the highest probability of success. Formally, it is straightforward to show that there exists a threshold β2 above which this pooling, “non-reactive” equilibrium exists:
6 We 7 The

refer the reader to the proof of proposition 2 for details on the deﬁnition of the D1 reﬁnement. reader should remember that µ2 is the probability that the state of nature is 1 when project 2 has been selected.

Optimal Independence in Organizations

77

β ≥ β2 =

1 ¯ αF (α¯ − (1 − α)F ( 2 b)B b)B 1 ¯ [F ( 1 ¯ − F ( 2 b)][(1 − α)B] + [F (α¯ − F (αb)]αB] b) b) 2

(3.8)

This threshold β2 can be either strictly superior to 1 (in which case this pooling equilibrium fails to exist – this is for instance the case with a perfectly informative signal (i.e. α = 1)) or strictly inferior to 0 (in which case it exists for all values of congruence – this is the case for instance when the Decision Maker is highly biased (i.e. B/B → ∞)). Therefore, we deﬁne, with a slight abuse of notation, the following threshold:

∗∗ ∗∗ β2 ≡ min{1, max{β2 , 0}}

(3.9)

Of course, as we already mentioned, another pooling equilibrium is possible, where the Decision Maker always select project 2 (i.e. ∀σ ∈ {1, 2}, O(σ) = 2). Using the D1 reﬁnement to restrict the choice of out-of-equilibrium beliefs, we can characterize this second, “non-reactive” equilibrium in a similar and symmetric fashion: there exists a threshold in congruence, β1 , below which this equilibrium exists. The intuition is the following: as β decreases, the Implementer becomes more likely to exert eﬀort on project 2. When β is very low, the Implementer’s preference for project 2 becomes so certain that the Decision Maker must select project 2 if she wants some implementation eﬀort to be exerted. More formally, the relevant incentive constraint of the Decision Maker is equivalent to the following condition: ¯ b)B (1 − α)F ( 1 ¯ − αF (αb)B 2 ¯ − F ( 1 b)][(1 − α)B] + [F (α¯ − F (αb)]αB] ¯ [F ( b) b) 2
1 2

β ≤ β1 =

(3.10)

This threshold β1 can happen to be strictly negative, in which case this pooling equilibrium fails to exist. This is the case, for instance, when the signal is perfectly informative (i.e. α = 1), as selecting project 2 after having received a signal 2 would then lead the organization to a certain failure. Note, however, that β1 is always inferior to 1. With a perfectly congruent agent, a Decision Maker facing signal 1 has no incentives to select project 2, as project 1 is (1) more likely to succeed (2) the Decision Maker’s preferred project and (3) the Implementer’s preferred project. With a slight abuse of notation, let us deﬁne the following threshold:
∗∗ ∗∗ β1 ≡ max{β1 , 0}

(3.11)

Proposition 2 summarizes the results derived in this section:

Optimal Independence in Organizations

78

Proposition 2 Our model admits, for each level of congruence, at most one pooling equilibrium. For low values of congruence (i.e. β ∈ [0, β1 ]) the Decision Maker always selects project 2, while, for high values of congruence (i.e. β ∈ [β1 , 1)), she always selects project 1. The two thresholds β1 and β2 , deﬁned by equations (3.9) and (3.11) verify: 0 ≤ β1 ≤ β1 ≤ β2 ≤ β2 ≤ 1 The scope for non-reactivity (i.e. 1 − β2 and β1 ) decreases with the signal precision (i.e. α) and increase with the Decision Maker’s bias and the Implementer’s bias. Proof See Appendix B.1.2.

The above analysis suggests that for some intermediate values of β (i.e., β ∈ [β1 ; β1 ] ∪ [β2 ; β2 ]), neither a “reactive”, nor a “non reactive” equilibrium exist. The following section shows that, in these intermediate cases, the equilibrium features a partially informative decision: the Decision Maker selects with a positive probability the project indicated by the signal. Semi-Reactive Equilibria This section deals with the conditions of existence of a mixed strategy equilibrium where the Decision Maker is allowed to randomize over the project she selects. We show in proposition 3 that there cannot be an equilibrium where the Decision Maker would strictly randomize, whatever the signal she receives, the project she selects. The intuition is that, because the Decision Maker is biased, she cannot be indiﬀerent between the two projects for all values of the signal, except in the extreme case where the signal has no informational content (i.e. when α = 1/2). There are therefore only two potential candidates for such mixed equilibrium, one where only a Decision Maker receiving signal 2 mixes between selecting project 1 and selecting project 2 and one where only a Decision Maker receiving signal 1 mixes between the two project. Let us focus on the ﬁrst type of mixed equilibrium. In such an equilibrium, the Decision Maker’s decision process is deﬁned as:   1 with probability ρ  2 with probability 1 − ρ

O(1) = 1 and O(2) =

In words, in such an equilibrium, the Decision Maker (1) always select project 1 when signal is 1 and (2) select project 2 with probability (1 − ρ) and project 1 with probability ρ when the signal is 2. Of course, ρ is endogenous and will be determined by equilibrium conditions. In the terminology deﬁned above, ρ = 0 corresponds to a reactive equilibrium and ρ = 1 to a non-reactive equilibrium. We now turn to the Implementer’s posterior belief. In such a “semi-reactive” equilibrium, when project 1 has been selected, the Implementer updates his beliefs on the probability that signal 1 was received by the Decision

Optimal Independence in Organizations

79

Maker. Ex post, project 1 may have been selected either because (1) signal 1 was observed or (2) the Decision Maker selected her preferred project in spite of having received a signal 2. Bayes’ rule imposes that the posterior belief be given by:

µ1 (ρ) =

α + ρ(1 − α) 1+ρ

Of course, when project 2 is ordered, it must be that the Decision Maker observed signal 2, so that: µ2 (ρ) = 1 − α In this semi-reactive equilibrium, the Decision Maker randomizes between project when the private signal is 2. For such mixed strategies to be sustainable, the Decision Maker must be indiﬀerent between the two options:

α βF (αb) + (1 − β)F (α¯ B b)
Decision Maker expected utility of selecting with signal

¯ = (1 − α) βF (µ1 (ρ)¯ + (1 − β)F (µ1 (ρ)b) B b) 2
Decision Maker expected utility of selecting with signal

(3.12)

1

2

2

This last equation 3.12 pins down the value of ρ as a function of the other parameters of our model. In particular, equation 3.12 deﬁnes, for each value of the congruence parameter β ∈ [β2 , β2 ], a unique ρ(β) ∈ [0, 1]. When β = β2 , ρ(β2 ) = 1 : the Decision Maker never reacts to the signal and the equilibrium is non-reactive, as shown above. When β = β2 , ρ(β2 ) = 0. The Decision Maker always reacts to the signal, and the equilibrium is fully reactive, consistent with the above analysis. A symmetric analysis can be performed for β ∈ [β1 ∗ , β1 ]. For each value of β in this interval, there is an equilibrium where the Decision Maker (1) always selects project 2 when the signal indicates 2 and (2) selects project 1 with probability 1 − ξ when the signal is 1. For lower values of β in this interval, the Decision Maker becomes more likely to select project 2 when receiving signal 1. We sum up the result of this section if the following proposition: Proposition 3 The only mixed equilibria of our model involve the Decision Maker randomizing strictly only for a particular value of the signal: randomizing strictly for both value of the signal cannot be an equilibrium strategy. Therefore, there are only two “types” of mixed, “semi-reactive”, equilibria: 1. When β ∈ [β1 ∗ , β1 ], the Decision Maker always selects project 2 when the signal indicates 2. When private signal is 1, the Decision Maker nevertheless selects project 2 with probability ξ(β) and project 1 with probability ∗∗ ∗ 1 − ξ(β). ξ() is a decreasing function of β such that ξ(β1 ) = 1 and ξ(β1 ) = 0.
∗ ∗ 2. When β ∈ [β2 , β2 ], the Decision Maker always selects project 1 when the signal indicates 1. When private signal is 2, the Decision Maker nevertheless selects project 1 with probability ρ(β) and project 1 with probability ∗∗ ∗ 1 − ρ(β). ρ is an increasing function of β, such that ρ(β2 ) = 1 and ρ(β2 ) = 0.

Proof See Appendix B.1.3.

Optimal Independence in Organizations

80

3.3.3

Summary and Discussion

Implemented Project by a 1- decision maker 64444744448 4 4
Signal 1 Project 2  Project 1 Project 1 Project 1 1 - ξ : Project1 Signal 2 Project 2 Project 2 Project 2
 ρ : Project 1  1- ρ : Project 2

ξ : Project 2

Project 1

0

β 1**

β 1*

β 2*

* β 2*

1

Figure 3.2: Implemented Project of a 1 Decision Maker according to the Signal Figure 2 summarizes the result of the analysis of section 3.3.2. For each level of diversity, there exists a unique perfect Bayesian equilibrium satisfying the D1 reﬁnement: For low levels of the congruence parameter β (β ∈ [0; β1 ]), project 2, which is the Implementer’s most likely preferred project is always selected. As β increases, the order progressively entails more and more informational content, as the Decision Maker uses her private signal more often in her decision process (i.e. for β ∈ [β1 ∗ , β1 ] ). For intermediate values of β (β ∈ [β1 ; β2 ]), the Decision
∗ ∗ Maker always selects the project indicated by the signal. Then, when β further increases (β ∈ [β2 , β2 ]), the Decision

Maker becomes more reluctant to choose project 2, even in cases where her signal indicates 2, as the Implementer becomes more likely to dislike this project. Finally, when β ∈ [β2 ; 1) , project 1 is always selected, whatever the value of the signal. Of course, some of these equilibria may not exist, depending on the parameters’ value (see proposition 1, 2 and 3 for more details). This suggests that reactivity is easier to obtain for intermediate levels of congruence, while non-reactivity prevails for large or small values of congruence β. Put diﬀerently, uncertainty about the Implementer’s taste, i.e. Implementer’s independence from the Decision Maker, is key to achieving reactivity in our organization. The reason is the following. There are two forces driving the Decision Maker decision rule: she can choose a project with a high, “objective”, probability of success or she can select a project for which the Implementer’s motivation is intrinsically high. The Decision Maker’s incentive to use the signal in her decision rule is all the more important when the Implementer’s own preferences are unknown, which happens for intermediate level of congruence. In other

Optimal Independence in Organizations

81

words, for intermediate level of congruence, the Decision Maker cannot anticipate what project would please the Implementer and therefore bases her decision only on (1) the objective probability of success and (2) her own bias for project 1. As the Implementer’s preferences becomes less uncertain (which happens for congruence level close to 0 or 1), the Decision Maker can now anticipate the eﬀect of her decision on the Implementer’s motivation, which makes the objective signal less important in her decision rule. At this point, we need to stress that reactivity does not just arise because the implementer and the DM’s biases cancel each other out. As we just pointed out, taking an implementer with an opposite bias to that of the DM (i.e. β = 0) is often detrimental to reactivity. As it turns out, it is the uncertainty about the Implementer’s preferences, not his neutrality between the two projects, that plays a key role in promoting reactivity. This is not obvious in our model because intermediate congruence β (i.e. independence) implies both uncertain Implementer’s preferences and an ex ante unbiased Implementer. A simple extension of this model highlights the key role played by uncertainty in our model (see Appendix B.3). Assume that there are no longer 2, but N diﬀerent projects. Also assume that the Decision Maker is unbiased and thus only cares about the probability of success. There are N types of implementers: each type i ∈ [1; N ] gets private beneﬁt b > 0 if project i is completed and succeeds, and 0 else. There is a probability β that the implementer prefers project i = 1. When the signal is j, the right project is j with probability α. The advantage of such a model is that β parameterizes the uncertainty about the implementer’s preferences, not his bias. As we show in Appendix B.3, for high values of β (i.e. low uncertainty on the Implementer’s preferences), the DM has a strong incentive to pander to the implementer’ preferences and reactivity is impaired. We also need to emphasize that the main eﬀect in this paper, namely that uncertainty about the Implementer’s preferences may enhance organizational reactivity, does not depend on the “signaling” aspect of the model (i.e. the fact that project selection may, or may not, be informative for the Implementer). Indeed, in a model where the Decision Maker’s private signal would be public, it would still be the case that the organization be reactive only for intermediate level of congruence, i.e. when the Implementer’s preferences are independent of the Decision Maker.8
Very light calculations reveal that when the signal is public information, the equilibrium features (1) reactivity for h i h i ˜ ˜ ˜ β ∈ β1 , β2 (2) the Decision Maker always selecting project 1 for β ∈ β2 , 1 and (3) the Decision Maker always selecting project 2 for h i ˜ ˜ ˜ β ∈ 0, β1 . β1 and β2 are deﬁned by: V ¯ (1 − α)BF ((1 − α)b) − αBF (αb) bβ = b ˜ ¡ ¡ b ` 1 ¯ − F ((1 − α)b) + αB F (α¯ − F (αb) ¯ (1 − α)B F ((1 − α)b) b) ¯ b αBF (α¯ − (1 − α)BF ((1 − α)b) b) b ˜ b β2 = ¡ ¡ X ¯ (1 − α)B F ((1 − α)¯ − F ((1 − α)b) + αB F (α¯ − F (αb) b) b)
8 footnotesize

Optimal Independence in Organizations

82

3.3.4

Organizational Design

The above analysis suggests that intermediate levels of congruence foster organizational reactivity. Reactivity, however, comes at a cost, as Implementers are more likely to have to implement ex post a project they intrinsically dislike and for which they will exert low eﬀort. To study this tradeoﬀ, this section looks for the optimal congruence β from the Owner’s perspective. For intermediate levels of congruence (i.e. β ∈ [β1 , β2 ]), we know that the Decision Maker equilibrium decision rule is to select the project as indicated by the signal. We can therefore write the expected proﬁt of such an organization as: ∀β ∈ [β1 , β2 ], V R = α F (α¯ + F (αb) b) 2 R (3.13)

This value does not depend on β. The reason for this is simple: because the organization is reactive, the ex ante probability that either project 1 or 2 is chosen is 1/2. Therefore, on average, whatever his preferred project, the Implementer will face with probability 1/2 a project he likes and with probability 1/2 a project he dislikes. Because in such a reactive equilibrium, posterior beliefs that the selected project is the successful one are always equal to α, this implies that, whatever the congruence level, the Implementer will on average exert the same level of eﬀort, namely the average between high eﬀort F (α¯ and low eﬀort F (αb). Thus, provided that congruence lies within the b) reactivity range, there is no particular level of congruence that maximizes the Owner’s proﬁt. Of course, we should stress again that, depending on the level of α and on the Implementer’s and Decision Maker’s bias (i.e. the bs and the Bs), the reactive equilibrium may either (1) fail to exist (when β2 < 0) or (2) always exist (when β1 ≤ 0 and β2 > 1).9 We now turn to non-reactive organizations where the Decision Maker always selects her preferred project. Such an equilibrium, if it exists, is supported by high level of congruence (β ∈ [β2 , 1)). The expected proﬁt of such a non-reactive organization is given by: ∀β ∈ [β2 ; 1), V N R = 1 2 βF ¯ b 2 + (1 − β)F b 2 R (3.14)

Project 1, the Decision Maker’s preferred project, is ex ante successful with probability 1/2. With probability β, action 1 is also the Implementer’s preferred project: in this case, he makes high eﬀort with probability F (¯ b/2). With probability 1 − β, the Implementer prefers project 2 and makes high eﬀort with probability F (b/2) only. In contrast to reactive organizations, the order here never conveys any information. From the implementer’s viewpoint, the posterior belief that the ordered project (always 1) is the successful one therefore remains equal to 1/2.
9 We

refer the reader to proposition 1 for details on the existence of the reactive equilibrium.

Optimal Independence in Organizations

83

As one directly sees, the value of a non-reactive organization (3.14) is maximized for perfect congruence between the Decision Maker and the Implementer (i.e. β = 1). As the organization always implements the same project, here project 1, it is best to ensure maximal implementation eﬀort for project 1, which is best made by hiring a perfectly aligned Decision Maker. Thus, the optimal non-reactive organization is worth: V NR = 1 F 2 ¯ b 2 R (3.15)

As it turns out, the value (3.15) of the optimal non-reactive organization implementing project 1 is similar to the value of the optimal non-reactive organization implementing project 2. The intuition for this is just that in such an organization, as project 2 is always selected on the equilibrium path, it is best suited to have an Implementer that will prefer for certain project 2, which makes him more likely to put in high eﬀort on such a project. The ex ante probability of success of this non-reactive, project-2-oriented, organization is therefore just 1/2 times the probability of eﬀort of the implementer (i.e. F (¯ b/2)), so that the organization’s proﬁt is indeed provided by equation (3.15). As we show in proposition 4, the semi-reactive organizations described in proposition 3 are never optimal from the Owner’s perspective, provided that a reactive organization can be put in place (i.e. that β2 ≥ 0). The intuition is the following: consider, for instance, a value of β for which the Decision Maker’s decision rule strictly randomizes between project 1 and 2 when her private signal indicates 2 (β ∈ [β2 , β2 ]). In this interval, the Decision Maker is, by deﬁnition, indiﬀerent between following signal 2 or not. Given that the Decision Maker intrinsically prefers project 1, this must imply that, for such values of congruence β, the ex ante unbiased Owner would strictly prefer project 2 to be undertaken. Thus, the shareholder can increase ﬁrm value by reducing β, which fosters reactivity, ρ, and thus increases the likelihood of project 2 being selected. Therefore, we can infer from this argument,10 that in this range of congruence supporting semi-reactive equilibria, organization’s value is increasing with congruence,
∗ so that the optimal level of congruence in this range is the extreme β = β2 , where the probability of no reactivity

is zero and the equilibrium becomes fully reactive. Overall, as long as both types of equilibria exist, the Owner’s maximization program boils down to a simple tradeoﬀ between a non-reactive organization (obtained with extreme congruence, i.e. β equal to 0 or 1) and fully reactive organizations (obtained with any intermediate β ∈ [β1 , β2 ]). The net gain of reactivity over non- reactivity((3.13) (3.15)) can be broken down into three terms, highlighting the trade-oﬀs between these two organizational forms:
10 We

provide a formal proof along with proposition 4.

Optimal Independence in Organizations

84

V R − V NR =

α−

1 2

α F (α¯ − b) F (α¯ − F (αb) + b) 2
Cost of Mismatch, < 0

¯ b F (α¯ − F ( 2 ) b) 2 Credibility Gain, > 0

(3.16)

Reactivity Gain, > 0

Let us brieﬂy comment on this tradeoﬀ: • Reactivity gain: This gain comes from the fact that from an ex ante point of view, the probability that the successful project is selected is strictly higher in a reactive organization, as the Decision Maker’s decision rule in such an organization is driven by her private information. As we noted earlier, such a gain is independent from the signalling aspect of our model. • Cost of mismatch: In reactive organizations, the Implementer must implement, with probability 1/2, a project he dislikes and for which he is more reluctant to exert high implementation eﬀort. This never happens in an optimal non-reactive organizations, as the Decision Maker always selects the Implementer’s preferred project. Here again, this cost is independent of the signaling aspect of the model. • Credibility gain: In a reactive organization, project selection perfectly conveys information on the Decision Maker’s private signal, whereas in a non-reactive organization, the Implementer can extract no additional information about the true state of nature. Thus, in a reactive organization, the Implementer’s faith in the project he’s implementing is always higher than with a non-reactive organization. In other words, reactivity gives credibility to the Decision Maker’s project selection. This eﬀect is a byproduct of the signaling game, and would disappear were the Decision Maker’s signal public information. As is obvious from equation (3.14) and (3.15 ), the reactive organization’s value is strictly increasing in α while the optimal non-reactive organization’s value is independent of α. Therefore, as the signal’s precision increase, the optimal organization switches from a non-reactive organization where Implementer’s preferences are perfectly known to a reactive organization where Implementer’s preferences becomes uncertain, i.e. where the Implementer is more independent. While this argument is intuitive, we still have to check that both reactive and non reactive equilibria exist when α varies. This is done in the following proposition: Proposition 4 The optimal organizational form goes from non-reactive to reactive as the signal’s precision increases. Semi-reactive organizations may be optimal for intermediate level of signal’s precision when a reactive equilibrium fails to exist. More precisely, deﬁne α0 such as the unique α ∈]1/2, 1[ such that: b/2), α0 F (α0¯ + F (α0 b) = F (¯ b) then:

Optimal Independence in Organizations

85

¯ • If α0 BF (α0¯ ≥ (1 − α0 )BF (αb), then the optimal organization is: b) 1. a non-reactive organization for all α ≤ α0 (with β = 1) 2. a reactive organization for α ≥ α0 (with any β ∈ [β1 , β2 ]) ¯ • If α0 BF (α0¯ < (1 − α0 )BF (αb), then there exists (α1 , α2 ) ∈]α0 , 1[, such that the optimal organization is: b) 1. a non-reactive organization for α ≤ α1 (with β = 1) 2. a semi-reactive organization for α ∈ [α1 , α2 ] (with β = 0) 3. a reactive organization for α ∈]α2 , 1) (with any β ∈ [β1 , β2 ]) Proof See Appendix B.1.4

3.4

Robustness

This section investigates the robustness of our results in front of three speciﬁcation changes that we believe shed light on our mechanism. First (section 3.4.1), we allow the Decision Maker to design the organization (i.e. to choose herself for the level of congruence with the Implementer). Second (section 3.4.2), we investigate whether diversity remains useful once we allow the Owner to provide powerful monetary incentives to the Decision Maker

3.4.1

Delegation of Organizational Design to the Decision Maker

In this section, we want to investigate the implication of allowing the Decision Maker to hire the Implementer. This amounts to delegating organizational design (i.e. the choice of congruence β) to the Decision Maker. In such a case, would the Decision Maker choose an optimal organization from the Owner’s perspective ? To some extent, the normative implications of our model hinge on the answer to this question, as we want to characterize the situations where the Decision Maker does not choose the eﬃcient congruence level β (eﬃcient from the Owner’s perspective) so that it becomes eﬃcient to limit her control over hiring decisions. In the corporate governance application we will explore later on, it would suggest instances where shareholders, instead of the CEO, should hire the company’s top executives. When in charge of choosing the congruence level β, the Decision Maker is facing the following trade-oﬀ: on the one hand, she shares part of the Owner’s objective, i.e. she wants to maximize the Implementer’s eﬀort for all type of selected project. In particular, she wants to increase the credibility of her order vis ` vis the Implementer (i.e. a the “credibility gain” above) and avoid to have an Implementer implementing a disliked project(i.e. the “cost of mismatch”above). On the other hand, conditional on success, she has intrinsic preferences toward project 1, whereas the Owner is indiﬀerent between the two project. Therefore, even when the signal indicates project 2, given that

Optimal Independence in Organizations

86

this signal is imperfect (α < 1), she is more willing to experiment project 1 than the Owner would be. Thus, for a given level of signal precision, the Decision Maker is marginally more willing to design a non-reactive organization (i.e. β = 1 when such a congruence supports a non-reactive equilibrium) than is the Owner. Note though that if the Owner is indiﬀerent between the two potential non-reactive organizations (i.e. one where only project 1 is implemented and one where only project 2 is implemented), this is not the case of the Decision Maker who, because of her intrinsic preferences, strictly prefers the pooling equilibrium where she always selects project 1. The above discussion suggests that the Decision Maker and the Owner are likely to choose similar organizations when the signal is either very precise or very imprecise. Yet, their choice may diﬀer for intermediate values of α. The formal analysis conﬁrms this intuition. Consider ﬁrst the case of a reactive equilibrium (i.e. β ∈ [β1 , β2 ] when this interval is not empty). The Decision Maker’s utility is then given by:

U R (β) =

α ¯ α (1 − β).F (α¯ + β.F (αb) .B [β.F (α¯ + (1 − β).F (αb)].B + b) b) 2 2 proba. that project 1 succeeds proba. that project 2 succeeds

(3.17)

which is increasing in β. Because she gets higher utility when she orders action 1, the Decision Maker prefers Implementers who are intrinsically more motivated by project 1 than by project 2: the positive eﬀect of increased motivation on project 1 dominates the negative eﬀect of decreased motivation on project 211 . As a result, conditional on choosing to be reactive and conditional on a reactive equilibrium being feasible, the Decision Maker prefers the highest possible level of congruence, β = β2 , while the Owner would be indiﬀerent between all level of congruence β in [β1 , β2 ]. Using the deﬁnition of β2 , we can re-write the Decision Maker’s utility in a reactive organization with β = β2 :

UR =

1 ¯ β F (α¯ + (1 − β2 )F (αb) B b) 2 2

(3.18)

When she decides to set up a non-reactive organization, provided a non-reactive equilibrium is feasible, the Decision Maker anticipates that she will always select project 1. She thus naturally prefers to hire an Implementer with a certain preference for project 1 (i.e. she sets β = 1):

U NR =

1 ¯ ¯ b F ( )B 2 2

(3.19)

Therefore, when the Decision Maker has control over the organizational design, she will hire a non-congruent
11 These

two eﬀects would exactly oﬀset each other from the Owner’s perspective.

Optimal Independence in Organizations

87

Implementer (i.e. β = β2 ) when: ¯ α ¯ α (1 − β2 ).F (α¯ + β2 .F (αb) .B ≥ 1 F ( b )B ¯ [β2 .F (α¯ + (1 − β2 ).F (αb)].B + b) b) 2 2 2 2
∗ Given that, by deﬁnition, β2 is the level of congruence for which the Decision Maker is indiﬀerent between

following a signal 2 and ordering 1, this condition can be rewritten as: F (α¯ − F b) ¯ b 2
∗ ≥ (1 − β2 ) F α¯ − F ¯ b b DM’s cost of mismatch DM’s credibility gain

(3.20)

The Decision Maker thus compares the gains from credible choices (i.e. that the Implementer believes more in the selected project success) to the costs of potential mismatch (i.e. the possibility that the Implementer intrinsically dislikes the selected project). As the careful reader will notice, the reactivity gain does not appear in the Decision Maker’s trade-oﬀ, as she is ex post indiﬀerent between selecting project 1 or project 2 after having received a private signal 212 . In other words, the Decision Maker does not value reactivity per se. However, she values the credibility that reactivity brings to her decisions as it increases the overall supply of eﬀort by the Implementer. From the Decision Maker’s perspective, organizational diversity has the value of a commitment device. As the gain from credible choices increases with the signal’s precision, the Decision Maker will be more willing
∗ to hire an independent Implementer (i.e., chooses a level of congruence β = β2 ) for high levels of signal precision.

However, as reactivity, per se, is worthless to her, the signal needs to be more precise for her to hire an independent Implementer than it needs to be for the Owner. The proof we have just informally sketched does not take into account the possibility that, for some parameter values, reactive equilibrium may fail to exist and also ignores the semi-reactive equilibria described in proposition 3. The following proposition gives a more formal results on delegation of organizational design to the Decision Maker. Proposition 5 Let us assume that: ¯ b) α0 BF (α0¯ ≥ (1 − α0 )BF (αb) (3.21) In that case, there exists a unique α 1 such that: (1) for α ∈]α0 , α[, the Decision Maker prefers a non-reactive ¯ 2 α organization while the Owner prefers a reactive organization and (2) for α ∈ [1/2, α0 ] [¯ , 1), both the Decision Maker and the Owner have the same preferences over organizational design. In other words, for intermediate level of signal’s precision, eﬃciency requires organizational design to be a prerogative of the organization’s Owner. Proof See Appendix B.1.5.
12 The reason is that, within the set of reactive equilibria, the Decision Maker always prefers organizations with the most congruent Implementer possible (i.e. β = β2 ) and that the deﬁnition of this congruence level is that the Decision Maker is indiﬀerent, conditional on signal 2, between project 1 and project 2.

Optimal Independence in Organizations

88

3.4.2

Allowing for Financial Incentives

Our analysis has shown that, in an organization where the only source of motivation is intrinsic, organizational design, i.e. the choice of congruence between the Decision Maker and the Implementer, could signiﬁcantly improve the quality of decision making, by constraining the Decision Maker to use her private information in her decision rule. Is this still true once extrinsic motivation is introduced in the organization ? Instead of using independence as a way to constrain the Decision Maker’s intrinsic motivation and to relax the “implementation constraint”, one may wonder whether extrinsic motivation cannot act as a more eﬃcient device to achieve this goal. We establish in this section that this is not the case. We show that diversity is still valuable to the organization even when “very complete” contracts are used, i.e. wages that are contingent on success or failure and also on preferences of both the Implementer and the Decision Maker. The intuition for this result is simple and relies on the existence of a limited liability constraint for the Decision Maker: with such a constraint, using monetary incentives to alleviate the implementation constraint is costly, while the use of independence in preferences comes for free. A loose sketch of the proof (see Appendix B.1.6 for a rigorous presentation) goes as follows: it is always the case that for suﬃciently high precision of the signal, α, the Owner wants to put in place a reactive organization. Then, assume that the optimal organization is such that the Decision Maker gets a strictly positive monetary reward for selecting action 2 and that congruence is perfect, i.e. β = 1. This clearly cannot be optimal: the Owner would be better oﬀ by decreasing both the Decision Maker’s wage and the congruence. Indeed, decreasing the congruence alleviates the Decision Maker’s incentive constraint and it also decreases the monetary compensation needed to make her react to her private information. Moreover, it would not aﬀect the Implementer’s motivation, as costs from mismatch (i.e. the possibility that the Implementer must implement a project he dislikes) are ex ante constant in a fully reactive organization. Therefore, diversity will be a useful tool for the Owner as long as there exists a non-zero measure set of parameters for which (1) the Owner ﬁnds reactivity optimal and (2) with no monetary compensation and a perfectly congruent Implementer, the organization is not reactive. In such a case, the Owner will want to maintain a certain level of diversity within the organization, in order to minimize the monetary compensation to give to the Decision Maker. Unfortunately, the reasoning we just sketched cannot be generalized for all values of α: it may well be the case that for levels of signal’s precision α such that reactivity is optimal, reactivity can be implemented even with perfect congruence (in other words, independence, i.e. β < 1, may not be necessary to make the Decision Maker willing to react to her private information as the signal is already precise enough). As proved in proposition 4,

Optimal Independence in Organizations

89

such a case never happens when monetary compensations are not allowed, as the optimal reactive organization always features strict independence(i.e. β < 1) for some level of signal’s precision
13

. When monetary contracts are

introduced, the possibility of rewarding the Implementer for success may leave no role for diversity as the optimal Implementer’s compensation may decrease his bias (e.g. when the distribution function F is “very” concave) and therefore make the “implementation constraint” less binding, so that even a high level of congruence may support a reactive equilibrium. Fortunately, this is not a general result, and we prove in proposition 6 that at least for a particular distribution (the uniform distribution) strict diversity is still required at the optimum. Proposition 6 Consider the special case where F is uniform over some interval [0, X]. Assume the Owner can write monetary contracts with both the Decision Maker and the Implementer. Also assume that these contracts are “very complete”, i.e. can be contingent on (1) success or failure of the project (2) the Decision Maker’s and the Implementer’s preferred project. In this case, there still exists a range of precision level [α1 , α2 ], such that when α ∈ [α1 , α2 ], the Owner should ˆ ˆ ˆ ˆ hire an independent Implementer (i.e. should optimally set β ∈]0, 1[) Proof See Appendix B.1.6

3.5
3.5.1

Applications
Bottom-Up Corporate Governance

A ﬁrst application of our theory is related to corporate governance. Both practitioners and academics have framed the issue of corporate governance as a top-down problem: How can the board eﬃciently monitor the CEO of a public corporation? What charter provisions and what incentive packages can give a CEO appropriate incentives to maximize shareholder value? By contrast, our theory suggests that bottom-up corporate governance (i.e., diversity within the executive chain of command) might be an important margin to consider in this debate: on a daily basis, the CEO might be more constrained by his executives (and their willingness to implement his choices) than by board members. The mechanism underlying this bottom-up pressure might not only be the “whistle-blowing” eﬀect emphasized by the popular press.
14

After all, cases of fraud are the exception and certainly a minor phenomenon in

the aggregate compared to ineﬃcient, but legal, decisions. Our model suggests that the channel of this bottom-up pressure might be the passive resistance of subordinates to orders that they disapprove. Such a need to elicit the top executives’ support acts as a disciplining device on the CEO and prevents her from undertaking controversial actions.
13 Indeed, we showed in proposition 4 that (1) under condition 3.21, reactive organizations are optimal for α ≥ α0 and can be implemented only with β < 1 in the vicinity of α0 while (2) if condition 3.21 does not hold, then for α ∈ [α1 , α2 ], β = 0 is optimal (and implement a semi-reactive equilibrium) and in the right vicinity of α ≥ α2 , reactive organizations are optimal and need a β strictly inferior to 1 to be put in place. 14 For a rigorous formalization, see Friebel and Guriev (2005).

Optimal Independence in Organizations

90

In a companion paper (Landier, Sraer and Thesmar (2005)), we provide empirical evidence supporting strong performance eﬀects of such a bottom-up governance mechanism. From a panel of US-listed corporations, we deﬁne an index of internal governance as the fraction of top-ranking executives who joined the ﬁrm before the current CEO was appointed. Our identifying assumption is that, in most cases, top executives hired under a CEO’s tenure are more likely to be congruent with him, either because they were hired by him (this is the case where α ∈ [α, α] ¯ in the model), or due to behavioral reasons such as loyalty feelings. First of all, we ﬁnd our index to be robustly and strongly correlated with various proﬁtability measures, such as return on assets, return on equity or market-to-book value of assets. These ﬁndings are not aﬀected when we control for traditional, “external” corporate governance measures (based on board independence or takeover defenses). Secondly, it seems that CEOs with independent executives make better acquisitions: the long-run stock returns of acquiring ﬁrms with high levels of dissent are larger than those of homogeneous ﬁrms. This suggests that a crucial role for boards of directors is to design the degree of executives’ independence from the CEO, rather than simply engaging in active direct monitoring. Our theory suggests that this design of internal dissent is relevant when the extent of private information at the top of the company (α) is high. From a human resource perspective, eﬃcient boards of directors should not only focus on CEO successions issues, as is traditionally the case, but should also be involved in the hiring decisions of top executives. As shown in Section 3.4.1, when this decision is left to the CEO, there is an important risk that he will hire executives congruent with his own preferences, de facto eliminating the counter power of executive independence. An interesting example of such a phenomenon is the testimony of Jean-Marie Messier (1999), former CEO the French Utility/Telecom/Media Conglomerate Vivendi, written in his (1999) autobiography. This was about three years before the burst of the tech bubble would lead to his fall by revealing the disastrous burden on the ﬁrm’s ﬁnancial health of expensive acquisitions. Messier was well aware of the role of top executives as counter-powers: “The danger of my job is isolation: if nobody criticizes you, you end-up making errors. I want top executives that aﬃrm their convictions, not people who say “yes” to get a promotion. . . But it’s diﬃcult to keep contestation alive.” A few paragraphs later he also reveals his taste for alignment when it comes to hiring executives: “All the top executives of Vivendi except two have been hired by myself. I chose them because I feel good with them at a personal level. Before nominating them, I check that we share the same vision, the same values.” Not only should the board be involved in recruiting top executives: an implication of the theory is that it might be optimal to shield some key executives from being ﬁred by the CEO. For instance, making the CFO of a company accountable to the board of directors and not the CEO when private information at the top of the company is high

Optimal Independence in Organizations

91

would be a normative prescription of our theory.

3.5.2

Public Administration and Ideological Bias

In most democracies, a key constitutional issue is the division of power between elected party politicians and professional bureaucrats. Elected oﬃce holders choose reform goals according to their partisan orientation and their understanding of the political feasibility of reforms, while administrators carry the reforms to reality. If they do not actively cooperate, the reforms will fail.15 Our model applies quite directly to this context: when the bureaucrats share the partisan preferences of the current policy maker, they are more likely to put forth eﬀort in implementing the attempted reforms. However, a non-partisan bureaucracy might give politicians eﬃcient incentives to adjust policy reforms to the context, even when such reforms go against their political bias. This mitigating eﬀect of bureaucracy on reform choices might be welfare improving, particularly so when the informational context is relevant. When politicians have private information on the feasibility or the suitability of reforms, a non-partisan administration will make them more reactive to information going against their bias. For example, a politician prone to reform social security will be more reactive to private signals suggesting the project is quite unpopular if the social security administration is a non-partisan body. When reactivity to information by the politician is of no importance, it is preferable to have an Implementer (i.e., a bureaucracy) aligned with the Decision Maker (i.e., the political power). According to our model, if politicians can choose bureaucrats once elected, they tend to choose too often people aligned with their own ideological bias. This may, in some cases, eliminate the “bottom-up” pressure of the bureaucracy. Therefore, our model sheds light on a controversial political topic: to what extent should elected politicians be allowed to ﬁre and hire top-administrators? The eﬃciency gains from partly shielding bureaucratic careers from political power is stated by Horn (1995): “civil service rules exist to constrain the ability of elected legislators to hire and ﬁre appointed administrators.”Such a constraint is valuable to ﬁlter decisions from extreme partisan bias and to make it more suited to political demand: “The need to elicit the cooperation of their subordinates creates a strong pressure on Bureau heads to act in a non-partisan way, even when they know their term won’t outlast the current administration” (Horn (1995)). The rules establishing employment conditions in the civil service are therefore crucial. Over the course of the XIXth century, the US bureaucracy evolved to be more and more indepen15 Albeit with a diﬀerent model, Persson, Roland and Tabellini (1997) apply a similar insight to the study of the balances of powers in western democracies. In their model, dissent between sources of power increases discipline–they have no proper decision making stage. Discipline arises from a mechanism similar to our “implementation constraint.”In their model, the parliament and the executive are the two powers. Some decisions require the approval of both powers, but both powers have conﬂicting interests because they ﬁght over rents. As a result, the rent extraction of each power is kept in check by the other. This is akin to the constraint the Implementer’s eﬀort choice imposes on the Decision Maker’s order, except that it goes in both direction. But in the end, the result is a mere improvement of discipline, not better decision making.

Optimal Independence in Organizations

92

dence from elected politicians. Roughly speaking, the US has switched from a “patronage system” where elected politicians could hire, ﬁre, and promote “administrators” as their own private employees, to a “merit system” that restricts their ability to interfere with the administrative career process, including compensation (the Pendleton Act of 1883). A second feature of governmental organization is in line with our model: Using a large survey on European governments, Aberbach, Putnam and Rockman (1981) report a signiﬁcant preference polarization of politicians vs. bureaucrats. Bureaucrats exhibit a strong ideological conservatism–they “resist change”–while politicians are attracted by more radical reforms. The idea that administrative conservatism acts as a useful constraint on the political will to reform is explicitly stated by these scholars: “ Bureaucratic centrists provide ballast and stability, but they cannot provide direction and innovation.”“ The merit democracy operates like a brake; it discourages excessive swings of the pendulum” (Morstein-Marx (1959)). This feature is one of the predictions arising when the reactivity equilibrium is optimal: the choice of an implementation body with “independent” preferences is then optimal.

3.5.3

Organizing for Change

Both the recent practitioner-based literature (e.g., Intel’s ex-CEO A. Grove (1999)) and the academic management literature (e.g., Utterback (1994), Christensen (1997)) insist on the vital need for companies to organize for innovation and ﬁght inertia. In the face of increased competition and increased volatility (see e.g., Comin and Philippon (2005)), the ability to perform radical innovations and “reinvent” the company is put forth as a crucial purpose of organization design. Scholars such as J. March (1991) and C. Argyris (1990] warn against a natural tendency of organizations to produce resistance to change. In the trade-oﬀ between exploration and exploitation staged in March (2001), such resistance to change can be optimally mitigated by the regular hiring of new members coming from outside the organization. This comes at the cost of lower short-term productivity, as the new hires lack experience. But, should this “injection of fresh blood” occur at the top or at the bottom of the organization? Our model sheds some new light on this issue. Without assuming that exogenous status quo biases fatally have to arise, we show that in order to implement reactivity within organizations, it is optimal to choose a prochange Decision Maker and pro-status quo Implementers. There are two main reasons for this result. First, as change is by deﬁnition the exception and not the rule, it is valuable for the company to have implementers who “enjoy” the status quo project, which is the most likely choice of the organization (even a reactive one). Second, pro-status quo implementers discipline the bias of the Decision Maker towards change. If she were not constrained

Optimal Independence in Organizations

93

by the resistance of implementers, a pro-change CEO might be tempted to implement change even when it is not “objectively” eﬃcient. This “change for the sake of change” trap is avoided by the bottom-up pressure imposed by status quo-biased Implementers. Our model therefore suggests that (1) reactivity becomes optimal when instability increases and that (2) reactivity should be implemented through a “fresh blood at the top” policy rather than a “fresh blood at the bottom” policy. This double result ﬁts quite well with the large increase in the hiring of outside CEOs documented by Murphy and Zabojnik (2003). They ﬁnd that the fraction of CEOs hired from outside has almost doubled between the 1970s and the 1990s. This trend is parallel to the rise in volatility faced by ﬁrms. For example, Comin and Philippon (2005) establish that idiosyncratic volatility (measured as sales volatility or market leader turnover) has also doubled during that period. The management literature explicitly links the hire of outside CEOs to the need to implement change. Khurana (2002) shows (with a critical message) that the mission assigned to externally hired CEOs is often to be the “corporate saviors” reinventing the company by adapting its strategy to a new market context. Schein (1992) also emphasizes the role of leaders in implementing radical changes against the prevailing corporate culture. In his view, organizational change comes from the top against the will of the bottom layers of the hierarchy, as our model predicts. To formally derive those results, we need to augment our base model to allow for various degrees of uncertainty in the optimality of projects. In this section, we will thus go back to the model and assume away symmetry of the state of nature: state 1 occurs with probability θ > 1/2 and state 2 will now occur with probability 1 − θ ≤ 1/2 only. State 1 is the “status quo”situation: a large θ means low uncertainty and a higher probability of success from the status quo decision. To ease the exposition, we will here assume that the eﬀort cost is distributed according to a uniform distribution, i.e., F (x) = x. The various equilibria (reactive, non-reactive and semi-reactive) should now be analyzed in two diﬀerent cases: (1) when the Decision Maker is biased in favor of the status quo and (2) when the Decision Maker is biased in favor of change. We look for the optimal Decision Maker´ preferences and the congruence parameter β s with the status quo that are optimal. We show in Appendix B.2 how the results of section 2 can be generalized to this extension. The crucial insight brought by this extension is that, for a given level of congruence β that is not too low, reactive organizations are easier to design with a pro-change Decision Maker than with a pro-status quo one. The intuition is obvious: when the Decision Maker is biased toward change, he is less likely to listen to pro-status quo- inclined Implementers, and more inclined to follow his own signal. As a result, a non-reactive equilibrium where the order is always the status quo is less likely to be sustainable.

Optimal Independence in Organizations

94

We now turn to the organizational design problem. Whatever the Decision Maker’s bias, the value of a reactive organization is given by: V R (β, θ) = θα θα βb + (1 − β)b θα + (1 − θ)(1 − α) (1 − θ)α +(1 − θ)α βb + (1 − β)b θ(1 − α) + (1 − θ)α

which is an increasing function of β. The principal source of this monotonicity is the following: as the status quo is more likely to be the optimal decision (θ > 1/2), it is more eﬃcient to hire an Implementer who is intrinsically motivated by implementing the status quo. This eﬀect was not present in our benchmark model where both states were equally likely to be optimal. A second source of the monotonicity is that the eﬀort gain from alignment between the order and the Implementer’s preferences is higher for the status quo project16 , simply because the ex post prior for this state is higher. A third eﬀect is absent here because we assumed F is linear. It is a “saturation eﬀect” that counteracts the ﬁrst two eﬀects. It is dominated as long as F is not too concave or if θ is large enough. This eﬀect is dominated by the ﬁrst two as long as F is not too concave or if θ is large enough. Reactivity can be sustained for higher levels of β when the Decision Maker is pro-change, as her incentive constraint is slacker than for a pro-status quo Decision Maker (the formal proof is in appendix). As the value V R of a reactive organization is an increasing function of β, the optimal reactive organization therefore sets the highest
∗∗ sustainable β, β = β2 (θ), with a pro-change Decision Maker. 17

The value of non-reactive organizations depends on the Decision Maker’s bias:   θ(β(θb) + (1 − β)(θb)) if DM favors “status quo” V N R (β, θ) =  (1 − θ)(β((1 − θ)b) + (1 − β)((1 − θ)b)) if DM favors “change” where it is obvious that a pro-status quo Decision Maker with perfectly aligned Implementer (β = 1) strictly dominates. The intuition is, again, that when the status quo becomes more likely to be the optimal decision, it becomes more eﬃcient to hire Implementers favoring this status quo project. The stage is set to compare the values of reactive and non-reactive organizations in the following proposition: Proposition 7 1. The optimal non-reactive organization has (1) a “status quo” -biased Decision Maker, and (2) fully “status quo” -biased Implementers (β = 1). 2. The optimal reactive organization has (1) a pro-“change” Decision Maker, and (2) moderately “status quo” ∗∗ -biased Implementers (β = β2 (θ)).
b−b 17 F concave means that the returns to private beneﬁts of already enthusiastic Implementers is lower than for non-enthusiastic ones. As a result, increasing congruence has a large negative impact on “unpopular” decisions (unenthusiastic Implementers cut down eﬀort a lot), and a small positive impact on popular decisions (enthusiastic Implementers already give all they can). As a result, congruence may destroy value.
16 θα θα+(1−θ)(1−α) (1−θ)α θ(1−α)+(1−θ)α

b − b vs.

The Corporate Wealth Eﬀect

95

∗∗ 3. The net advantage of reactive organizations, V R (β2 (θ), θ) − V N R (1, θ), increases as θ decreases to 1/2. Nonreactive organizations are always optimal in the neighborhood of θ = 1. Reactive organizations are optimal in

the neighborhood of θ = 1/2 as soon as α > Proof See Appendix B.2.

b/ 2.(b + b)

Hence, our model predicts that ﬁrms operating in high-uncertainty environments (θ close to 1/2) tend to have (1) a pro-change Decision Maker and (2) high levels of diversity. Firms operating in relatively safe environments (θ closer to 1) have (1) pro-status quo Decision Makers and Implementers and (2) very low levels of diversity. Result 2 in the above set of results relates to some pieces of the informal management literature that argue that it is best to hire an outsider to implement change. For example, Lou Gerstner (quoted from Khurana (2002), p.65), IBM’s retired CEO, states in the following terms the rationale for bringing in an outsider to implement change: “You don’t see many examples of internal candidates getting to the top of the system and then laying waste to the existing culture”. An outsider is less likely to share exactly the same view as her subordinates. Our model suggests that this is good, as it allows the two parties to communicate more eﬀectively.

3.6

Conclusion

This paper has shown that independence may enhance corporate decision making quality. Because Decision Makers must internalize the motivation of intrinsically motivated Implementers, independence, measured as the congruence of these intrinsic preferences with those of the Decision Maker, may act as a moderating device in the decision making process. This moderating mechanism is diﬀerent from whistle-blowing or explicit opposition, and relies explicitly on the “separation of implementation and control” that is casually observed in organizations: the mere presence of a potentially independent Implementer along the chain of command compels the Decision Maker to use more private, “objective” information in her selection process. This mechanism is robust: even when monetary contracts are allowed or when organizational design is delegated, preferences heterogeneity can always be part of an eﬃcient organization. As we view it, the mechanism highlighted in this paper is very general and has many implications to every-day organizations. In the area of corporate governance, we provide in a companion paper evidence consistent with independent executives (1) increasing overall proﬁt and (2) increasing the quality of strategic decisions such as acquisitions. This has important normative implications for the role of boards of directors. Optimal independence can also serve as an interesting framework to understand the long-standing debate on the divide of power between elected politicians and professional bureaucrats.

The Corporate Wealth Eﬀect

96

Finally, our theoretical analysis may be extended in several directions. First, we believe that our organizational setting can bring new insight on the understanding of collusion within hierarchies. Indeed, one could try to derive the consequences of allowing the Implementer to bribe the Decision Maker in order to convince him to stick to his preferred project. Second, we have explicitly assumed that Implementers cannot leave a ﬁrm with an independent Decision Maker to join an organization with a Decision Maker sharing the same intrinsic preferences. However, there are obvious reasons why an Implementer may prefer to work with a fully congruent manager. An interesting theoretical question that then arises is whether independent organizations, even though optimal, may survive to such sorting equilibrium on the labor market.

Chapter 4

The Corporate Wealth Eﬀect: From Real Estate Shocks to Corporate Investment
4.1 Introduction

In the presence of ﬁnancial frictions, the value of a ﬁrm’s collateral plays a key role in determining the amount this ﬁrm can borrow, and the projects this ﬁrm can invest in. Barro (1976), Stliglitz and Weiss (1981) and more recently Kiyotaki and Moore (1997) point out that with either moral hazard or adverse selection, collateral will enhance a ﬁrm’s ability to issue debt and to invest. Despite an important theoretical literature, there is scant evidence on the role of collateral in determining corporate investment. The existing literature instead has focused on the eﬀect of cash on investment. Yet, cash balance is not the only asset that a ﬁrm can use to ﬁnance new investment. For instance, real estate property, but also trade credit, inventories or even productive equipment typically serve as collateral to back new loans (Davydenko and Franks, 2005). This paper is a detailed empirical study of the eﬀect of shocks to the value of collateral. We ﬁnd robust evidence that ﬁrms transform capital gains on land into new investment. Firms invest more when their land value increases. Instead of selling the land, they ﬁnance this additional investment by issuing new debt. Increased land value seems to decrease the risk and asymmetric information attached to these new loans. New debt contracts are more long term, more likely to be syndicated, and are less likely to include covenants protecting the creditors. Such a relaxation of ﬁnancing constraints reduces average proﬁtability of capital for ﬁrms whose shareholders are weak and increases it for ﬁrms with strong shareholders. We believe these results are important for at least two reasons. The ﬁrst implication is positive: it suggests that large, exogenous shifts in the value of shareholders’ equity - land in this case - have sizeable eﬀects on corporate demand for equipment goods. Such a “corporate wealth eﬀect” might explain how purely ﬁnancial shocks generate persistent macroeconomic ﬂuctuations, as argued in the macroeconomics literature since Bernanke and Gertler 97

The Corporate Wealth Eﬀect

98

(1989). Our paper uncovers the micro foundations of such a macroeconomic model in a precise manner. The second implication of our analysis is normative. As positive shocks to land value alleviate ﬁnancing constraints, holding real estate on the balance sheet may provide a useful corporate hedging instrument. Following up on Holmstrom and Tirole (2000, 2001), our analysis suggests that ﬁrms should beneﬁt more from holding land if their returns are less correlated with their liquidity needs. Because their eﬀects are easy to measure, we focus on shocks on the value of land holdings of US ﬁrms. The ﬁrst reason to study real estate is that land holdings at the ﬁrm level are directly available using standard datasets. Besides, real estate is an appealing type of collateral to study credit constraints because it is a commonly used source of collateral, either in developed (Davydenko and Franks (2005)) or in developing economies (World Bank Survey (2005)). The third virtue of real estate is that its value ﬂuctuates, so that the amount of collateral that a given ﬁrm can mobilize varies from one year to the next. At least part of these ﬂuctuations are likely to be exogenous to changes in investment opportunities for ﬁrms outside the ﬁnance, insurance, construction and real estate industries. These speciﬁc features of real estate property will allow us to identify the eﬀect of collateral on investment. Yet, we believe that our analysis extends easily to other forms of capital, like foreign exchange denominated securities, trade credit or inventories. First, we look at the sensitivity of investment to local land prices at the ﬁrm level. We expect in general such a correlation to be positive, simply because land prices comove with demand shocks, and ﬁrms tend to build up capacities in order to serve this demand. This is why we compare ﬁrms that have land to those that have none, and interpret the diﬀerential sensitivity as the eﬀect of ﬂuctuation of land value on investment and other ﬁnancial policy variables. This strategy rests on two sources of identiﬁcation. The ﬁrst one comes from the comparison, within a region, of the investment responses to price shocks between land owning and land leasing ﬁrms. Our second source of identiﬁcation is, within the set of land owning ﬁrms, the comparison of investment in high and low real estate inﬂation regions. This approach has been used, for instance, by Case, Quigley and Shiller (2001), in their study of home wealth eﬀects on household consumption. In spite of all these precautions, it could still be argued that the diﬀerence between land owning and land leasing ﬁrms is that the former tend to be more “local” than the latter. As a result, and any collateral consideration aside, land owning ﬁrms should respond more to land prices because they proxy for local demand ﬂuctuations. We treat such criticism seriously, and submit our results to a series of robustness checks. In particular, we identify a source of land prices variation that is orthogonal to local demand shocks. We use local diﬀerences in the supply of land to predict the response of local land prices to shocks on national interest rates. Such diﬀerences in the supply of land aﬀect local land prices, but not local demand. Using them as instrument, we ﬁnd estimates that are very similar to

The Corporate Wealth Eﬀect

99

those obtained using more straightforward estimation procedures Leaning on this empirical strategy, we ﬁrst report robust causal evidence that real estate inﬂation has a positive and signiﬁcant impact on the investment behavior. The point estimate suggests that for each additional $1 of land holdings, ﬁrms invest $.60 more. Overall, when real estate prices increase by one standard deviation, ﬁrms with signiﬁcant real estate ownership, relative to ﬁrms with no real estate assets, increase their level of capital expenditures by around 2% of the standard deviation of investment. The eﬀect is large, given that land is in general only a very small fraction of assets1 . Hence, while real estate is less liquid than cash, it appears that ﬁrms are still able to make use of such collateral to ﬁnance additional investments. Second, we investigate the channel through which the appreciation of land value is converted into additional investment. We ﬁnd that ﬁrms with signiﬁcant land holdings in states with increased real estate prices signiﬁcantly modify their capital structure. They do so by mostly increasing their long-term leverage. We then look more precisely into debt contracts. We ﬁnd that new debt contracts tend to be more “asymmetric information free”. They are more likely to be syndicated, to have long maturity, and less likely to include debt covenants, special clauses imposing lower bounds to ﬁrm performance whose goal is to protect creditors from moral hazard and adverse selection. Third, we investigate the proﬁtability of the investment made from this increased collateral value. As discussed by Blanchard, Lopez de Silanes and Shleifer (1994), investment may respond to such “windfalls” for two reasons: because of adverse selection on ﬁnancial markets (Myers and Majluf, 1984) or because of managerial moral hazard (Jensen and Meckling, 1976). In the ﬁrst case, a relief on ﬁnancial constraints should increase value while in the second, it should decrease value. We separate ﬁrms whose shareholders are strong from ﬁrms whose management is strong, using standard corporate governance indices. We ﬁnd that, for ﬁrms with strong shareholders, collateral windfall do not result in lower proﬁtability of invested capital. But consistently with Blanchard et al’s hypothesis, we ﬁnd that collateral windfalls are followed by a decline in proﬁtability when shareholders are weak. Thus, the shocks in collateral that we expose in this paper also provides us with a good way to evaluate some real eﬀects of the shareholder - manager conﬂict. While most of the existing theory relates investment to debt capacity, and debt capacity to collateral, the empirical literature has focused on the direct eﬀect of cash ﬂows on investment (Fazzari, Hubbard, and Petersen (1988)). Erickson and Whited (2000) and Hennessy and Whited (2007) have criticized this approach on the ground that, for both theoretical and empirical reasons, such coeﬃcients are diﬃcult to interpret. They advocate the use of GMM estimation. Another branch of the literature has tried to isolate cash ﬂows shocks that are orthogonal to investment opportunies (Blanchard, Lopez-de-Silanes and Shleifer (1994), Lamont (1997), Rauh (2006)). This
1 On average, land accounts for 2% of Property, Plants and Equipment. When we restrict ourselves to land owning ﬁrms, the mean goes up to 4%.

The Corporate Wealth Eﬀect

100

paper clearly belongs to that tradition. Even closer to the present paper, Almeida, Campello and Weisbach (2004) have focused on the role of cash holdings, instead of cash ﬂows. They show that credit constrained ﬁrms tend to store cash on their balance sheet to avoid forgoing valuable investment opportunities in the future. Rather than looking at cash (ﬂows or stock), we focus on exogenous ﬂuctuations in the value of collateral, in a large panel of ﬁrms. To our knowledge, the only existing papers on collateral shocks are Peek and Rosengreen (2000), Goyal and Yamada (2001) and Gan (2006). These contributions focus on corporate investment in the very speciﬁc context of the 1980s Japanese real estate bubble. Our paper is on US ﬁrms, uses a more stringent identifying strategy - triple, instead of double, diﬀerences and investigates in detail the mechanism through which collateral is converted into investment. In doing this, we touch two issues that have been unexplored by most of the empirical literature on ﬁnancing constraints: ﬁnancing policy and corporate performance. First, we investigate the eﬀect of corporate wealth shocks on capital structure. This allows us to add to the literature on ﬁnancing choices, which has so far mostly used endogenous and temporary shocks to corporate wealth (see for instance Myers and Shyam-Sunders, 1999). In response to an exogenous and permanent balance sheet shock, we ask if ﬁrms adjust their leverage. Our results complement the ﬁndings of Benmelech, Garmaise and Moskowitz (2005) that more liquid assets (or more “redeployable” assets) are ﬁnanced with loans of longer maturities and durations. They also complement a recent paper by Suﬁ (2007), who shows that syndicated debt is associated with loans that exhibit less asymmetric information. We ﬁnd debt to be more likely to be syndicated in the presence of collateral. A second issue we address is the proﬁtability of constrained investment. This allows us to test if ﬁnancing constraints originate in the agency cost of separation of ownership and control, or in asymmetric information on ﬁnancial markets. We ﬁnd that relaxing ﬁnancing constraints reduces proﬁtability if ﬁrms whose managers are strong. Apart from Blanchard, Lopez de Silanes and Shleifer (1994), few papers have investigated this issue. The rest of the paper is structured as followed. Section 4.2 presents the construction of the data as well as some summary statistics. Section 4.3 provides the main results on corporate investment and section 4.4 on capital structure. Section 4.5 explores the link between corporate governance and investment performance. Section 4.6 concludes.

4.2

Data

We use accounting data of US listed ﬁrms, merged with real estate prices measured both at the level of the state and of the Metropolitan Statistical Area (MSA), where the ﬁrm’s headquarters are located. We complement this information with governance data from various sources, as well as data on land supply constraints (again, at the

The Corporate Wealth Eﬀect

101

state and MSA levels).

4.2.1

Accounting and Governance Data

We begin with the entire sample of active COMPUSTAT ﬁrms between 1984 and 2004, with non-missing total assets (COMPUSTAT item #6). This provides us with a sample of 21,122 ﬁrms and a total of 185,300 ﬁrm-year observations. We then keep ﬁrms whose headquarters are located in the United States and exclude from the sample ﬁrms operating in the ﬁnance, insurance, real estate and construction industries, as well as ﬁrms involved in a major takeover operation. Finally, we discard observations were land ownership is not reported. 4.2.1.1 Real Estate Ownership

As mentioned in the introduction, we seek to understand how ﬁrms transform capital gains on their real estate into additional collateral and investment. We ﬁrst measure land holding using COMPUSTAT item #260, labeled “PPE - Land and improvements at cost”. This balance sheet item includes all land directly bought by the ﬁrm, at its purchase price (Kieso, Weygandt, Warﬁeld, 2006). It excludes constructions built on land, which enter the item “PPE - Buildings at cost” (#263). These variables are reported in COMPUSTAT starting in 1984, hence our choice of period. While “PPE - Buildings at Cost” is also related to real estate prices, we will here focus on land. The main reason for doing this is that facilities tend to be highly ﬁrm speciﬁc, and are thus likely to be worth little in themselves for other ﬁrms or real estate developers. Land, however, is not ﬁrm speciﬁc and its value as collateral should therefore be more sensitive to changes in real estate market conditions. While we believe it is more sensible to focus on land and exclude buildings, it must be noted that this convention does not aﬀect our empirical results at all. This is not surprising. In our data, both items are strongly correlated (R2 = 0.63). Aside from this, both items tend to be simultaneously positive, or simultaneously equal to zero. Among ﬁrms who report to own land, only 2% report no building. Almost 94% of ﬁrms who report at least some land also own buildings. First, we check manually the reliability of these variables by looking at the 10K forms ﬁled by listed corporations with the Security and Exchange Commission. We retrieve these documents from the SEC’s EDGAR website (http://www.sec.gov/edgar.shtml). We ﬁrst list all ﬁrms of our sample that are present in ﬁscal year 2004, and sort them alphabetically by ticker symbol. We take the ﬁrst 20 ﬁrms whose item #260 is strictly positive, as well as the ﬁrst 10 ﬁrms whose item #260 is equal to zero. For each of these ﬁrms, we then look in the 10K for item 2 (entitled “Properties”), which provides an often verbal - with some quantitative elements - description of the facilities that the ﬁrm owns or leases. The results of this investigation is reported in Table 4.1. The ﬁrst two columns report the accounting value of land and buildings held by ﬁrms according to COMPUSTAT. The third column equals 1 when land value (from COMPUSTAT) is strictly positive, and zero else. This is to be compared with column 8, which reports a 1 when the ﬁrms declares owned land in its 10K. All 20 ﬁrms with a positive land

The Corporate Wealth Eﬀect

102

holding report that they own at least one facility somewhere. All 10 ﬁrms with zero land holdings lease all their buildings and warehouses. Thus, COMPUSTAT information on land ownership is highly reliable.

Table 4.1: Checking the Reliability and Meaning of COMPUSTAT item #260
COMPUSTAT Information Land at Cost ($m) Cost ($m) property owned in HQ state in non HQ state in HQ state property Building at Any HQ Other owned Any ownership Any property Any owned 10-K information (Item 2: Properties)

Ticker

Company

Symbol

Name

20 Firms owning Real Estate, According to COMPUSTAT 97 462 2.786 2.082 187.624 .104 42.959 37.085 5.526 .479 2012 338.428 .326 .15 2.466 7.85 34.085 19.089 16.474 .251 100% 61% 61% 85% 2.796 1 0 1 78% 212.238 1 1 1 1 106.003 1 1 1 1 1 1 492.309 1 1 1 1 61.987 1 1 1 1 1 38.133 1 1 0 0 1 5.048 1 0 1 0 1 0 0 1 0 2519.492 1 1 1 1 1 6203 1 1 1 1 1 1 1 0 1 1 1 74.69 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100% 132.886 1 0 0 1 0 1 233.397 1 0 0 1 0 1 1.374 1 1 0 0 1 1 373.862 1 0 0 1 0 1 26.805 1 1 1 1 1 1 32.267 1 1 1 0 1 1 6177 1 1 1 1 1 1684 1 1 1 1 1 1

A

Agilent Technologies, Inc.

The Corporate Wealth Eﬀect

AA

Alcoa Inc.

AAIIQ

aaiPharma, inc.

AAON

Aaon, inc.

AAP

Advance Autopart, inc.

AATK

American Access Technologies, Inc.

ABC

Amerisource Bergen Co.

ABG

Asbury Automotive Group

ABL

Biltrite inc.

ABLE

Able Energy inc.

ABS

Albertson’s Inc.

ABT

Abbott Laboratories

ACET

Aceto corporation

ACME

Acme Communications, Inc.

ACMR

A.C. Moore Arts and Crafts, Inc.

ACO

Acmol International Corporation

ACR

American Retirement Corporation

ACS

Affiliated Computer Services Inc.

ACV

Alberto-Culver Company

ACU

ACME United Corporation

Overall

10 Firms not owning Real Estate, According to COMPUSTAT 0 0 0 0 0 0 0 0 0 0 0% 0 0 0 0 0 0 0% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0% 0 0 0 0 0 0 0 0 0 0 0% 0 0 0 0 0 0 0 0 0 0 0%

AACC

Asset Acceptance Capital Corp.

ABAX

Abaxis Inc.

ABCO

The Advisory board Company

ABGX

Abgenix inc.

ABIX

Abatix Corp.

ABMD

Abiomed, Inc.

ABTL

Autobytel, Inc.

ABXA

ABX Air, Inc.

ACAD

Acadia Pharmaceuticals Inc.

ACCL

Accelrys, Inc.

Overall

103

Notes: We restrict ourselves to ﬁrms present in our sample in 2004. We then split observations into ﬁrms with some land (item 260 > 0) and ﬁrms without any land (item 260 =0). For each group, we then sort ﬁrms by ticker symbol. We focus on the ﬁrst 20 ﬁrms with land, and the ﬁrst 10 ﬁrms without land. For each ﬁrm in this list, we look for the 10K form that correspond to activity in year 2004. Under item 2 (property), we ﬁnd the information that we report here.

The Corporate Wealth Eﬀect

104

Item #260 reports the book value of real estate owned by the ﬁrm when it was purchased (historical cost). It is not appreciated when real estate prices go up, so that capital gains do not show up in this accounting variable2 . To measure (unrealized) capital gains, we thus need to interact this variable with actual real estate price indexes that vary over time. To factor the proper price, though, we need to know where the land held is located. Of course, COMPUSTAT does not provide us with the geographic location of each speciﬁc piece of land owned by a ﬁrm. But fortunately, the data reports headquarter location (variables STATE and COUNTY), which we take as a proxy of where the ﬁrm holds its real estate. A concern could be that ﬁrms may either (1) not own any real estate in the state where their headquarters are or (2) own a lot of real estate in other states - or other countries. We ask if this is the case for the thirty ﬁrms of Table 4.1. Column 4 asks if the ﬁrms owns its executive oﬃces, while column 5 asks if the ﬁrm owns any other property in the state where these oﬃces are located. Out of the 20 land owning ﬁrms, 61% actually own their headquarters. Among the remaining ﬁrms, a third leases its headquarters, but owns other facilities in the same state (actually, the same town). Overall, 79% of these ﬁrms own some land in the state (in these cases, city) where they are headquartered. Thus, land ownership is a good indicator that the ﬁrm owns land in the state (even city) where its executive oﬃces are located. As shown in column 6, there is, however, substantial measurement error left: 85% of these land owning ﬁrms also own land out of state, suggesting that COMPUSTAT item #260 vastly overestimates the fraction of land held in the HQ’s location. This discussion suggests that (1) COMPUSTAT item # 260 provides a good proxy that the ﬁrm owns land in the state (city) where its HQ are located, but that (2) this variable in general overestimates the amount of land the ﬁrm actually owns in the area. This is why we use in most regressions a dummy variable equal to 1 if and only if item #260 is strictly positive. There is another reason why we should not trust the cross sectional dispersion in the value of land and buildings: since land is reported at historical cost, the value of this variable depends crucially on when the ﬁrms has bought the facility. Put otherwise, ﬁrms who bought early will look like ﬁrms who bought little real estate later on. One last reason to discretize item #260 is that the bulk of the cross sectional variation in land ownership comes from diﬀerence between ﬁrm owning and ﬁrms leasing real estate. Indeed, almost 46% of all observations correspond to ﬁrms owning neither land nor buildings. 4.2.1.2 Other Accounting Data

Aside from real estate, we also use other accounting variables, and construct ratios as is done in most of the corporate ﬁnance literature. Let us start with the variables used in the investment equations. We compute investment rate as the ratio of capital expenditures (COMPUSTAT item #128) to past year’s Property Plant & Equipment (item
2 But,

under US GAAP, it is depreciated (impaired) when its market value falls below the purchase price.

The Corporate Wealth Eﬀect

105

#8). We compute market to book value of assets as follows: we take the total market value of equity as the number of common stocks (item #25) times end-of-year close price of common shares (item #24). To this, we add the book value of debt and quasi equity, computed as book value of assets (item #6) minus common equity (item #60) minus deferred taxes (item #74). We then normalize the resulting ﬁrm’s “market” value using book value of assets (item #6). We also use the ratio of cash ﬂows (item #18 plus item #14) to past year’s PPE (item #8). We use COMPUSTAT to measure debt issuance. We measure long term debt issue as long term debt issuance (item #111) normalized by PPE (item #8). We also compute long term debt repayment (item #114) divided by PPE. Finally, COMPUSTAT does not give us access to short term debt issuance and repurchase, so we content ourselves with net change in current debt (item #301) divided by PPE. Net change in long term debt is deﬁned as long term issuance minus long term repayments normalized by PPE. In our last regressions, we will measure accounting performance using Return on Assets: operating income before depreciation (item #13) minus depreciation (item #14) divided by total assets (item #6). Finally, to ensure that our results are statistically robust, we windsorize all variable deﬁned as ratios. Table 4.2 provides summary statistics, as well as the total number of observations after windsorization, for all accounting variables.

Table 4.2: Summary statistics
Std. Dev. Observations 25th percentile 75th percentile

Mean

Firm Level Data .386 .377 -.053 .328 .29 .006 1.76 .536 .06 -.001 2.482 -.086 8.718 1.975 2.173 1.305 17,946 1 .814 12,182 1 2.679 19,535 6.8 1.539 64,858 -.235 .50 10.6 3 3 3.59 60,105 .943 3.42 .315 70,415 -.206 .20 .10 65,239 0 .37 .499 72,405 0 1 6.623 54,749 7.17 12.14 .218 38,583 -.043 .048 .461 68,783 .009 .316 .562 67,875 0 .369 .313 71,488 -.091 .117 .457 56,156 .112 .439 .454 64,218 .112 .444

Investment

Investment net of RE Inv.

ROA

Debt Issuance

The Corporate Wealth Eﬀect

Debt Repayment

Changes in Current Debt

Average Interest Rate

LAND0

(pt Land0 /Kt−1 )

ABLAND0

Market/Book

Cash

Governance 1

Governance 2

Governance 3

MSA Level Data .034 2.094 .134 .161 2.039 .356 1,288 .367 1,288 0 1.8 .147 920 .0145 .997 3,404 1.4 .244 1,785 -.062 .111 2.43 .2505 1 2.25

Index Growth Rate (MSA level)

Index (MSA level)

Land Constraint (Geography)

Land Constraint (Rent Control)

Land Constraint (Regulation)

State Level Data .048 1.93 .057 .024 955 .71 1,014 .047 965 .023 1.36 .04 .062 2.3 .072

Index Growth Rate (State level)

Index (State Level)

Personal Income Growth Rate

106

Notes: This table provides summary statistics at the ﬁrm level (panel 1) at the MSA level (panel 2) and at the state level (panel 3). Investment is deﬁned as capital expenditure (item #128) normalized by lagged value of tangible assets (item #8). Investment net of RE investment is deﬁned as Investment (item #128) minus investment in land (item #260 in year t+1 - item #260) normalized by lagged tangible assets (item #8). ROA is deﬁned as operating income before depreciation minus depreciation and amortization normalized by total assets ((item #13-item #14)/item #6). Debt Issuance is deﬁned as item #111 normalized by tangible assets (item #8). Debt Repayment is deﬁned as item #114 normalized by tangible assets (item #8). Changes in current debt is deﬁned as item #301 normalized by tangible assets (item #8). Average Interest Rate is deﬁned as Interest expenses (item #15) normalized by total debt (item #9+item #34). LAND0 is a dummy indicating land holding (item #260) in 1984 or in the ﬁrst year of the ﬁrm’s appearance in COMPUSTAT. (pt LAN D0 /Kt−1 ) is a proxy for the current market value of initial land holdings (See Appendix C.1 for details on the construction of this variable). ABLAND0 is deﬁned as the residual in column (4) of table 4.3. Market/Book is deﬁned as market value of equity normalized by its book value (item #199 /(item #60 /item #25)). Cash is deﬁned as Income before extraordinary items + depreciation and amortization (item #14 + item # 18) normalized by lagged tangible assets (item # 8). Governance 1 is the Gompers et al. (2004) Index ; Governance 2 is board size (equals to 1 (resp 2 and 3) if the ﬁrm is in the bottom (resp. medium and top) third of the distribution of board size) ; Governance 3 is Bebchuck’s entrenchment index. Index Growth Rate (MSA Level – resp. State Level) is the growth rate of the MSA (resp. State) real estate price index ; Index (MSA Level – resp. State Level) are levels of real estate price index, and are normalized to 1 in 1984. Personal income is the state’s personal income, normalized to 1 in 1980.

The Corporate Wealth Eﬀect

107

Governance Data We then merge this dataset with corporate governance data. We use three diﬀerent sources. First, the IRRC corporate governance and directors dataset provides us with board based measures of governance: the fraction of independent directors, the number of directors sitting on the board and the fraction of former employees sitting on the board. These variables are often used in the corporate governance literature. They are available for the 1996-2001 period only, and mostly for large ﬁrms. Second, we use the increasingly popular Gompers, Ishii, and Metrick’s (2003) (hereafter GIM) index of corporate governance, which compiles various corporate governance provisions included in the CEO’s compensation package, in the corporate charter and the board structure. The GIM index is available for 1990, 1993, 1995, 1998 and 2001. In other years, we assume that it takes the value that it had in the most recent year when it was non missing. Third, we will also use the Bebchuk et al. (2004) Entrenchment Index. This index, available from 1990 to 2003, is based on six provisions - four constitutional provisions that prevent a majority of shareholders from having their way (staggered boards, limits to shareholder bylaw amendments, supermajority requirements for mergers, and supermajority requirements for charter amendments), and two takeover readiness provisions that boards put in place to prevent hostile takeovers (poison pills and golden parachutes). The merging process leaves us with a sample of 2,211 ﬁrms for which the GIM index is non-missing, 2,358 ﬁrms for which the Entrenchment index is non-missing and 1,281 ﬁrms for which board information is available. Although it can be debated, we will, as their authors do, consider that a large value of these index indicates strong managers and weak owners. Thus, everything else equal, we expect such ﬁrms to further away from maximizing value.

4.2.2
4.2.2.1

Real Estate Data
Real Estate Prices

Data for real estate prices come from the Oﬃce of Federal Housing Enterprise Oversight3 . The O.F.H.E.O. is an independent entity within the Department of Housing and Urban Development, whose primary mission is “ensuring the capital adequacy and ﬁnancial safety and soundness of two government-sponsored enterprises (GSEs) - the Federal National Mortgage Association (Fannie Mae) and the Federal Home Loan Mortgage Corporation (Freddie Mac)”. The O.F.H.E.O. provides a Home Price Index (HPI), which is a broad measure of the movement of singlefamily house prices. Because of the breadth of the sample, it provides more information than is available in other house price indexes. In particular, the HPI is available at the state level since 1975. It is also available for the 61 largest Metropolitan Statistical Areas, with a starting date between 1977 and 1987 depending on the MSA considered. We match the state level HPI with our accounting data using the state identiﬁer from COMPUSTAT. To match the
3 http://www.ofheo.gov/index.asp

The Corporate Wealth Eﬀect

108

MSA level HPI, we aggregate FIPS codes from COMPUSTAT into MSA identiﬁers using a correspondence table available from the OFHEO website. We use private household price data rather than commercial real estate data for two reasons. First and foremost, these are the only data freely available over such a period of time and at such a level of disaggregation. Moreover, real estate property is a relatively homogeneous good, which makes private single-family a good proxy for real estate. Second, having in mind endogeneity issues, we are concerned about a potential correlation between local real estate prices and local business conditions that may aﬀect the proﬁtability of investment. In that respect, private single-family house prices are a priori less correlated with local investment opportunities, than commercial real estate. 4.2.2.2 Measuring Land Supply

We measure geographical restrictions to land supply using data from Rose (1989). Rose computes, for the 40 most populated MSAs in the US, measures of the availability of land for urban use. He takes the sum of weighted annular areas, except water, around the city center. The weights decrease exponentially to zero, at a rate determined by population density. These measures are then normalized by the hypothetical value they would take in the absence of water. Rose’s index of land supply ranges from 1 in Atlanta and Phoenix (areas without water), to .521 in San Francisco and .561 in Chicago4 . As recently argued by Glaeser, Gyourko and Saks (2005) and Green, Malpezzi and Mayo (2005), regulation also plays a key role in restricting the construction of new homes, and therefore in limiting the expansion of land supply. First, regulation can aﬀect the return to new homes and therefore aﬀect the willingness of investors to build them, through, for instance, rent control (as in NYC, Boston and LA). Regulation issued at the state or at the city level can also directly impede the construction of new homes. At the state level, regulation usually take the form of environmental regulation (to protect the coast, to preserve wetlands), or planning. At the city level, the key restriction is zoning (land devoted to commercial real estate, to single family homes, to multiple family homes etc.), as well as the ability for a household or a real estate developer to rezone a given residential subdivision, and obtain a building permit. Another city level restriction that matters is the adequacy of infrastructure, although this part may be more endogenous to the local economy. We use measures of rent control (at the city level) and state level regulation from Malpezzi (1996). These measures are available for the 56 largest MSAs in the United States, and have been shown to be strongly correlated with measures of land supply elasticities by Green, Malpezzi and Mayo (2005). Taking state regulation, the ordering of MSAs changes somewhat, but the correlation between these indexes is very high.
4 In the regression analysis that follows, we use 1 minus the Rose measure instead of the Rose measure, so that it is increasing with land restrictions, and therefore homogenous with the other regulation measures we use.

The Corporate Wealth Eﬀect

109

In section 4.3.3, we interact these measures of land supply restriction with a measure of interest rate. We use the “contract rate on 30 year, ﬁxed rate conventional home mortgage commitments” from the Federal Reserve website, between 1977 and 2006.

4.2.3

Loan Contracts

Beyond debt issuance and its maturity, we look at debt contracts with more detail. Here, the dataset is DEALSCAN, which is described in detail by Chava and Roberts (2006) and Suﬁ (2007). Every year between 1987 and 2004, this dataset collects loans made by banks to many medium to large ﬁrms. Each observation is one tranche of a loan made to a given ﬁrm in a given year. The information on these tranches is collected directly from banks or through the specialized press. Obviously, this sample is biased toward large ﬁrms and large loans. The median loan amount if $94m; the average amount is $278m. More than 70% of the tranches correspond to syndicated debt. The coverage improves over time: there are 1,216 loans in 1987, 6,013 in 1996 and 6,445 in 2004. DEALSCAN gives a lot of detailed information on the debt contracts. We take out variables that measure the degree of information asymmetry between the lender and the borrower: the spread, maturity in months, the principal of the tranche, whether the tranche is part of a syndicated loan. We also retrieve information on debt covenants, i.e. promises made by the borrower about minimum interest coverage, minimum assets level, and about giving to the lender the proceeds in case of an asset sale. For each covenant, a dummy variable is constructed to be equal to one if the covenant is present. We then match this information with our COMPUSTAT data; this leads to a dataset of ﬁrm-tranche-year observations that we use in section 4.4.2.

4.3
4.3.1

Real Estate Prices and Investment
Empirical Strategy

In this section, we explore the consequences of variations in real estate prices on investment policy. For ﬁrm i, at date t, with headquarters located in state s, we start with the following equation:

s s IN Vit = αi + δt + β.LAN Di × Pts + controlsit +

it

,

(4.1)

where IN V is the ratio of investment to previous year capital stock, LAN Di is a measure of real estate ownership and Pts measures real estate prices in state s at time t, normalized by 1980’s prices. As controls, we use two variables usually included in the literature: the ratio of cash ﬂows to assets and one year lagged market to book value of assets s (see section above for a deﬁnition). We also include a ﬁrm ﬁxed eﬀect αi , as well as state-year dummies δt , designed

to capture state speciﬁc investment shocks, i.e. ﬂuctuations in the local economy. Shocks εit are clustered at the state × year level. This correlation structure is conservative given the explanatory variable of interest LAN Di × Pts

The Corporate Wealth Eﬀect

110

is deﬁned at the ﬁrm level (see Bertrand, Duﬂo and Mullainathan [2004]). We are particularly careful with our land ownership variable LAN Di . First, as mentioned above, it is best to use a dummy variable equal to 1 when the ﬁrm reports at least some land in its ﬁxed assets. To remove as much endogeneity problems as possible, we deﬁne LAN Di as the dummy equal to 1 when the ﬁrm has land in the ﬁrst year when it enters our panel (this date equals 1984 for all ﬁrms listed before that date, approximately half of the sample). As a result LAN Di is not time-varying, and its level is not identiﬁed separately from the ﬁrm ﬁxed eﬀect αi. Again, while sensible and conservative, this convention does not turn out to aﬀect much our results, because the land dummy is very stable over time. Among ﬁrms who report at least some land in their ﬁrst year of presence in the data, 96% of subsequent observations also do. Among initially non land owning ﬁrms, 87% still do not own any piece of land after the ﬁrst year in sample. Estimating β thus amounts to comparing the response of investment to real estate inﬂation, between land owning s and non land owning ﬁrms. This comparison allows to abstract from state speciﬁc macro shocks δt . One potential

problem with this approach, however, is that land ownership is itself an endogenous choice. Firms can choose to own or lease their assets. After all, our manual check from Table 4.1 showed that even among land owning ﬁrms, 40% were leasing their headquarters in 2004. Such a choice may induce strong endogeneity biases if real estate prices proxy for local demand shocks. Some ﬁrms may be more exposed than others to such local shocks, and it is possible that these ﬁrms tend to own more land, as for example in the case of local retailers. It could also be the case for, say, small ﬁrms. In such a case, the estimate of β would be misleading as it would also capture the eﬀect of these characteristics on the pro-cyclicality of ﬁrms’ investment. To alleviate this endogeneity problem, we ﬁrst control for observables that may aﬀect both the sensitivity to local demand and the propensity to own - instead of lease - land. For the ﬁrst observation of each ﬁrm, we regress the land ownership dummy on its economic determinants - such as size, age - and retrieve the residual of this equation. The determinants of land ownership we include are close to those used by Sharpe and Nguyen (1995) in their study of the share of the lease vs buy decision. We include two-digit industry dummies, as well as a measure of ﬁrm size (log of total assets), a measure of ﬁrm age (years since IPO), ﬁrm proﬁtability and a measure of capital intensity (tangible assets over total assets). We further include book leverage and state of headquarter location dummies. These last regressors do not appear in Sharpe and Ngyuen’s study, but may a priori aﬀect both the propensity to own land, as well as the sensitivity of investment to local demand. Table 4.3 presents the result of the regression of initial land holding on the various observables we use. A quick inspection of the R2 suggests that industry dummies have the largest explanatory power (almost 60% of the cross sectional variance): obviously, supermarkets or restaurant chains are more likely to own land than internet start-ups. Most other coeﬃcients have the expected sign: larger and older ﬁrms are more likely to own real estate. This is also the case for proﬁtable and capital intensive ﬁrms.

The Corporate Wealth Eﬀect

111

More surprisingly, leverage turns out to be positively correlated with land ownership, suggesting a possible reverse causality: land collateral may allow ﬁrms to take on more debt. We will actually shed some light on this mechanism in section 4.4. Table 4.3: Explaining Initial Real Estate Ownership
Initial RE Asset Dummy (1) Log(Assets) .087*** (.0018) Firm Age (2) .086*** (.0018) .052*** (.0083) IPO after 1984 .029 (.037) Tangible/Asset (3) .084*** (.0018) .053*** (.0083) .032 (.037) .29*** (.021) ROA (4) .079*** (.0021) .055*** (.0087) .028 (.037) .29*** (.022) .12*** (.012) Leverage .054*** (.0098)

Year Dummies

Yes

Yes

Yes

Yes

State Dummies

Yes

Yes

Yes

Yes

Industry Dummies

Yes

Yes

Yes

Yes

Observations

8,456

8,456

8,456

8,192

Adj. R2

.56

.57

.57

.58

Notes: This table explains the initial real estate asset ownership of a sample of COMPUSTAT ﬁrms. The dependent variable is a dummy indicating land holding (item #260) in 1984 or in the ﬁrst year of the ﬁrm’s appearance in COMPUSTAT. The explanatory variables are: Log(Assets) (item #6), Firm Age measured as the ﬁrst year in COMPUSTAT, a dummy indicating whether the ﬁrm became public after 1984, Tangible net of real estate assets (item #8-item #260)/Assets(item #6), ROA ((item #13-item #14)/item #6) and leverage ((item #9 + item #34)/ item #6). All regressions also control for state of location, year and industry ﬁxed eﬀect. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at 10, 5 and 1% level of signiﬁcance.

The Corporate Wealth Eﬀect

112

We label ABLAN Di the residual of the equation presented in column 4 of Table 4.3 and use, in most cases, the following speciﬁcation: s s IN Vit = αi + δt + β.ABLAN Di × Pts + controlsit + it

(4.2)

This speciﬁcation amounts to running equation (4.1), but including various observables (industry, size, age etc.) interacted with prices Pts . By virtue of Frisch-Waugh theorem, the estimate of β that we get using (4.2) is exactly the same. We thus control for observables that may induce a false correlation between investment and local real estate prices. Yet, there may be unobservables that could generate a strong correlation between land ownership and pro-cyclical behavior: for instance, more ambitious ﬁrms may follow demand more aggressively and prefer to buy land. We deal with this potential critique in section 4.3.3.

4.3.2

Main Results

Table 4.4 reports various estimates of equation (4.2). Column 1 is just the standard investment equation, estimated on our sample; it simply assumes that β = 0 and includes both Tobin’s q and cash ﬂows as explanatory variables. Both traditional determinants come out statistically very signiﬁcant, as in most studies. Yet, as widely argued in the literature, it is diﬃcult to interpret the positive correlation of cash ﬂows and investment as evidence of ﬁnancing constraints, both empirically and theoretically (see for instance Erickson and Whited (2000)). The explanatory power of variables included in this traditional model is not huge: for instance, a one standard deviation increase in cash ﬂows increases investment by 3% of ﬁxed assets, which corresponds to less than 9% of the cross sectional s.d. of investment.

Table 4.4: Real Estate Prices and Investment Behavior
Capital Expenditure (2) .12*** (.021) .046*** (.018) .019 (.018) .60*** (.18) (.013) (.014) (.015) .039*** .034** .04*** (3) (4) (5) (6) (7) - RE Inv. Cap. Exp.

(1)

The Corporate Wealth Eﬀect

LAND0 × Index

ABLAND0 × Index

ABLAND0 × Personal Income

(pt Land0 /Kt−1 )

Cash (.0036) .02*** (.0015) (.0011) (.0015) .017*** .02*** .02*** (.0014) (.0025) (.0036) (.0037)

.033***

.034***

.034***

.032***

.034***

(.0037)

Market/Book

.02***

(.0015)

Year×State Fixed Eﬀects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Firm Fixed Eﬀect

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

53,237

53,237

52,107

52,107

52,100

53,237

45,046

113

Notes: This table presents the impact of real estate prices on ﬁrm investment decisions, depending on a ﬁrm initial land holding. Columns (1), (2), (3), (4), (5) and (6) use capital expenditure (item # 128 normalized by lagged item #8) as dependent variable. Columns (7) use capital expenditure less real estate investment (item # 128 - (item #260 in year t + 1 - item #260) normalized by lagged item # 8). Column (1) estimates a basic investment equation, using cash (item #14 + item # 18 normalized by lagged item # 8) and lagged Market to Book ratio (item #199 /(item #60 /item #25)) as control variables. Column (2) controls for initial land holding and housing prices using LAN D0 as a measure of initial land holding. Column (3) uses ABLAN D0 (constructed as the residuals in column (4) of table 4.3) as the measure of initial land holding. Column (4) estimates equation 4.2 at the state level, while column (5) adds a direct measure of local activity (Personal Income deﬁne as an index with a value of 1 in 1980) to control for potential endogeneity issue. Column (6) uses a the current market value of initial land holding normalized by lagged tangible assets (i.e. (pt Land0 /Kt ) – see Appendix C.1 for details) as an explanatory variable. Finally, column (7) estimates equation 4.2 using Investment minus real estate investment as a dependent variable. All speciﬁcations use year-state as well as ﬁrm ﬁxed eﬀect and allow for correlation of residuals within a given state-year. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at 10, 5 and 1% level of signiﬁcance.

The Corporate Wealth Eﬀect

114

In column 2, we remove those controls and include the interaction term LAN Di × Pts . We test whether β > 0 and ﬁnd that land owning ﬁrms (LAN Di = 1) invest signiﬁcantly more when real estate prices (Pts ) increase. A one standard deviation increase in real estate prices (+15 points) increases capital expenditure by about 1.8% of ﬁxed assets more for ﬁrms who own real estate. This magnitude may not appear very large, as the cross sectional standard deviation of investment stands around 33% of tangible ﬁxed assets (PPE). Yet, it is important to keep in mind that land holding accounts, for ﬁrms with positive land holding, for only 4% of the capital stock. Given that land is such a small portion of all assets, the estimated economic magnitude is surprisingly large, when compared to, for instance, cash ﬂows in the traditional investment equation. Column 3 uses a similar speciﬁcation to that of column 2, but replaces the land ownership dummy LAN Di by the abnormal land ownership variable constructed above ABLAN Di . In contrast to columns 2, we now take into account the fact that land owning ﬁrms tend to be larger, older, more proﬁtable, more indebted and concentrated particular industries. We still exclude the usual determinants of investment. Again, the eﬀect remains statistically very signiﬁcant (below 1%). Including these controls, however, cuts the estimated size of the coeﬃcient, by nearly 60%. In column 4, we add cash ﬂows and Tobin’s q. The coeﬃcient of interest is unchanged, and remains statistically signiﬁcant at 1% (with a t-statistic of 3). Taking the various correlates of land into account, as well as cash ﬂows and market to book, a 1 s.d. deviation increase in real estate prices (15 points) leads to a diﬀerential increase in investment of some .4% of total PPE, between ﬁrms that stand 2 s.d. apart from each other in terms of ABLAN Di . This eﬀect is small (about 1% of the sample s.d. of investment), but again, it has to be compared to the share of land in PPE (4%). The methodology used so far allows to control for observable heterogeneity in sensitivity to local demand. Yet, some of this heterogeneity may be unobservable. One very ﬁrst way to tackle this line of criticism is to actually control for local demand shocks. We do this in Table 4.4, column 5. There, to measure local demand, we take state level Disposable Personal Income (DPI) series available from the Bureau of Economic Analysis, and estimate the following equation:

s s IN Vit = αi + δt + β.ABLAN Di × Pts + δ.ABLAN Di × DP It +

it

s where DP It is personal income in state s at date t. We normalise state level DPI to be equal to 1 in 1984. As one

can see from column 6 of Table 4.4, adding the controls for local activity to our baseline regression does not change at all the estimates of our coeﬃcient of interest. It decreases very slightly the precision of the estimation but the result is still signiﬁcant at the 1% level (t-value of 2.4). In addition, real estate owning ﬁrms do not behave diﬀerently in the wake of a demand shock: as expected, the coeﬃcient on the DPI interaction term is slightly positive, but far from being statistically signiﬁcant. This is comforting: at least part of the variability in real estate inﬂation is

The Corporate Wealth Eﬀect

115

orthogonal to the dynamics of local demand, and still aﬀects ﬁrm investment. It remains that this test may lack s power if DP It does not accurately reﬂect local demand shocks.

One other way to test for this alternative interpretation is by focusing on ﬁrms that are most likely to sell locally. To do this, for each manufacturing industry, we run regression (4.2) separately and obtain distinct β coeﬃcient. We then ask if β is larger for industries that tend to sell locally. We measure the propensity of each manufacturing industry to sell locally by using transport costs from National Accounts. In non reported results, we do not ﬁnd any correlation between industry β and industry transport cost. In other words, in the cross section of sectors, it is not the case that “local” industries are responsible for the average β we ﬁnd in Table 4.4. Yet again, this test lacks power as our measure of industry level transport costs is a very noisy proxy of whether a ﬁrm adresses local, or global markets. To gain such power, we will run similar regressions using city level home prices in section 4.3.3. The overall eﬀect of real estate prices on corporate investment that we describe in columns 2–5 of table 4.4 appears to be rather small. As we already mentioned, this is mainly due to the fact that land holdings account, on average, for a very small share of PPE (4%). In column 6, we quantitatively assess the role of real estate assets on a ﬁrm’s investment. We estimate the same equation as in column 4, replacing ABLAN Di × index by the market value of initial land holding, expressed as a fraction of total ﬁxed assets (variable pt × Landi /Kt−1 ). This “continuous” speciﬁcation allows us to interpret quantitatively our results. For each extra $1 of collateral, ﬁrms invest an $.60. We must insist on the fact that the market value of land holding is not directly observable. We have to rely on a proxy for this variable.5 The point estimate ($.60 investment for each $1 of collateral) is large and economically signiﬁcant. It is in line with previous estimate in the literature. In their study of ﬁnancial contracts in the property development industry, Benmelech, Garmaise and Moskowitz [2005] ﬁnd leverage ratios of nearly 90% for loans secured on real estate. This coeﬃcient should however be interpreted with caution. It relies on strong assumptions about the date of purchase of land holdings (see Appendix C.1), and the variable we use to derive it, COMPUSTAT item #260, is likely to be mismeasured (see Section 4.2.1.1). In any case, this regression conﬁrms that our eﬀect is not driven by the choice of a dummy rather than a “continuous” variable. Finally, one last caveat with the estimates from columns 2-6 is that investment contains land purchase. As a result, our strong coeﬃcients may simply reﬂect the fact that ﬁrms buy more land when its price goes up, a recommendation expressed by several real estate practitioners (see Pomazal, 2001). In non reported regressions, we looked at the elasticity of land holdings to real estate prices. We only found a slightly negative, and insigniﬁcant at the 41% level, relation between real estate inﬂation and the change in land ownership at cost, controlling for other investment determinants. The negative sign suggests that perhaps a fraction of the ﬁrms with positive land
5 We

defer the reader to Appendix C.1 for details on the computation of this variable.

The Corporate Wealth Eﬀect

116

holdings are realizing some capital gains and transform them into cash. In column 7, we report instead estimates of an equation where we replace capital expenditure by capital expenditure net of contemporaneous change in land ownership. The coeﬃcient on ABLAN Di is similar to the one reported in column 5. We ran further robustness checks that we do not report here. First, we replaced LAN D0 by a dummy equal to one when the sum of land and buildings is strictly positive. The idea behind this regression is that some buildings may not be ﬁrm speciﬁc and have resale value that is aﬀected to ﬂuctuations on the real estate market. None of our results were aﬀected, as ﬁrms owning land also tend to own buildings. Second, we ran “placebo” regressions in the spirit of Bertrand, Duﬂo and Mullainathan [2004]. For each observation in our sample, we used the model estimated in Table 4.3 to predict the probability of owning land. For each observation, we then make a draw from a uniform distribution U[0;1] : if the draw is below the predicted probability of owning land, the ﬁrm is labeled “placebo land owner”. The conditional distribution of these placebo land owners is the same as the distribution of actual land owners. We then run regressions (4.2) using the placebo land ownership dummy. We save the point estimate, and replicate the procedure 200 times. At the end of this process, we ﬁnd placebo estimates centered around zero and spread between -.01 and +.01, while the point estimate using the actual LAN Di dummy turned out to be .12 (see Table 4.4, col 2). Thus, it is the information contained in the LAN Di variable that generates the correlation of investment with real estate prices, not some hard wired eﬀect of our estimation procedure. Finally, as an ultimate robustness check, we conducted a similar analysis using French data, and obtained very similar results (see Chaney et al. (2006)).

4.3.3

City level results

In this section, we replace, in equation (4.2), state level home prices by home prices measured at the level of the city (Metropolitan Statistical Area) where the ﬁrm’s headquarters are located. In this new equation, we assume implicitly that ﬁrms owning land tend to own land in the MSA where their headquarters are located (which is likely, given that 60% of the ﬁrms manually checked in Table 4.1 that own some real estate actually own their executive oﬃces). This results in the amended version of equation (4.2): m m IN Vit = αi + δt + β.ABLAN Di × Ptm + controlsit + εit

(4.3)

for ﬁrm i, at date t, with headquarters located in MSA m. Controls are cash ﬂows and market to book ratio. We cluster the error terms εit at the MSA×year level. The result of this regression is reported in Table 4.5, column 1. The number of observations is lower than in Table 4.4 as home prices are not available for all MSAs in the US. The coeﬃcient β that we obtain is equal to 0.1, signiﬁcant at 1% (with a t-stat of 4.3). The estimated sensitivity of investment to land value is more than twice as large when we measure value using MSA level prices, rather than state level prices (we obtained, in column 5 of Table 4.4, a β of 0.039). Yet precision remains high: our interpretation

The Corporate Wealth Eﬀect

117

for this diﬀerence is that, since ﬁrms tend to own their headquarters, the use of MSA level prices gives rise to less measurement errors, and therefore less downward bias in our measure of β. Table 4.5: Robustness Table: Looking at MSA level price variation
Investment (1) ABLAND0 × MSA Index .1*** (.023) (2) .08*** (.031) (3) .12*** (.047) (4) .16*** (.049)

ABLAND0 × Stock Index

.005 (.0044)

ABLAND0 × GDP

-.013 (.029)

Cash

.03*** (.0049)

.03*** (.0049) .022*** (.0019)

.03*** (.0049) .022*** (.0019)

.03*** (.0049) .022*** (.0019)

Market/Book

.022*** (.0019)

Year FE× ABLAND0

No

No

No

Yes

Year × MSA FE

Yes

Yes

Yes

Yes

Firm FE

Yes

Yes

Yes

Yes

Observations

23,781

23,781

23,781

23,781

Notes: This table investigates the robustness of results in table 4.4 looking at price variation at the MSA Level. The dependent variable is investment (item #128 normalized by lagged item #8). Column 1 simply estimates equation 4.2 at the MSA Level. Column 2 (resp. Column 3) controls for aggregate shocks in activity, controlling for the interaction of ABLAND0 and the GDP (deﬁned as an index taking the value 1 in 1980) (resp. the Stock Market Capitalization (also deﬁned as an index)). Column 4 controls for ABLAND0 interacted with year dummies. All regressions control for year-MSA as well as ﬁrm ﬁxed eﬀects, and cluster observations at the MSA-year level. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at 10, 5 and 1% level of signiﬁcance.

The Corporate Wealth Eﬀect

118

One potential source of bias in such estimates is that our ABLAN Di variable measures some particular exposition to the aggregate business cycle. In a fashion similar to the control used in Table 4.4, column 6, we ﬁrst control for stock market prices (column 2) or aggregate GDP growth (column 3). None of these controls aﬀects our estimate of β. Last, we choose to be more agnostic about patterns of the business cycle, and include direct interactions between ABLAN Di and year dummies. Although we lose some precision, our coeﬃcient remains high (.16) and statistically signiﬁcant (with a t-statistic above 3). These robustness checks are reassuring. Yet, it could still be argued that ABLAN Di measures diﬀerent exposition to local demand shocks. Although we have shown in the past section that our results are robust to the introduction to an interaction of ABLAN Di with state level personal income, one clear possibility is that state level income does not measure local demand cycle precisely enough. Fortunately, using MSA level data has the advantage of giving us natural instrumental variables. We interact measures of local constraints on land supply and aggregate shifts in the interest rate to predict real estate prices. A lowering of interest rates shifts the demand for real estate upward. If the local supply of land is very elastic, the increased demand will translate mostly into more construction (more quantity) rather than higher land prices. If the supply of land is very inelastic on the other hand, the increased demand will translate mostly into higher prices rather than more construction. We expect that in MSAs where land supply is constrained, a drop in interest rate should have a larger impact on real estate prices. We start by checking this prediction. We estimate, for MSA m, at date t, the following model of real estate prices Ptm : Ptm = αm + δt + γ.SupplyConstraintm × IRt + um t (4.4)

where SupplyConstraintm measures constraints on land supply at MSA level, IRt is the aggregate interest rate at which banks reﬁnance their home loans (see description above). αm is an MSA ﬁxed eﬀect, and δt captures macroeconomic ﬂuctuations in real estate prices, from which we want to abstract. These ﬁrst stage regressions are reported in table 4.6. Each of three columns of Table 4.6 takes a diﬀerent measure of constraints on land supply (see exact descriptions above). In column 1, we use physical land supply (lake or sea) preventing city expansion in one direction. This speciﬁcation has slightly fewer observations as the physical constraints variables compiled by Rose[1989] are only available for a 20 cities in the US. In column 2, we take a dummy equal to 1 in the presence of rent control for at least part of the homes. In column 3, we take city level regulation (zoning restrictions, building permits, infrastructure). In all cases, high values of SupplyConstraintm means an MSA with very constrained land supply. As a result, we expect the eﬀect of declining interest rates on prices to be strongest in MSA with high SupplyConstraintm . In line with our expectations, γ turns out to be negative, and signiﬁcant at 1%. In the following, we compare the price

The Corporate Wealth Eﬀect

119

Table 4.6: Impacts of Geography, Land Regulation and Rent Control on Housing Prices
Price Index (1) Geography×Mortgage Rate -.46*** (.18) (2) -.27*** (.071) (3) -.28*** (.059)

Year Dummies

Yes

Yes

Yes

MSA Fixed Eﬀect

Yes

Yes

Yes

Observations

874

1,196

1,196

Adj R2

.84

.85

.85

F-STAT.

6

14

23

Notes: This table investigates how geography, rent control and land regulation aﬀects real estate prices. The dependent variable is the real estate price index, deﬁned at the MSA level. Column 1 uses the presence of a lake or the sea (variable Geography) interacted with mortgage rates adjusted for the inﬂation rate. Column 2 uses rent control while column 3 uses building regulation at the MSA level. All three variables are increasing in land scarcity. All regressions control for year as well as MSA ﬁxed eﬀects, and cluster observations at the MSA-year level. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at 10, 5 and 1% level of signiﬁcance.

The Corporate Wealth Eﬀect

120

responses of real estate prices to a 100bp interest rate decline, between “constrained” cities (75th percentile of the supply constraint distribution) and “unconstrained” cities (25th percentile). In column 1 (physical constraints), we ﬁnd that prices in “constrained” cities increase by 11 more points than in “unconstrained” cities (out of sample s.d. in price increases of 15). The coeﬃcient is signiﬁcant, but the F-test is not larger than 6, suggesting that this instrument may be weak. In column 2, the F-stat is much larger (14) suggesting that rent control provides us with a stronger instrument. A 100bp decrease in aggregate interest rates leads to a rise of real estate prices that is 27 points larger in “constrained” cities than in “unconstrained” cities. In column 3, we also have a strong instrument (F-stat of 23) leading to a price response of 12 points. These eﬀects are large economically, and highly signiﬁcant.6 We then move to the second stage. Equation (4.4) allows us to predict prices Ptm at the MSA level using diﬀerential impact of interest rates between MSAs. We then use the predictors of price levels from this equation as inputs for our investment equation: m m IN Vit = αi + δt + β.ABLAN Di × Ptm + γ.ABLAN Di × δt + controlsit + εm it

(4.5)

which is identical to equation (4.3). The εm are clustered at the MSA-year level. The controls ABLAN Di × δt it are there to ensure that the identiﬁcation of β rests on the diﬀerential impact of interest rates according to land supply, not on the aggregate impact of interest rate. We report the results using the three measures of land supply in Table 4.7, columns 1-3. All these estimates have to be compared with column 4, in Table 4.5, where actual MSA level prices are used instead of predicted prices, but where controls are identical (estimate of β equal to .16). The number of observations declines somewhat as land supply measures are not available for all MSAs. In column 1, we predict price levels using physical constraints on land supply. We ﬁnd a coeﬃcient of .36, signiﬁcant at 1%, which is larger than the non instrumented estimate. Regulation and rent control diﬀerences give lower estimates (.22 and .25 respectively), less signiﬁcant (at the 5% level of signiﬁcance). In all estimations, we ﬁnd slightly larger coeﬃcients than in non instrumented regressions, but the diﬀerence is never statistically signiﬁcant. Our interpretation is that, if anything, MSA level home prices are still noisy proxies of the land value of ﬁrms, and that straight OLS (within) estimators deliver slightly underestimated, but broadly correct, coeﬃcients. In the following, we will therefore focus on OLS (within) estimators.
6 The order of magnitude of these eﬀects is, however, not unrealistic. Assuming for instance that land delivers a perpetuity of π, its value should be equal to: π V = r Thus, the price response to an increase in r should write:

dV 1 = − .dr V r Under this - simple and unrealistic - calibration, with interest rates of 5%, a 100bp decrease would generate an increase in prices of 20%.

The Corporate Wealth Eﬀect

121

Table 4.7: Real Estate Prices and Investment Behavior: IV estimates
Investment Geography ABLAND0 ×M SA Index .36*** (.14) Regulation .22** (.11) Rent Control .25** (.13)

Cash

.032*** (.0062)

.038*** (.0053) .021*** (.0025)

.038*** (.0053) .022*** (.0025)

Market/Book

.018*** (.002)

Year FE× ABLAND0

Yes

Yes

Year× MSA FE

Yes

Yes

Yes

Firm FE

Yes

Yes

Yes

Observations

13,433

17,282

17,228

Notes: This table presents IV estimation of equation 4.2 using restriction on land supply interacted with mortgage interest rates as instruments for the level of real estate prices (see table 4.6 for details). The instruments used are: geography (column 1) ; rent control (column 2) ; Building regulation (column 3). All regressions control for year-MSA as well as ﬁrm ﬁxed eﬀects, and cluster observations at the MSA-year level. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at 10, 5 and 1% level of signiﬁcance.

The Corporate Wealth Eﬀect

122

4.4

Collateral and Debt

In this section, we try to explore the channel through which ﬁrms are able to convert capital gains on their land holdings into further investment. In order to do so, we ﬁrst use COMPUSTAT and look at the response of debt issuance to real estate price shocks. We will then use DEALSCAN data to see what features of debt contracts are aﬀected by capital gains on real estate.

4.4.1

Debt Issuance

As we saw in section 4.3.2, ﬁrms, when confronted with an increase in the value of their land holdings, do not sell their real estate properties. It means that outside ﬁnancing must be increased to explain the observed increase in investment. One clear candidate at this stage is the issue of new debt, secured on the incremental value of land holdings. Table 4.8 reports results of the eﬀect of an increase in land value on debt issues. To simplify interpretations, we will remove controls from equation (4.2), and replace investment on the right hand side by debt issuance: s DebtIssues = αi + δt + β.ABLAN Di × Pts + εs it it

(4.6)

To obtain estimates comparable to investment results, our debt issuance variables are normalized by tangible ﬁxed assets (PPE). We start by running the investment regression, which is identical to the speciﬁcation of Table 4.4, column 2. The only diﬀerence is that we include all observations for which cash ﬂows and Tobin’s q controls of investment are missing in COMPUSTAT. In this slightly larger sample, the investment coeﬃcient is broadly the same as in Table 4.4, column 2: β = 0.044.

Table 4.8: Real Estate Prices and Debt Issues - COMPUSTAT Data
Investment debt (1) .044*** (.0054) (0.025) (.025) (.02) (.011) 0.054** .073*** .02 -.021* (2) (3) (4) (5) issuance repayment current debt Change in Long term debt Long term debt Change in

The Corporate Wealth Eﬀect

ABLAND0 × Index

Year× State Fixed Eﬀect

Yes

Yes

Yes

Yes

Yes

Firm Fixed Eﬀect

Yes

Yes

Yes

Yes

Yes

Observations

68,203

64,917

66,369

67,259

69,477

Notes: This table presents capital structure regressions, using COMPUSTAT data. The dependent variables are: (1) Investment deﬁned as (item #128) normalized by tangible assets (item #8), (2) Change in Total Debt (item # 111 - item #114) normalized by item #8, (3) Long Term Debt Issuance, deﬁned as (item #111) normalized by item #8 (4) Long Term Debt Repayment deﬁned as (item #114) normalized by (item #8) (5) Changes in Current Debt as (item #301) normalized by item #8. The speciﬁcation is otherwise similar to that of column 2 of table 4.4. The regressions also include state-year and ﬁrm ﬁxed eﬀects. All regressions cluster observations at the state-year level. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at 10, 5 and 1% level of signiﬁcance.

123

The Corporate Wealth Eﬀect

124

To see how this additional investment is ﬁnanced, we then look at net long term debt issue (column 2), which COMPUSTAT allows us to break down into long term debt issue (column 3) and long term debt repayment (column 4). We also investigate net change in current debt (column 5), which is provided in a single item by COMPUSTAT. Overall net (of repayment) long term debt issue responds to real estate inﬂation by the same order of magnitude as investment (β = 0.054 compared to 0.044). This suggest that almost all investment is ﬁnanced by long term debt issue. When we break down net issue of debt, the bulk of the eﬀect of capital gains is on long term debt issue, whose coeﬃcient β (0.073) is larger than that of investment (0.044). Put otherwise, the data is consistent with all new investment being ﬁnanced by an additional issue of long term debt. A 15 points increase in the index is in general accompanied by a diﬀerential increase in new long term issues of .4% of total assets, between ﬁrms that stand 1 s.d. of ABLAN Di apart from each other. This is a very signiﬁcant eﬀect, albeit small when compared to average new long term issues, which amount on average to some 30% of all tangible ﬁxed assets. Yet again, land is small fraction of all assets (4%). Column 5 shows that capital gains on land are in general accompanied by a slight decrease in short term debt. Overall, the coeﬃcients on investment (0.044) and decrease in current debt (0.021) add up almost to the coeﬃcient on long term debt (0.054). This suggests that the increase in land value is used both to invest more, but also to increase overall debt maturity, as both pecking order and trade oﬀ theories of capital structure would suggest.

4.4.2

Debt Contracts

The above results are consistent with ﬁrms with more valuable land holdings taking on more long term debt. We ask here whether the new debt issued tends to be less information sensitive, as the pecking order theory of capital structure would predict. To do this, we look at various features of the debt contracts themselves, and see how they correlate with capital gains, much in the spirit of equation (4.6). Since some ﬁrms may sign several debt contracts in a given year, the dataset we use in this section is a panel of debt contracts, matched with ﬁrm characteristics (see description in section 4.2.3). For debt contract j, issued by ﬁrm i, at t, located in state s, we estimate: s s Fj,it = αi + δt + β.ABLAN Di × Pts + εs it

(4.7)

where F is the chosen feature of the contract (maturity, spread etc.). As in equation (4.6), we do not include ﬁrm level, time changing, controls, mostly because the literature provides very little guidance as to what to control for. Error terms εs are clustered at the state year level. Results of these regressions are reported in Table 4.9. it

Table 4.9: Real Estate Prices and Debt Issues - Loans Characteristics
Spread Loan (1) -15 (14) Yes Yes Yes Yes Yes Yes Yes (.037) (.19) (.084) (.041) (.15) (.085) .095*** .17 .17** -.083** -.32** -.22*** (2) (3) (4) (5) (6) (7) Principal (Log) on Interest Coverage Asset Sweep Syndicated Loan’s Maturity Covenant Cov.on Asset

The Corporate Wealth Eﬀect

ABLAND0 × Index

Year×State Fixed Eﬀect

Firm Fixed Eﬀect

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

11,102

11,893

12,181

11,656

12,736

3,183

3,183

Notes: This table investigates loan characteristics, using the DEALSCAN and SDC datasets. The dependent variables are: the loan spread (column (1)), the loan principal (column (3)), a dummy variable indicating whether the loan is syndicated (column (2)), the loan maturity in days (in logarithm, column (4)), a dummy variable indicating the presence of a covenant on interest coverage (column (5)), on asset sales (column (6)) and a dummy indicating the presence of an asset sweep (column (7)). The speciﬁcation used in this regression is similar to that of column (2) in table 4.4. Each regression uses state-year as well as ﬁrm ﬁxed eﬀects and clusters observations at the state-year level. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at 10, 5 and 1% level of signiﬁcance.

125

The Corporate Wealth Eﬀect

126

We start with interest rates (column 1). We do not ﬁnd that loans are signiﬁcantly cheaper for ﬁrms whose land has appreciated. The spread is slightly lower, but the diﬀerence is slim. The diﬀerential response of spread to a 15 points increase in real estate prices between two ﬁrms standing 1 s.d. of ABLAN Di apart from each other is less than 1 bp. The eﬀect is therefore neither statistically signiﬁcant nor economically meaningful. We interpret this as being consistent with the fact that collateral does not allow the ﬁrm to borrow at a cheaper rate, but rather allows the ﬁrm to borrow more. This is consistent with rationing on the credit market. Collateral helps the ﬁrm to take on loans (and possibly more expensive ones) that banks did not want to grant before. The rest of the evidence presented in Table 4.9 is consistent with new loans being less information sensitive. First, these new loans are more likely to be syndicated (column 2). A 15 point increase in prices raises the probability of syndication by .5%, again comparing two ﬁrms apart from each other by 1 s.d. of ABLAN Di . Suﬁ (2007) shows that syndication is evidence of low asymmetry of information between lenders and borrowers; our evidence suggests that collateral may indeed reduce such information problems. Second, these new loans are more likely to be long term (column 4), which is consistent with COMPUSTAT evidence, although there may be doubts that DEALSCAN is actually representative of all ﬁrms’ new loans (the sampling procedure does not ensure this). A 15 points increase in real estate prices increases maturity by nearly 1%. Thus, new loans seem less risky from the bank viewpoint. Third, they are less likely to include three types of covenants. Covenants are promises in debt contracts that can take many forms, and are usually interpreted by the literature as features designed to reduce information asymmetries between the lender and the borrower (see for instance Chava and Roberts (2006)). Breaching a covenant usually leads to renegotiation, although technically the bank could terminate the loan and demand immediate payback (such events are labeled “technical default”). Hence, the outcome of this renegotiation is in general in favor of the bank. Column (5) uses as the dependent variable a dummy equal to 1 if the loan speciﬁes a threshold on the interest coverage ratio (the ratio of operating income to interest expenses). If the ﬁrm breaks this threshold, it enters into technical default. Column (6) looks at the presence of a threshold on the ﬁrm’s assets, below which the terms will most certainly be renegotiated. Column(7) focuses on “asset sweep” covenants, which force the ﬁrm to pay back the loan if it sells one of its assets. For each of these features, an increase in land value is associated with a signiﬁcantly lower probability that the new debt contracts includes these covenants. As with syndication and debt maturity, the presence of additional collateral seems to alleviate information asymmetries and ease the contractual conditions under which ﬁrms can have access to new debt.

4.5

Corporate Governance and Investment Performance

In the previous sections, we have shown that additional collateral coming from increasing land value reduces information asymmetries between ﬁrms and lenders. It allows ﬁrms to borrow more long term and therefore invest

The Corporate Wealth Eﬀect

127

more. Firms are therefore credit constrained, and quickly take advantage of new collateral to invest more. Yet, the question of why such sensitivity is present in the data remains open. Blanchard, Lopez-de-Silanes and Shleifer [1994] suggest two reasons for this. One ﬁrst possibility is that managers are benevolent maximizers of shareholder value, but ﬁrms face adverse selection on the credit market. This creates rationing, so that a positive shock to collateral alleviates informational asymmetries and leads ﬁrms with valuable investment opportunities to invest more. The second possibility is that managers are simply empire-builders, not value maximizers. As a result, cash or collateral windfalls are a cheap way to help managers pursue their growth strategies, be they proﬁtable or not. Both theories are likely to contain some truth. To investigate their respective explanatory power, we look at the cross section of ﬁrms. We compare ﬁrms where shareholders are strong to ﬁrms where the management is strong. Our hypothesis is that additional investments made possible by collateral windfalls are more likely to be proﬁtable in ﬁrms with strong shareholders. To test such a prediction, we measure the impact of shocks to real estate prices on the proﬁtability of land owning ﬁrms. We compare this response for ﬁrms with strong shareholders (well governed) and ﬁrms with weak shareholders (badly governed). We expect proﬁts of well governed ﬁrms to increase more - or decrease less - than that of badly governed ones. To check this, we take three governance measures from the existing literature and estimate the following equation:

s ROAs = αi + δt + γGOVi × Pts + β.ABLAN Di × Pts + ζ.GOVi × ABLAN Di × Pts + εs , it it

(4.8)

where ROA is Return on Assets (operating income on total assets), and GOVi measures the quality of corporate governance. The last interaction term GOVi × ABLAN Di is not included because of the ﬁxed eﬀect αi . Table 4.10 shows estimates of (4.8). Column 1 ﬁrst assumes homogenous governance in the sample (GOVi = 1 for all ﬁrms). We ﬁnd here that performance does not respond to collateral shocks. This can be consistent with both empire building and asymmetric information theories working in opposite directions. One other possibility is that the asymmetric information theory holds, but some ﬁrms have decreasing returns to scale, such that even a proﬁt maximizing ﬁrm my reduce its average proﬁtability by investing more (see for instance the discussion in Banerjee and Duﬂo (2004])). Some other ﬁrms may have increasing returns to scale - such that all investment results in improved productivity. On average, both eﬀects may cancel out. Yet, for given returns to scale, it is likely that proﬁtability of badly governed ﬁrms improves less, or deteriorates more, than for well governed ones. In columns 2, 3 and 4, we interact the ABLAN Di × Pts term with three governance indices taken from the corporate governance literature, respectively the Gompers et al. (2003) index (GIM index), board size from IRRC, and the Bebchuck et al. (2004) Entrenchment Index. Note that each is an inverse measure of corporate governance, so that a high GIM index, a large board size and a large Entrenchment index all mean poor governance, i.e. weak shareholders. All results point to a statistically strong detrimental eﬀect

The Corporate Wealth Eﬀect

128

Table 4.10: Performance and Collateral Windfall - Corporate Governance
Return On Assets (×100)

GIM Index (1) ABLAND0 × Index .0078 (.013) (2) .048*** (.017)

Board Size (3) .024* (.012)

Entrenchment Index (4) .035*** (.0097)

ABLAND0 × Governance × Index

-.0049** (.002)

-.02*** (.0066) -.0035** (.0015)

-.0085** (.0042) -.004*** (.0011)

Governance× Index

-.002*** (.00045)

Year×State Dummies

Yes

Yes

Yes

Yes

Firm Fixed Eﬀect

Yes

Yes

Yes

Yes

Observations

69,851

19,238

12,044

17,657

Notes: This table relates corporate governance to investment quality, providing an estimation of equation 4.8. Dependent variable is ROA, deﬁned as ((item #13-item #14)/item #6). Corporate governance measures are: the Gompers Ishii Metrick Index (column 2); board size (column 3) ; the Bebchuk et al’s Entrenchment Index (column 4). All governance measures are constant for a given ﬁrm across time. Note that a high GIM or entrenchment index indicates poor governance. All speciﬁcation use year-state as well as ﬁrm ﬁxed eﬀect. All regressions also cluster observations at the state-year level. ∗ , ∗∗ , and ∗∗∗ means statistically diﬀerent from zero at the 10, 5 and 1% level of signiﬁcance.

The Corporate Wealth Eﬀect

129

of poor corporate governance on the quality of investment. Again, however, given that land is only a very small fraction of assets, the economic magnitude of the resulting eﬀect is small. Take, for instance, results from column 4. Well governed ﬁrms have an entrenchment index of 1. The eﬀect of collateral increase, as measured by the diﬀerence between two ﬁrms that are 1 s.d. apart in terms of ABLAN Di , is nearly equal to zero. For badly governed ﬁrms (index of 3), the eﬀect of collateral increase is to reduce ROA by 0.1 percentage point. These results suggest that while ﬁrms with sound corporate governance do not translate a positive collateral shock into higher performance, proﬁtability declines among badly governed ones. Measures of corporate governance therefore seem to entail informational content about a ﬁrm’s ability to transform ﬁnancing into value. In a recent paper, Franzoni (2007) obtains comparable results by looking at stockmarket reactions to negative cash ﬂow shocks. He ﬁnds that such responses are less negative in ﬁrms whose governance is weak. His results and ours comfort Blanchard et al (1994)’s view that ﬁrms diﬀer in their willingness to maximize shareholder value. This also suggests that the quality of governance has real eﬀects on ﬁrm proﬁtability. Probably because they have focused on the cross sectional dispersion in proﬁts, few papers have managed to exhibit real eﬀect of corporate governance - an exception being Bertrand and Mullainathan (2003). This paper set itself a much less ambitious goal, as it focuses on the dispersion in performance among well and badly governed corporations, conditional on experiencing an exogenous shock to collateral value and assuming that governance is independent from this shock. Still, we believe these results are important considering the few “real eﬀects” available in the literature.

4.6

Conclusion

The key implication of our analysis is normative. As positive shocks to land value alleviate ﬁnancing constraints, holding real estate in the balance sheet may provide a useful corporate hedging mechanism. Following up on Holmstrom and Tirole (2000, 2001), our analysis suggests that ﬁrms should beneﬁt more from holding land when its returns are less correlated with their own cash ﬂows. Thus, the decision to lease or buy land should be part of the corporate hedging policy. The present paper opens up many leads for further research. One interesting hypothesis would be to use shocks to real estate value to investigate how internal capital markets function. On a restricted sample of oil conglomerates, Lamont (1997) has shown that capital markets indeed respond to cash ﬂow shocks of one of the conglomerate’s activities. Because so many ﬁrms have land in their balance sheet, studying such land value shocks allows to replicate Lamont (1997)’s study on a larger sample. Such a new approach would allow us to study the organizational determinants of well functioning capital markets. While US data are not necessarily well suited for this kind of study - COMPUSTAT does not provide land ownership at the segment level - French ﬁrms, with their group structure, provide the ideal ﬁeld on which to test the various theories of internal capital markets that have

The Corporate Wealth Eﬀect

130

emerged in recent years.

References

131

References
Aberbach, J., Putman, R., and B. Rockman, 1981, “Bureaucrats and Politicians in Western Democracies”. Cambridge, MA: Harvard University Press. Abowd J., Kramarz, F., and D. Margolis, 1999. “High-Wage Workers and High-Wage Firms.” Econometrica, 67, 251-333. Adams, R., Almeida, H. and Ferreira, D., 2005. “Understanding the Relationship between Founder-CEOs and Firm Performance.” Working Paper, SITE-Stockholm School of Economics and Stern School of Business. Adams, R. B., H. Almeida, and D. Ferreira, 2002, “Powerful CEOs and their Impact on Corporate Performance,” Review of Financial Studies (forthcoming). Aghion, P., and J. Tirole, 1997. “Formal and Real Authority in Organizations,”Journal of Political Economy, vol. 105, no. 1, pp. 1-29. Almeida, H., Campello, M. and M. Weisbach, 2004. “The Cash Flow Sensitivity of Cash,” Journal of Finance, American Finance Association, vol. 59(4), pages 1777-1804, 08. Amit, R. and B. Villalonga, 2006. “How Do Family Ownership, Control, and Management Aﬀect Firm Value?” Journal of Financial Economics, 80, 385-417. Anderson, R. and D. Reeb, 2003. “Founding Family Ownership and Firm Performance: Evidence from the S&P 500.” Journal of Finance, 58, 1301-1327. Anderson R., Mansi, S. and D. Reeb, 2003. “Founding Family Ownership and the Agency Cost of Debt.” Journal of Financial Economics, 68, 263-285. Argyris, C., 1990, “Overcoming Organizational Defenses,” Allyn and Bacon. Baghat, S., and B. Black, 2003, “The Non-Correlation Between Board Independence and Long Term Firm Performance,” Journal of Corporation Law.

References

132

Banerjee, A. and E. Duflo, 2004. “Do Firms Want To Borrow More ? Testing Credit Constraints Using a Directed Lending Program”, mimeo MIT. Barontini, R. and L. Caprio, 2005. “The Eﬀect of Family Control on Firm Value and Performance. Evidence from Continental Europe.” ECGI - Finance Working Paper No. 88/2005. Barro, R., 1976. “The Loan Market, Collateral, and Rates of Interest,” Journal of Money, Credit and Banking, Ohio State University Press, vol. 8(4), pages 439-56, November. Bebchuk, L. , Cohen, A. and A. Ferrell, 2004. “What Matters in Corporate Governance?”, Harvard Law School John M. Olin Center Discussion Paper No. 491. Available at SSRN: http://ssrn.com/abstract=593423.

Bebchuk, L., and A. Cohen, 2004, “The Costs of Entrenched Boards,” Journal of Financial Economics (forthcoming). Benmelech, E. Garmaise and T. Moskowitz, 2005. “Do Liquidation Values Aﬀect Financial Contracts? Evidence from Commercial Loan Contracts and Zoning Regulation,” The Quarterly Journal of Economics, 120(3), pp. 1121. ´ ´ Bennedsen, M., Nielsen, K., Perez-Gonzalez, F. and D. Wolfenzon, 2006. “Inside the Family Firm: The Role of Families in Succession Decisions and Performance.” The Quarterly Journal of Economics, forthcoming. Bernanke, B. and M.Gertler, 1989. “Agency Costs, Net Worth and Business Fluctuations” American Economic Review, 79(1), pp. 14-34. Bertrand, M. and S. Mullainathan, 1999. “Is There Discretion in Wage Setting? A Test Using Takeover Legislation.” Rand Journal of Economics, 30, 535-54. Bertrand, M. and A. Schoar, 2003. “Managing with Style: The Eﬀect of Managers on Firm Policies.” Quarterly Journal of Economics, 118, 1169-208. Bertrand, M., Johnson, S., Samphantharak, K. and A. Schoar, 2005. “Mixing Family With Business: A Study of Thai Business Groups and the Families Behind Them.” Working Paper, MIT Sloan. Bertrand, M., Duflo E. and S. Mullainathan, 2004. “How Much Should We Trust Diﬀerence in Diﬀerence Estimators ?”, Quarterly Journal of Economics, vol 119, pp 249-275 Bertrand, M. and S. Mullainathan, 2003. “Enjoying the Quiet Life ? Corporate Governance and Managerial Preferences” Journal of Political Economy

References

133

Blanchard, O., Lopez-de-Silanes, F. and Shleifer, A., 1994. “What do Firms do with Cash Windfalls?,” Journal of Financial Economics, Elsevier, vol. 36(3), pages 337-360, December. Bradley, M., and Sundaram, A., 2004, “Do Acquisitions Drive Performance or Does Performance Drive Acquisitions?”, mimeo Calvo, G. and S. Wellisz, 1979. “Hierarchy, Ability, and Income Distribution”, Journal of Political Economy, 87, 991-1010. Carhart, M., 1997, ”On Persistence in Mutual Fund Performance.” Journal of Finance, 52:1, pp 57-82. Case, K., Quigley, J. and R. Shiller, 2001. “Comparing Wealth Eﬀects: The Stockmarket Versus the Housing Market,” NBER Working Paper N◦ 8606. Chaney, T., Sraer, D. and D. Thesmar, 2006. “L’Impact de la Bulle Immobili`re Parisienne sur les Come portements d’Investissement des Entreprises Fran¸aises”, Mimeo CREST-LEI c Chava, S. and M. Roberts, 2007. “Is Financial Contracting Costly ? An Empirical Analysis of Debt Covenants and Corporate Investment”, mimeo Wharton School Cho, I. K., and J. Sobel, 1990. “Strategic Stability and Uniqueness in Signaling Games,” Journal of Economic Theory, 50, 381-413. Christiansen, C., 1997, “The Innovator’s Dilemma,” Harvard Business School Press. Comin, D., and Philippon, T., 2005, “The Rise in Firm-Level Volatility: Causes and Consequences,” NBER macroannuals. Cremers, M., K. John, and V. Nair, 2005, “Takeovers, Governance and the Cross-Section of Returns,” mimeo. Cukierman, A., and Tomasi, M., 1998, “When Does It Take a Nixon to Go to China?” American Economic Review, vol. 88, N. 1, 180-197. Dahya, J., J. MC Connel, and N. Travlos, 2002, “The Cadbury Commitee, Corporate Performance and Top Management Turnover,” The Journal of Finance, 67, 461–480. Davydenko, S. and J. Franks, 2005. “Do Bankruptcy Codes Matter? A Study of Defaults in France, Germany and the UK” . EFA 2005 Moscow Meetings Paper Available at SSRN: http://ssrn.com/abstract=647861 Dessein, W., 2002, “Authority and Communication in Organizations,” Review of Economic Studies, vol. 69,

References

134

pp. 811-838. Dewatripont, M., and Tirole, J., 2005, “Modes of Communication ,” Journal of Political Economy (forthcoming). Erickson, T. and T. Whited, 2000. “Measurement Error and the Relationship Between Investment and q”, Journal of Political Economy, Vol. 108, pp. 1027 - 1057.

Faccio, M. and L. Lang, 2002. “The Ultimate Ownership of Western European Corporations.”Journal of Financial Economics, 65, 365-395. Fahlenbrach, R., 2005. “Founder-CEOs and Stock Market Performance.”Working Paper, Ohio State University Fama, E. and K. French, 1996, “Multifactor Explanations of Asset Pricing Anomalies”, Journal of Finance, 51, 55-84 Fama, E. and K. French, 1997, “Industry Costs of Equity”, Journal of Financial Economics, 43:2, 153-193 Fazzari, S., Hubbard, G. and B. Petersen, 1988. “Financing Constraints and Corporate Investment.” Brookings Papers on Economic Activity, Vol. 1, pp. 141-195. Franzoni, F., 2007. “Financing Constraints and Agency Costs: The Market’s Reaction”, mimeo HEC School of Management Friebel, G., and Guriev, S., 2005, “Earnings Manipulation and Internal Incentives,” Mimeo. Gan, J., 2006. “Collateral, Debt Capacity, and Corporate Investment: Evidence from a Natural Experiment”, Journal of Financial Economics, forthcoming. Glaeser, E., Gyourko, J. and R. Saks, 2005. “Why Have Housing Prices Gone Up ?”, Harvard Institute for Economic Research. Gompers, P., Ishii, J. and A. Metrick, 2003. “Corporate Governance And Equity Prices,” The Quarterly Journal of Economics, MIT Press, vol. 118(1), pages 107-155, February. Goyal, V. and T. Yamada, 2001. “Asset Price Shocks, Financial Constraints and Evidence: Evidence From Japan”, mimeo HKUST Green, R., Malpezzi, S. and S.Mayo, 2005. “Metropolitan-Speciﬁc Estimates of the Price Elasticity of Supply

References

135

of Housing, and Their Sources”, American Economic Association Papers and Proceedings, v95, pp 333-339 Grove, A., 1999, “Only the Paranoid Survive,” Currency. Hayes, R., P. Oyer, and S. Schaeffer, 2005, “Co-Worker Complementarity and the Stability of Top Management Teams,” mimeo. Hennessy, C. and T. Whited, 2006. “How Costly if External Financing ? Evidence From a Structural Estimation” Forthcoming Journal of Finance. Hermalin, B., 1997, “Toward An Economic Theory of Leadership: Leading by Example,”American Economics Review, vol. 88, no. 5, pp. 1188-1206. Hermalin, B., and M. Weisbach, 2003, “Boards of Directors as an Endogenously Determined Institution: A Survey of the Economic Evidence,” Economic Policy Review, 9, 7–26. Holderness, C. and D. textscSheehan, 1988. “The Role of Majority Shareholders in Publicly-Held Corporations.” Journal of Financial Economics, 20, 317-346. Holmstrom, B. and J. Tirole, 2000. “Liquidity and Risk Management,” Journal of Money, Credit and Banking, Ohio State University Press, vol. 32(3), pages 295-319, August. Holmstrom, B. and J. Tirole. “LAPM: A Liquidity-Based Asset Pricing Model,” Journal of Finance, 2001, v56, 1837-1867. Holtz-Eakin, D., Joulfaian D. and H. Rosen, 1993. “The Carnegie Conjecture: Some Empirical Evidence.” Quarterly Journal of Economics, 108, 413-436. Horn, M., 1997, “The Political Economy of Public Administration,”Cambridge University Press. Jensen, M., 1986. “Agency Costs of Free Cash Flow, Corporate Finance and Takeovers.” American Economic Review, 76, 323-339. Jensen, M. and W. Meckling, 1976. “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure” The Journal Of Financial Economics, vol. III, pages 305-60. Kaplan, S. and Minton, B., 1994, “Appointment of Outsiders in Japanese Boards: Determinants and Implications for Managers,” Journal of Financial Economics, 36, 225–258. Khurana, R., 2002, “Searching for a Corporate Savior,” Princeton University Press.

References

136

Kieso, D., Weygandt, J., and T. Warfield, 2006. “Intermediate Accounting” , Wiley & Sons. Kiyotaki, N., and J. Moore, 1997. “Credit Cycles,” Journal of Political Economy, 105(2), pages 211-248. Kohlberh, E. and J. F. Mertens, 1986. “On the Strategic Stability of Equilibria,” Econometrica, 54 (5), 1003-38. Kreps, D., 1990. “Corporate Culture and Economic Theory.” In James E. Alt and Kenneth A. Shepsle, eds., Perspectives on Positive Political Economy, New York: Cambridge University Press. La Porta, R., Lopez De Silanes, F. and A. Shleifer, 1999. “Corporate Ownership Around The World.” Journal of Finance, 54, 471-517. Lamont, O., 1997. “Cash Flow and Investment: Evidence from Internal Capital Markets,” Journal of Finance, March, Vol. 52, No. 1, pp. 57-82 Landier, A., Sraer, D., and Thesmar, D., 2005, “Bottom-up Corporate Governance,”Mimeo. Landier, A., Sraer, D. and Thesmar, D., 2007, “Optimal Independence in Organizations,” mimeo Loughran, T. and A. Vijh, 1997. “Do Long-Term Shareholders Beneﬁt from Corporate Acquisitions?”Journal of Finance, 52, 1765-1790. Lyon, R., Barber, B. and C. Tsai, 1999. “Improved Methods for Tests of Long-Run Abnormal StockReturns.”Journal of Finance, 54, 165-201. Majluf, N. and S. Myers, 1984. “Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have,” Journal of Financial Economics, Vol. 13, No. 2, pp. 187-221. Malpezzi, S., 1996. “Housing Prices, Externalities and Regulation in the U.S. Metropolitan Areas”, Journal of Housing Research, Vol. 7, No. 2, pp. 209-241. March, J. G., 1991, “Exploration and Exploitation in Organizational Learning,” Organization Science, vol. 2, no. 1, pp. 71-87. Marino, A. M., Matsusaka, J. G., and Zabojnik, J., 2006. “Authority and Disobedience,” Working paper, USC. Messier, J. M., 1999, “:J6M.COM : Faut-il avoir peur de la nouvelle ´conomie ?”Hachette Litteratures. e Mitchell, M. and E. Stafford, 2000. “Managerial Performance and Long-Term Stock-price Performance.”

References

137

Journal of Business, 73, 287-320. Modigliani, F. and M. Miller, 1958. “The Cost of Capital, Corporation Finance and the Theory of Investment.” American Economic Review, 48, 261-297. Moeller, S., Schlingemann F. and R. Stulz, 2005. “Wealth Destruction on a Massive Scale? A Study of Acquiring-Firm Returns in the Recent Merger Wave.”Journal of Finance, 60, 757-782. Mork, R., Strangeland D. and B. Yeung, 1998. “Inherited Wealth, Corporate Control and Economic Growth: the Canadian Disease?”NBER Working Paper No. W6814.

Morck, R., Shleifer, A., and R. Vishny, 1989, “Alternative Mechanisms of Corporate Control,” American Economic Review, 79: 842-852 Morstein Marx, F., 1959, “Elements of Public Administration,” Engelwood Cliﬀs, NJ: Prentice-Hall. Murphy, K., and Zabojnik, J., “Managerial Capital and the Market for CEOs,” Working paper, USC. Myers, S. and L. Shyam-Sunder, 1999. “Testing Static Tradeoﬀ Against Pecking Order Models of Capital Structure,” Journal of Financial Economics 19, pp. 219-244. Pagano, M. and P. Volpin, 2005. “Managers, Workers and Corporate Control.” Journal of Finance, 60, 841-868. Peek, J. and E. Rosengren, 2000. ”Collateral Damage: Eﬀects of the Japanese Bank Crisis on Real Activity in the United States,” American Economic Review, American Economic Association, vol. 90(1), pages 30-45, March. ´ ´ Perez-Gonzalez, F., 2006. “Inherited Control and Corporate Performance.”American Economic Review, forthcoming. Persson, T., Roland, G., and Tabellini, G., 1997, “Separation of Powers and Political Accountability,”Quarterly Journal of Economics, vol. 112, no. 4, pp. 1163-1202. Pomazal, K., 2001. ”If It Appreciates, Buy It. If It Depreciates, Lease It”, Journal of the Mineral, Metals and Material Society, 53 (8) (2001), p. 48. Prendergast, C., 1993, “A Theory of Yes Men”, The American Economic Review, Vol. 83, No. 4., pp. 757-770. Rau, R. and T. Vermaelen, 1998. “Glamour, Value, and the Post-Acquisition Performance of Acquiring

References

138

Firms.”Journal of Financial Economics, 49, 223-253. Rosenstein, S., and J. Wyatt, 1990, “Outside Directors, Board Independence and Shareholder Wealth,” Journal of Financial Economics, 26, 175–191. Rotenberg, J. , and Saloner G., 1994. “Beneﬁts of Narrow Business Strategies.” American Economic Review 84: 1330-1349. Rauh, J., 2006. “Investment and Financing Constraints: Evidence from the Funding of Corporate Pension Plans,” Journal of Finance, American Finance Association, vol. 61(1), pages 33-71, 02. Rose, L., 1989. “Topographical Constraints and Urban Land Supply Indexes”, Journal of Urban Economics, vol. 26, pages 335-347. Schein, E., 1992, “Organizational Culture and Leadership,” Jossey-Bass. Sharpe, S. and H. Ngyuen, 1995. “Capital Market Imperfection and the Incentive to Lease” Journal of Financial Economics 39, pp. 271-294. Shleifer, A., and L. Summers, 1988. “Breach of Trust in Hostile Takeovers.” In Alan J. Auerbach Ed., Corporate Takeovers: Causes and Consequences, The University of Chicago Press. Simon H. A., 1957, “Authority, ”in Arensberg, C. M., Barkin, S., Cambers, W. E., Wilensky, H. L., Worthy J. C., and Dennis, B. D. (´ds.), “Research in Industrial Human Relations. A Critical Appraisal,” New York: Harper e & Brothers Publishers, pp. 103-115. Sloan, A., 1963, “My Years with General Motors,” Garden City NY: Doubleday&Co. Spence, M., 1973. “Job Market Signaling,” Quarterly Journal of Economics, 87, 296-332. Stiglitz, J. and A. Weiss, 1981. “Credit Rationing in Markets with Imperfect Information,” American Economic Review, American Economic Association, vol. 71(3), pages 393-410, June. Sufi, A., 2007, “Information Asymmetry and Financing Arrangements: Evidence from Syndicated Loans” Journal of Finance World Bank, 2005, “World Bank Survey.” Thesmar D. and M. Thoenig, 2004. “Financial Development and the Rise in Firm-Level Uncertainty.” Working paper, CREST.

References

139

Utterback, J. M., 1994, “Mastering the Dynamics of Innovation,” Harvard Business School Press.

Vafeas, N., 1999, “Board Meeting Frequency and Firm Performance,” Journal of Financial Economics, 53, 113–142. Van Den Steen, E., 2005, “Organizational Beliefs and Managerial Vision,” Journal of Law, Economics and Organization, 21, 256–283. Van Der Heyden, L., 2004. “The Performance of Family-Owned Firms in the French Stock Market: 19932002.” Mimeo INSEAD. Weisbach, M., 1988. “Outside Directors and CEO Turnover.” Journal of Financial Economics, 20, 431-460. Yermack, D., 1996, “Higher Market Valuation of Companies with a Smaller Board of Directors,” Journal of Financial Economics, 40, 185–211. Zabojnik, J., 2002, “Centralized and Decentralized Decision Making in Organizations,” Journal of Labor Economics, vol. 20, no. 1, pp. 1-22. Zingales, L., 2004, “Want to Stop Corporate Fraud? Pay Oﬀ Those Whistle-Blowers,” Washington Post (outlook section). White, Halbert (1980). “Using Least Squares to Approximate Unknown Regression Functions.” International Economic Review, 21, 149-170.

References

1

Appendix A

Performance and Behavior of Family Firms
A.1 Data

This study uses 5 distinct sources of data.

A.1.1

Corporate Accounts

The initial sample is drawn from the DAFSA yearbook of all ﬁrms listed on the French stockmarkets.1 over the 1994-2000 period2 There are on average some 700 such ﬁrms each year. This yearbook mostly collects data from annual reports and therefore provides us with the ﬁrm’s consolidated accounts (balance sheet and proﬁt accounts) as well as more ”organizational information” such as: total employment, major shareholders, all board members and part of the top management - including the CEO. Firms’ age and industry are also provided, although the industry classiﬁcation only has 13 diﬀerent names.3 Overall, there were 682 ﬁrms listed on the French stockmarket in 1999, and 549 for which we have value added ﬁgures - therefore excluding mostly ﬁnancial ﬁrms. Also the number of these ﬁrms is modest when compared to the overall population (some 2 millions of ﬁrms exist in France, among which some 700,000 corporations), these ﬁrms tend to be heads of groups, and thus to control a large number of other ﬁrms. Most of the time, these subsidiaries are legally diﬀerent ﬁrms, but eﬀectively mere “divisions” of the group. Less frequently, these ﬁrms really are separate entities that are controlled by the listed holding, but with other shareholders. This is why it is critical here to use consolidated accounts; without them, our information on employment, assets, sales and costs would be almost meaningless (a holding company has no sales and just checks in dividends, for example). Given that listed ﬁrms
1 Until 1997, France had no less than 7 stock exchanges (Bordeaux, Marseilles, Nancy, Nantes, Lille, Lyons and Paris), where most ﬁrms (70%) were listed in Paris. All exchanges were merged in 1997. 2 The DAFSA yearbooks in fact collected ﬁrm level information since at least the mid-1960s, but they have been computerized only since 1994. 3 Another, ﬁner and more standard, classiﬁcation was also provided, but it turns out that under this classiﬁcation more than a third of all ﬁrms simply appears as ”holdings”, with no further information on the group’s activity. This is why we chose to focus on the data-speciﬁc, unconventional, industry classiﬁcation.

References

2

tend to be large and group leaders, it turns out that they represent a large share of aggregate activity. Total sales generated by these ﬁrms represent some 900 bn euros, or 66% of French GDP. For those 549 ﬁrms for which we have value added ﬁgures in 1999, the sample we have represents 14% of total GDP. Total employment amounts to some 6 million jobs (a third of private sector employment), although many of them abroad - in particular in very large groups. Last, total market capitalisation of ﬁrms listed on the French stock exchange amounts to 119% of GDP.

A.1.2

Family Ownership and Management

Taking all ﬁrms listed on the stock exchange in 1999, we begin by determining whether these ﬁrms are “family ﬁrms” or not. To do so, we look at the ﬁrm’s shareholders. When we ﬁnd that the founder or the founder’s family was a blockholder of the company, we label the ﬁrm as “family ﬁrm”. We also add as an additional requirement that the blockholder owns more than 20% of the voting rights, taking into account the pyramidal structure of certain family groups. This requirement is almost useless as in only very few cases did a family own less than 20% of the voting rights. A few additional remarks are needed at this point. First, we deal with the problem of multiple founders by considering that it is suﬃcient that one of the family is still present among the shareholders to label the ﬁrm as a family ﬁrm. Second, in few instances, we stumble upon raiders, that is individuals who started with a very tiny ﬁrm - sometimes a family ﬁrm - and became progressively major players through a series of successful market operations and acquisitions - for instance Fran¸ois Pinault, or Vincent Bollor´. We label these ﬁrms as family ﬁrms (and more c e precisely as “founder-controlled”) - even though they did not create, per se, the companies in question. In addition to the basic DAFSA ﬁles, the informations on the company’s history and the identity of the owner are collected from three main sources. First, we directly look into the annual company reports obtained mainly through the Internet. As it turns out, in many cases, the ownership structure provided in the annual reports remains somewhat opaque, especially since in many circumstances French families tend to hold control through pyramids of holding corporations (see Faccio and Lang (2002)). To get at the identity of the ultimate controlling owner, we use the information collected since 1997 by the Conseil des March´s Financiers (CMF). This administrative body is an e outlet of the Treasury supposed to monitor French ﬁnancial markets ; an act passed in 1997 made it mandatory to individuals or ﬁrms who cross various thresholds in a listed ﬁrm’s capital to declare it to the CMF.4 In turn, the CMF has to make it public, and, in order to improve the transparency of the French ﬁnancial system, the CMF publication provides us with the ultimate owners of the holdings generally responsible for the transactions. Last, we complemented these two sources of information with the use of various French business newspaper websites (L’Expansion, Le Nouvel Economiste). Following Anderson et al. (2003), we then break this category down into three sub categories. First, the ﬁrm
4 These

thresholds are 5%, 10%, 20%, 33% and 50% of all votes.

References

3

is said to be “founder-controlled” when the founder of the ﬁrm still holds the family block and is CEO. Second, the ﬁrm is said to be “heir-managed” when the founder no longer holds control over the ﬁrm - most of the time because he retired or died - but when heirs of the founder collectively control the company votes, and an heir direct or indirect - of the founder is the actual CEO of the company. Third, a ﬁrm is said to be family-owned, but professionally managed, when the family (founder or heirs) still holds the controlling block, but the CEO position is held by an outsider. Following this methodology for our starting year 1999, we are able to track the family status of 470 companies among the 549 non ﬁnancial / non real-estate ﬁrms present on the market this year. We then track back any family status changes between 1994 and 1999 by looking at CEO changes in the period. We ﬁnd 161 such CEO changes and try, for every one of them, to determine whether the nature of the family status is aﬀected: only 52 of them actually turn out to correspond to such transitions. We also track with the same method any family status changes in the year 2000. Finally, we repeat this whole operation on ﬁrms exiting the market before 1999 but present at some point on the market after 1994, so that we ﬁnally look at the family status of any ﬁrm present on the French stock market between 1994 and 2000. Out of a total sample (i.e. including all years) of 731 non ﬁnancial / non real-estate ﬁrms, we are able to track the family status for 595 ﬁrms.

A.1.3

Employment Data

Total ﬁrm employment (consolidated) is reported in the DAFSA yearbook. Computing the skill, seniority and age structure within the ﬁrm requires more detailed ﬁrm-level employment data. To do this, we use the social security ﬁles made available to the statistical oﬃce by the tax administration. For each subsidiary that belongs to the listed group, these ﬁles provides us with the wage, occupation, age and seniority of 4% of the employees - all employees born in October of an even year. We then use another survey (“Liaisons Financi`res”, described for instance in Thesmar and Thoenig (2004)) to track the group that each subsidiary belongs e to. This survey on ﬁnancial relations between ﬁrms is exhaustive on all ﬁrms that have more than 500 employees or more than 1.5 million euros of shares of other ﬁrms. Consequently, most subsidiaries of our listed groups are likely to be covered by the sampling technique. We restrict ourselves to subsidiaries that are 100% controlled, directly or indirectly by the group leader (who is in general the listed ﬁrm of the group). We ﬁrst us these employee level data to re-compute total employment and average wage in the ﬁrm. In general, computed total employment is smaller than employment reported by DAFSA,5 for two reasons. First, most of these
5 In 1998 for instance, domestic employment of French listed ﬁrms (computed using social security ﬁles) accounted for about 37% of total employment of these ﬁrms (computed using reported employment in the DAFSA Yearbook).

References

4

ﬁrms, in particular the largest ones, have foreign subsidiaries and thus foreign employees that do not enter our social security records. Secondly, some of these ﬁrms include in their consolidated statements employees of ﬁrms that they do not 100% control. Since accounting regulation is not clear on consolidating rules, there is little we statisticians can do on that front. We then use these data to add information on ﬁrms’ gender, occupational, age and seniority structure. Unfortunately, education based measures of skill are not available from this dataset ; yet, as can be made clear from the Labour Force Survey which includes both informations, the occupation variable proxies education. We computed the fraction of Managers, Supervisors, Skill workers/clerks and Unskilled workers/clerks as well as the average age, age squared and average seniority of workers. Finally, we also retrieved the fraction of women employed.

A.1.4

Stock Prices

Daily stock price data over the 1991-2002 period are provided by Euronext, the French stock exchange. For each day the stock market is open, Euronext provides in particular, for each ﬁrm listed, the price at which the last transaction of the day was realized. For each month, we take the price of the last transaction of the last day of the month, and compute monthly returns using these prices. Euronext price data take account of dividends payments, but not always of stock splits, which creates sometimes huge variations in calculated monthly stock returns. To account for this, we simply trimmed the stock returns data by deleting the top and bottom 1% of the distribution. To compute abnormal returns, we need a model of expected returns. We estimate three diﬀerent models of expected returns. The ﬁrst model simply assumes that a stock’s expected return equals the market return. Unfortunately we cannot directly compute the market return with our data, because Euronext does not provide any ﬁgure for the number of shares outstanding before 1997. To simplify, we use as proxy for the market return the monthly return on the leading French stock market index, the CAC40, which is provided since 1988. The second model of expected returns is the CAPM. We ﬁrst take our measure of the riskless rate of return from EUROSTAT, which provides us with a monthly time series on the rate of return of 10-years French Treasury bills since 1980. After de-annualization, we use this measure to compute excess returns on various stocks and the market. Then, for each ﬁrm, we regress the excess stock return on the excess market return, and take the residuals of these regressions as our second measure of abnormal return. The third model takes into account the fact, well documented in the asset pricing literature, that small ﬁrms, and value ﬁrms (with low market to book value of assets), show consistently positive abnormal return in a CAPM model. As it has become standard in this literature, we follow Fama and French (1996) and add to our model of expected returns, in addition to the market return, the excess return of small ﬁrms (SMB), and the excess return on value ﬁrms (HML). SMB is computed by sorting ﬁrms according to the past year capitalisation. We call big, the

References

5

20% largest ﬁrms, and small the 20% smallest. To determine SMB, we subtract, each month, the value weighted monthly returns of the largest ﬁrms (by previous year’s standard) from the value weighted monthly returns of the 20% largest ﬁrms (again, by previous year’s standard). To compute HML, we sort ﬁrms by past year book to market value of assets. We call “value ﬁrms” ﬁrms with the 20% highest book to market in the previous year, and glamor, ﬁrms with the lowest 20%. HML is the diﬀerence in value-weighted monthly returns between the value and the gloamor portfolio. Finally, for each ﬁrm, we regress excess returns on the time series of the market excess return, on SMB and on HML, and take the residual of this regression as our third measure of abnormal returns. In the last two cases, the models of expected returns are estimated separately for each transaction (acquisition), in all the months available since 1991 before the deal. We also require that the acquiring ﬁrm has at least 12 months of stock returns prior to the transaction.

A.1.5

Acquisitions

The data source for large acquisitions is SDC platinium, a ﬁrm that collects publicized transactions (repurchases, LBOs, M&A) undertaken by companies in various countries. For France, SDC reports since 1990 some 1,000 completed acquisitions per year. We focus on all successfully completed acquisitions where the bidder (1) belongs to our sample, (2) owns less than 50% before the transaction and more than 50% afterwards. From SDC, we then retrieve the month and year of the acquisition. For the ﬁrms in our sample, we end up with some 100 acquisitions per year between 1994 and 1998. In 1999 and 2000, we have some 150 acquisitions, which is not surprising given excellent ﬁnancial market acquisitions. The number of ﬁrms making acquisitions is somewhat lower, given that some ﬁrms undertake several acquisitions (sometimes as much as 5 in a given year): over the whole period, some 80 ﬁrms (out of 650) make at least one acquisition. For the few observations for which target size is reported (a third of the total), we ﬁnd that the average cost of the transaction stands around $ 180 millions.

References

6

A.2

Additional Tables
Table B1: Year by Year Regressions of Performance of Family Firms
ROA(×100) Founder CEO Heir CEO Professional CEO Founder CEO ROE(×100) Heir CEO Professional CEO

1994

1.3 (1.1)

2.2∗∗ (1.0) 1.1 (1.0) .8 (1.0) 1.2 (1.1) 2.0∗ (1.1) 1.8∗ (1.1) 2.7∗∗ (1.2) 1.8∗∗∗ (.3)

1.7 (1.2) 1.7 (1.1) .7 (1.1) .6 (1.2) 1.5 (1.2) 1.4 (1.2) 2.5∗∗ (1.2) 1.4
∗∗∗

12.5∗∗∗ (3.3) 8.8∗∗∗ (3.2) 5.3 (3.5) 11.4∗∗∗ (3.3) 13.8∗∗∗ (3.0) 8.9∗∗ (3.5) 8.6∗∗ (4.4) 9.9∗∗∗ (1.1)

12.3∗∗∗ (3.1) 4.4 (3.0) 9.8∗∗∗ (3.3) 11.7∗∗∗ (3.1) 9.6∗∗∗ (2.8) 5.3 (3.5) 7.4∗ (4.4) 8.6∗∗∗ (1.2)

10.0∗∗∗ (3.5) 8.5∗∗∗ (3.3) 3.9 (3.6) 8.7∗∗∗ (3.3) 9.7∗∗∗ (3.0) 11.2∗∗∗ (3.7) 7.3∗ (4.5) 8.5∗∗∗ (.9)

1995

1.7 (1.1)

1996

.4 (1.1)

1997

1.4 (1.2)

1998

2.4∗∗ (1.1)

1999

.8 (1.2)

2000

3.7∗∗∗ (1.2)

Fama-Mc Beth Coeﬃcient

1.7∗∗∗ (.4)

(.2)

Source: year by year OLS estimates. See A.1 for details on data construction and sources. Dependent variable are ROA (column 1 to 3) and ROE (column 4 to 6). Control variables are similar to those used in table 1.3. Estimates of the coeﬃcients on all three family dummies, along with their standard errors, are reported for each year (lines). *,** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

References

7

Table B2: Performance of Family Firms: Further Robustness Checks
ROA (×100) ROE (×100) Market to book Log(wage bill / empl.)

Founder CEO

2.3∗∗∗ (.9)

10∗∗∗ (2.2) 11∗∗∗ (1.9) 8.9∗∗∗ (2.2) 1.3∗∗ (.53) -3.3∗∗∗ (.92) 1.5 (2.5) .26 (3.1) -23∗∗∗ (4.5) -23∗∗∗ (6.8) -.48 (2.9)

.31∗∗∗ (.1) .14 (.093) .16 (.12) .064∗∗∗ (.023) -.18∗∗∗ (.044) -.4∗∗∗ (.14) .12 (.14) -.97∗∗∗ (.17) -1.2∗∗∗ (.24) .27∗∗ (.14)

-.025 (.048) -.12∗∗∗ (.048) -.091∗∗ (.047) .016 (.011) -.0015 (.017) .099∗ (.053) .026 (.062) -.25∗∗∗ (.065) -.083 (.14) .051 (.058)

Heir CEO

2.6∗∗∗ (.79)

Professional CEO

1.8∗∗ (.85)

Log(Assets)

-.29∗ (.16)

Log(Firm Age)

-.6∗ (.36)

Former SOE

-1 (.82)

Fraction equity of largest block Debt/Assets

.24 (1.2) -10∗∗∗ (1.4)

Stock Return Volatility

-9.7∗∗∗ (2.2)

Focus

1.1 (1.1)

Industry FE Year FE

yes yes

yes yes

yes yes

yes yes

Heir=Professional Observations Adj. R2

.33 1,841 .21

.4 1,845 .17

.89 1,762 .29

.45 1,753 .3

Source: Huber-White-Sandwich estimates, allowing for correlation of observations within a given ﬁrm. See A.1 for details on data construction and sources. Note: this table is similar to table 1.3 with focus (herﬁndhal index using 2 digits classiﬁcation of industries - equal 1 when ﬁrm operates in a single sector and goes to 0 as the ﬁrm becomes very diversiﬁed) as an additional control. *, ** and *** mean statistically diﬀerent from zero at 10%, 5% and 1% level of signiﬁcance.

References

8

A.3

Breaking down Performance
∀i, t : ROAi,t = α + β1 F1,i + β2 F2,i + β3 F3,i + γXi,t + ¡ βf = ROAf − ROA0 + β Xf − X0 , i,t Diﬀerences in ROA among the diﬀerent groups of family ﬁrms are estimated with through equation (1):

For family status f , this diﬀerence is given by the βf coeﬃcient, which is identiﬁed as:

where ROAf is the mean value of ROA for family status f . To understand where this diﬀerence comes from, we use he following equality, for each ﬁrm i: ¡ ROAi = (L/K)i (Y /L)i − wi ,

where L/K is the ratio of employment to capital, Y /L is the ratio of value added to employment and w is the mean wage paid in ﬁrm i. Using this equality, we can compute the mean of ROA for any type of family status f as: ROAf = (L/K)f (Y /L)f − wf + cov (L/K)f , (Y /L − w)f Therefore, the unconditional diﬀerence in the mean of ROA between ﬁrms with family status f and widely held ﬁrms comes as: ∆ (ROA) = ROAf − ROA0 = (L/K)f ∆ (Y /L) − ∆ (w) (A.1) + (Y /L)0 − w0 ∆ (L/K) + ∆ (cov ((L/K) , (Y /L) − w)) The diﬀerences in the right-hand side of the previous equation are unconditional diﬀerences. We are interested in the contribution of conditional diﬀerences to corporate performance. Therefore, we can estimate the following equations:6 V b ∀i, t : (L/K)i,t = a + b1 F1,i + b2 F2,i + b3 F3,i + cXi,t + ηi,t b ` ∀i, t : (Y /L)i,t = A + B1 F1,i + B2 F2,i + B3 F3,i + CXi,t + ui,t b b X ∀i, t : wi,t = A + B1 F1,i + B2 F2,i + B3 F3,i + CXi,t + νi,t Using the result from these estimations, one can link the conditional diﬀerences in proﬁtability (the βf coeﬃcients) to the conditional diﬀerences in labor to capital ratio (the bf coeﬃcients), labor productivity (the Bf coeﬃcients) and wages (the Bf coeﬃcients): ∆ (ROA) = ¡ (L/K)1 B1 − B1 + (C − C)(X1 − X0 ) + ¡ (Y /L)0 − w0 b1 + c(X1 − X0 ) + ∆ (cov (L/K, Y /L − w))

So that: βf = ¡ (L/K)f Bf − B1 | {z }
Diﬀ. in Labor Prod. -Diﬀ in Wage

+

(Y /L)0 − w0 b1 {z } |
Diﬀ. in cap. lab. ratio

(A.2)

¡ + Xf − X0 β + (L/K)f (B − B) + (Y /L)0 − w0 c + ∆ (cov (L/K, Y /L − w)) {z } | {z } |
Diﬀ. in observable covariance term

We therefore see that the conditional diﬀerences in ROA come from 5 diﬀerent sources: conditional diﬀerences in labor productivity ; conditional diﬀerences in wages ; conditional diﬀerences in capital to labor ratio ; diﬀerences in observables between family status ; diﬀerent covariance structure between labor to capital ratio and labor productivity net of wages. Table 6 simply displays each of these 5 terms for the diﬀerences in performance between founder-managed, heir-managed, professionally managed and widely held ﬁrms.

6 Notice

that these three models use the real ratios and not the logarithm of these ratios.

References

9

Appendix B

Optimal Independence in Organizations
B.1
B.1.1

Proofs
Proof of Proposition 1

The conditions for the existence of the separating equilibrium announced in the proposition has been shown in the text. This equilibrium exists only when β2 ≥ 0, otherwise there is no level of congruence for which the Decision Maker is willing to select project 2 when receiving signal 2. β2 ≥ 0 is equivalent to condition 3.6. That this equilibrium is the unique separating equilibrium is easy to prove. The only other candidate for a separating equilibrium would be one where the Decision Maker selects project 2 (resp. project 1) after having received signal 1 (resp. signal 2). In such an equilibrium, the posterior beliefs of the Implementer are necessarily given by: µ1 = 1 − µ2 = 1 − α In words, the Implementer is sure that the Decision Maker has received signal 1 (resp. 2) when project 2 (resp. 1) has been selected. Consider the two incentive constraint of the Decision Maker: @ ¡ ¡ ¯ (1 − α)B βF ((1 − α)b) + (1 − β)F ((1 − α)¯ ≥ αB βF (α¯ + (1 − β)F (αb) b) b) ¡ ¡ ¯ (1 − α)B βF ((1 − α)¯ + (1 − β)F ((1 − α)b) ≥ αB βF (αb) + (1 − β)F (α¯ b) b) Adding up these two constraints leads to the following inequality: ¡ ¡ ¯ ¯ (1 − α)(B + B) F ((1 − α)¯ + F ((1 − α)b) ≥ α(B + B) F (¯ + F (b) , b) b) which is clearly never satisﬁed. Therefore, the separating equilibrium of the proposition is indeed the unique separating equilibrium of the model. That β2 ≤ 1 and β1 ≥ 0 follow directly from the deﬁnition 5 and 3. Assume that β ≤ β1 . Then: ¡ ¡ ¯ αB βF (α¯ + (1 − β)F (αb) ≤ (1 − α)B βF (αb) + (1 − β)F (α¯ b) b)

Because α > 1 − α, this directly implies that: ¡ ¡ ¯ (1 − α)B βF (α¯ + (1 − β)F (αb) < αB βF (αb) + (1 − β)F (α¯ ⇒ β < β2 , b) b) which proves that β1 < β2 . ¯ Finally note t = B/B and τ = b/¯ Then one can rewrite the threshold β2 as: b. β2 = ¢ αF (α¯ − (1 − α)F (ατ ¯ b)t b) £ F (α¯ − F (ατ ¯ [1 − α + αt] b) b)

First, we look at how β2 evolves with α. Consider α2 > α1 and let β ≤ β2 (α1 ). This implies that:
¡ ¡ ¯ (1 − α1 )B βF (α1¯ + (1 − β)F (α1 b) ≤ α1 B βF (α1 b) + (1 − β)F (α1¯ b) b) | {z } | {z }
=ψ(α) =φ(α)

Note that φ() is strictly increasing with α as F is also strictly increasing. We have: ψ (α) = = ¡ ¯ B −βF (α¯ − (1 − β)F (αb) + (1 − α)β¯ (α¯ + (1 − α)(1 − β)bf (αb) b) bf b) ¡ ¡ ¯ −B β F (α¯ − (1 − α)¯ (α¯ + (1 − β) (F (αb) − (1 − α)bf (αb)) b) bf b)

References

10

Because F is strictly concave, we know that: ∀ x ≥ 0, F (x) ≥ xf (x), so that: F (α¯ ≥ α¯ (α¯ > (1 − α)¯ (α¯ and similarly: b) bf b) bf b) F (αb) ≥ αbf (αb) > (1 − α)bf (αb). This proves that ψ() is strictly decreasing so that we have: ψ(α2 ) < ψ(α1 ) ≤ φ(α1 ) < φ(α2 ) which proves that: β < β2 (α2 ) ⇒ β2 (α1 ) < β2 (α2 ) Of course, the proof that β2 is a decreasing function of α is similar. Let us brieﬂy sketch it. Consider α2 > α1 and let β ≥ β1 (α1 ). Then:
¡ ¡ ¯ b) b) (1 − α1 )B βF (α1 b) + (1 − β)F (α1¯ ≤ α1 B βF (α1¯ + (1 − β)F (α1 b) ⇔ ν(α1 ) ≤ ω(α1 )

It is easy to prove that ν() is a decreasing function of α while ω is an increasing function of α, so that: ν(α2 ) ≤ ν(α1 ) ≤ ω(α1 ) ≤ ω(α1 ) ⇒ β ≥ β1 (α2 ) This proves that β1 (α1 ) ≥ β1 (α2 ). We now investigate how β2 evolves with t. We directly compute the derivative: ∂β2 b) b) α(1 − α) F (α¯ + F (ατ ¯ = ≥0 2 F (α¯ − F (ατ ¯ ∂t b) b) (1 − α + αt) Therefore, an increase in the Decision Maker’s bias (i.e. a decrease in t) makes her more reluctant to select project 2 after having received signal 2. Similarly, we can compute the derivative of β1 with respect to t. We ﬁnd: ∂β1 F (α¯ + F (ατ ¯ b) b) α(1 − α) = ≥0 ∂t b) b) (α + (1 − α)t)2 F (α¯ − F (ατ ¯ Here, an increase in the Decision Maker’s bias (i.e. a decrease in t) makes her less reluctant to select project 1 after having received a signal 1. For this particular incentive constraint, a higher Decision Maker’s bias is actually helping reactivity to emerge. The natural question is therefore whether overall reactivity is impaire or enhanced by an increase in the Decision Maker’s bias. Fortunately, it is to compute the derivative of β2 − β1 with respect to t: ¡ ∂ β 2 − β1 α(1 − α)(1 − t2 )(2α − 1) F (α¯ + F (ατ ¯ b) b) = ≥0 ∂t b) b) (α + (1 − α)t)2 (1 − α + αt)2 F (α¯ − F (ατ ¯ Overall, an increase in the Decision Maker’s bias decreases the scope for reactivity. We turn to the sensitivity of β2 with respect to τ : ∂β2 α¯ (ατ ¯ (α¯ bf b)F b) αt − (1 − α) = ¡2 × ∂τ 1 − α + αt F (α¯ − F (ατ ¯ b) b) As one can see, the link between β2 and τ is ambiguous. A higher Implementer’s bias (i.e. a decrease in τ ) makes the Decision ¯ B Maker more reluctant to select project 2 after having received signal 2 only when the signal is suﬃciently precise (i.e. α ≥ B+B ). To ¯ have a more interesting comparative static for the scope for reactivity, we need thus to look at β1 . As one can easily compute: ∂β1 α¯ (ατ ¯ (α¯ bf b)F b) (1 − α)t − α = ≤0 ¡2 × ∂τ (1 − α)t + α F (α¯ − F (ατ ¯ b) b) Therefore, a more biased Implementer makes it more attracting for the Decision Maker to select project 2 after having received signal 1. Overall, we can now look at the scope for reactivity as a function of τ : ¡ ¡ ∂ β2 − β1 2 α2 − (1 − α)2 t α¯ (ατ ¯ (α¯ bf b)F b) = ¡2 × ∂τ ((1 − α)t + α) (αt + (1 − α)) F (α¯ − F (ατ ¯ b) b) Which proves the announced proposition. QED

B.1.2

Proof of Proposition 2

We ﬁrst investigate the pooling equilibrium where the Decision Maker always selects project 1. In such an equilibrium, the Implementer’s posterior belief is given by: µ1 = 1/2, as, on the equilibrium path, no information on the state of nature can be inferred from the Decision Maker’s project selection. However, out of the equilibrium path, any belief is a priori admissible, i.e. µ2 ∈ [1 − α, α].

References

11

We look for restrictions on these out-of-equilibrium beliefs imposed by the D1 reﬁnement. Let us brieﬂy introduce some notations. D D Call U1 (resp. U2 ) the equilibrium utility of a Decision Maker receiving signal 1 (resp. signal 2). Call U1 (µ2 ) (resp. U2 (µ2 )) the out-of equilibrium utility of a Decision Maker receiving signal 1 (resp. signal 2) when out-of-equilibrium beliefs are given by µ2 , i.e. the utility a Decision Maker gets by selecting project 2. ¨ © ¨ © D 0 D Call Di = µ2 ∈ [1 − α, α] | Ui < Ui (µ2 ) and Di = µ2 ∈ [1 − α, α] | Ui = Ui (µ2 ) Deﬁnition (D1 Reﬁnement) 0 If D1 D1 ⊆ D2 , then µ2 = 1 − α. In words, if each out-of-equilibrium belief µ2 that leads to a proﬁtable deviation for a Decision Maker with signal 1 also leads to a strictly proﬁtable deviation for a Decision Maker with signal 2, then the Implementer must believe that only Decision Makers with signal 2 deviate from the equilibrium, i.e. that µ2 = 1 − α.

D Therefore, let µ2 ∈ [1 − α, α] such that U1 ≤ U1 (µ2 ) (i.e. µ ∈ D1

0 D1 ). This means that: ¯ ¡ b b ¯ b) (1 − α)B βF ((1 − µ2 )b) + (1 − β)F ((1 − µ2 )¯ ≥ αB βF ( ) + (1 − β)F ( ) 2 2

But then, because α > 1 − α, we have that:
¡ D ¯ U2 (µ2 ) = αB βF ((1 − µ2 )b) + (1 − β)F ((1 − µ2 )¯ > (1 − α)B b) ¯ b b βF ( ) + (1 − β)F ( ) = U2 2 2

0 So that µ2 ∈ D2 , which implies D1 D1 ⊆ D2 . Therefore, with our equilibrium concept (i.e. Perfect Bayesian Equilibrium with D1 reﬁnement), this pooling equilibrium must necessarily verify µ2 = 1 − α.

There are a priori two incentives constraints to check for this pooling equilibrium to exist. The ﬁrst one is that a Decision Maker receiving signal 2 should prefer to select project 1. This can be written as: ¯ ¡ b b ¯ (1 − α)B βF ( ) + (1 − β)F ( ) ≥ αB βF (αb) + (1 − β)F (α¯ ⇔ β ≥ β2 b) 2 2 The other incentive constraint states that a Decision Maker receiving signal 1 should also prefer to select project 1. This incentive constraint is equivalent to: ¯ ¡ b b ¯ αB βF ( ) + (1 − β)F ( ) ≥ (1 − α)B βF (αb) + (1 − β)F (α¯ b) 2 2 But if β ≥ β2 , then, because α > 1 − α: ¯ αB
¯ b b βF ( ) + (1 − β)F ( ) 2 2

> ≥ >

¯ b b ¯ (1 − α)B βF ( ) + (1 − β)F ( ) 2 2 ¡ b) αB βF (αb) + (1 − β)F (α¯ ¡ (1 − α)B βF (αb) + (1 − β)F (α¯ , b)

so that the ﬁrst incentive constraint (a Decision Maker with signal 2 must select project 1) implies the second incentive constraint (a Decision Maker with signal 1 must select project 1). We now prove that β2 ≥ β2 . Let β ≥ β2 . Then: ¯ ¡ b b ¯ (1 − α)B βF ( ) + (1 − β)F ( ) ≥ αB βF (αb) + (1 − β)F (α¯ b) 2 2 Using the fact that α > 1/2, this implies that: ¡ ¡ ¯ (1 − α)B βF (αb) + (1 − β)F (α¯ ≥ αB βF (αb) + (1 − β)F (α¯ ⇒ β ≥ β2 b) b)

We thus have β2 ≥ β2 . The derivation of the other pooling equilibrium as well as the proof that β1 ≤ β1 is exactly similar and are thus skipped. We now turn to the comparative statics in the thresholds β2 and β1 . Let us focus on the proof for β2 , the proof for β1 being just symmetric. Let α2 > α1 and let β ≤ β2 (α1 ). This implies that: ¯ ¡ b b ¯ (1 − α1 )B βF ( ) + (1 − β)F ( ) ≤ α1 B βF (α1 b) + (1 − β)F (α1¯ b) 2 2 But α2 > α1 and 1 − α2 < 1 − α1 , so that: ¯ ¡ b b ¯ (1 − α2 )B βF ( ) + (1 − β)F ( ) ≤ α2 B βF (α1 b) + (1 − β)F (α1¯ ⇒ β ≤ β2 (α2 ) b) 2 2 This proves that: β2 (α2 ) ≥ β2 (α1 ).

References

12

We can compute the derivative of β2 with respect to t: ∂β2 ∂t α(1 − α)F (α¯ (¯ b)F b/2) = ¡ ¡¡2 ≥ 0 ¯ b) (1 − α) F (b/2) − F (b/2) + αt F (α¯ − F (αb)

Finally, let τ2 > τ1 and let β ≤ β2 (τ1 ). This implies that: ¯ ¡ ¡ b b (1 − α) βF ( ) + (1 − β)F ( ) ≤ ατ1 βF (αb) + (1 − β)F (α¯ < ατ2 βF (αb) + (1 − β)F (α¯ b) b) 2 2 So that β ≤ β2 (τ2 ) and β2 (τ1 ) ≤ β2 (τ1 ). QED.

B.1.3

Proof of Proposition 3

We ﬁrst show that there cannot be equilibrium where the Decision Maker would randomize strictly for both values of the signal. If such an equilibrium were to exist, then the Decision Maker would have to be indiﬀerent between both project, for both value of the signal, otherwise she would not be strictly randomizing. Call κ the probability of selecting project 1 after receiving a signal 1 and χ the probability of selecting project 1 after having received signal 2. The indiﬀerence condition mentioned above can be written as: @ ¡ ¡ ¯ b) b) αB βF (µ1¯ + (1 − β)F (µ1 b) = (1 − α)B βF ((1 − µ2 )b) + (1 − β)F ((1 − µ2 )¯ ¡ ¡, ¯ (1 − α)B βF (µ1¯ + (1 − β)F (µ1 b) = αB βF ((1 − µ2 )b) + (1 − β)F ((1 − µ2 )¯ b) b) where µi is the posterior belief that state of nature is 1 when project i has been selected. Dividing the two previous equations yields: α/(1 − α) = (1 − α)/α which cannot happen as α > 1/2. Therefore, a mixed equilibrium can only involve the Decision Maker randomizing strictly for one particular value of the signal. Let us focus on the equilibrium where the Decision Maker randomizes when she receives signal 2, ρ ∈]0, 1[ being the probability that she selects project 1 in that case. We proved earlier that posterior beliefs, imposed by Baye’s law, are given by: V b ` µ1 (ρ) = α + ρ(1 − α) 1+ρ b X µ (ρ) = 1 − α
2

Because she randomizes when receiving signal 2, the Decision Maker must be indiﬀerent between project 1 and project 2 when receiving such a signal. This implies that: ¡ ¡ ¯ b) b) (B.1) αB βF (αb) + (1 − β)F (α¯ = (1 − α)B βF (µ1 (ρ)¯ + (1 − β)F (µ1 (ρ)b) The other Decision Maker’s incentive constraint should verify that, after receiving signal 1, the Decision Maker wants to select project 1. Using the fact that α > 1 − α, this is indeed the case as: ¡ ¡ ¯ ¯ αB βF (µ1 (ρ)¯ + (1 − β)F (µ1 (ρ)b) b) > (1 − α)B βF (µ1 (ρ)¯ + (1 − β)F (µ1 (ρ)b) b) ¡ = αB βF (αb) + (1 − β)F (α¯ b) ¡ > (1 − α)B βF (αb) + (1 − β)F (α¯ b)

Therefore, when condition B.1 is veriﬁed, the semi-reactive equilibrium is an equilibrium. Condition 3.12 deﬁnes an implicit function β(ρ), for each ρ ∈]0, 1[, which is the only level of congruence for which the ρ-mixed equilibrium is an equilibrium. µ1 () is strictly decreasing in ρ and goes from 1/2 to α. β(ρ) is uniquely deﬁned as: β(ρ) = ¯ αBF (α¯ − (1 − α)BF (µ1 (ρ)b) b) ¡ ¡ ¯ αB F (α¯ − F (αb) + (1 − α)B F (µ1 (ρ)¯ − F (µ1 (ρ)b) b) b) (B.2)

It is easy to prove that β(ρ) is an increasing function of beta. Let 1 > ρ1 > ρ2 > 0. By deﬁnition of β(ρ1 ), we have: ¡ ¡ ¯ b) b) αB β(ρ1 )F (αb) + (1 − β(ρ1 ))F (α¯ = (1 − α)B β(ρ1 )F (µ1 (ρ1 )¯ + (1 − β(ρ1 ))F (µ1 (ρ1 )b)

We know that µ1 () is a strictly decreasing function of ρ, so that: ¡ ¡ ¯ αB β(ρ1 )F (αb) + (1 − β(ρ1 ))F (α¯ < (1 − α)B β(ρ1 )F (µ1 (ρ2 )¯ + (1 − β(ρ1 ))F (µ1 (ρ2 )b) b) b) ¯ αBF (α¯ − (1 − α)BF (µ1 (ρ2 )b) b) ¡ ¡ = β(ρ2 ) ⇔ β(ρ1 ) > ¯ αB F (α¯ − F (αb) + (1 − α)B F (µ1 (ρ2 )¯ − F (µ1 (ρ2 )b) b) b) So that β is indeed strictly increasing in ρ, which implies that ρ, deﬁned as a function of β is a strictly increasing function of β. It is direct to check that for β = β2 , the indiﬀerence condition B.1 is satisﬁed for ρ = 1 and that for β = β2 , it is satisﬁed for ρ = 0. The derivation of the semi-reactive equilibrium with the Decision Maker randomizing conditional on receiving signal 1 is exactly similar and is therefore skipped. QED.

References

13

B.1.4

Proof of Proposition 4

We begin this proof by showing that the value of a semi-reactive organization, when it exists, is a decreasing function of β. We consider the case of a semi-reactive equilibrium where the Decision Maker’s decision rule strictly randomizes over project 1 and 2 only when receiving signal 2 (Point 2 of proposition 3). The proof for the other semi-reactive equilibrium is similar. The value of such a semi-reactive equilibrium is given by: ¡ 1 ¡ 1 (α + (1 − α)ρ) βF (µ1 (ρ)¯ + (1 − β)F (µ1 (ρ)b) + α(1 − ρ) βF (αb) + (1 − β)F (α¯ b) b) R, V SR = 2 2 where ρ is given by the indiﬀerence equation 3.12. Using this indiﬀerence equation, we can rewrite organization’s value in a very simple fashion: ¡ 1 αB 1 ¯ V SR = (α + (1 − α)ρ) ¯ + 2 α(1 − ρ) βF (αb) + (1 − β)F (αb) R 2 (1 − α)B We can derive now the organization’s value with respect to β, being aware that ρ is an implicit function of β: dV SR ∂V SR ∂ρ ∂V SR = + dβ ∂ρ ∂β ∂β But we have: ¯ ¡ ∂V SR B−B = βF (αb) + (1 − β)F (α¯ αR ¯ b) ≤0 ∂ρ B And: ∂V SR = ∂β
¡ αB 1 1 ¯ (α + (1 − α)ρ) ¯ + 2 α(1 − ρ) F (αb) − F (αb) ≤ 0 2 (1 − α)B
∂ρ ∂β

(B.3)

Moreover, we have proved in proposition 3 that ρ is an increasing function of β, so that: evolves monotonically with β as: ∂V SR ∂ρ ∂V SR dV SR = + ≤0 dβ ∂ρ ∂β ∂β | {z } |{z} | {z }
≤0 ≥0 ≤0

≥ 0. Therefore, organization’s value

Therefore, setting aside the conditions of existence of a reactive equilibrium, we have that: V SR ≤ V SR (β = β2 ) = V R . If the reactive equilibrium exists, then the Owner always prefers this equilibrium to the semi-reactive equilibrium. Deﬁne now α0 as in proposition 4: ¡ α0 F (α0¯ + F (α0 b) = F (¯ b/2) b)

¯ First, assume that: α0 BF (α0¯ ≥ (1 − α0 )BF (αb). b) This condition implies that for α = α0 β2 > 0, so there exist a range of β ∈ [0, 1] with positive measure supporting a reactive equilibrium. Moreover, we also have for α = α0 that β2 < 1. Indeed, let α = α0 and let β = 1. Then: ¯ b/2) (1 − α0 )BF (¯ α0 F (α0 b) ≥ (1 − α0 )BF (¯ b/2)

≥ α0 BF (α0 b), 0 F (α0 ¯ < (1 − α0 )F (¯ ¯ since = F (b/2) − α b) b/2). This means that for β = 1 and α = α0 , the Decision Maker’s strategy consisting in always selecting project 1 is an equilibrium, or equivalently that β2 < 1. Therefore, for α = α0 , both a reactive and a non-reactive equilibrium exists. By deﬁnition of α0 , we know that at α = α0 , reactive and “optimal” (i.e. β = 1) non-reactive organizations deliver the same proﬁt to the Owner. Moreover, as β2 and β2 are increasing functions of α (see proof of proposition 1 and 2, we know that for α ≤ α0 , a non-reactive equilibrium always exists while for α ≥ α0 , a reactive equilibrium always exists. Finally, since the reactive organization’s value is strictly increasing with α while the non-reactive organization’s value is independent of α, we conclude that for α ≥ α0 the optimal organization is reactive while for α ≤ α0 , it is non-reactive. Note that semi-reactive organizations are always dominated in this case, as for α ≤ α0 , the non-reactive organization delivers a higher proﬁt than that of the reactive organization, which itself is more proﬁtable that the semi-reactive organization (see above), while, for α ≥ α0 , the reactive organization always exists and therefore dominates the semi-reactive equilibrium. This proves the proposition’s ﬁrst bullet point. Before turning to the second bullet point of proposition 4, let us prove that V SR is, for a given level of congruence β, an increasing function of α. Consider ﬁrst ρ as a function of α, for a given β. Then one can diﬀerentiate the indiﬀerence equation 3.12 with respect to α: ¡ ¡ ¡ ¯ b) b) bF b) B βF (αb) + (1 − β)F (α¯ + B βF (µ1 (ρ)¯ + (1 − β)F (µ1 (ρ)b) + αB βbF (αb) + (1 − β)¯ (α¯ | {z }
V W b b b b b b b b ¡a ∂µ1 (ρ) ¯ ∂ρ ` ¯ (1 − α)B β bf (µ1 (ρ)¯ + (1 − β)bF (µ1 (ρ)b) b) b ∂α b ∂ρ | {z }b b b b | {z } b b X Y ≥0
≤0 ≥0

=

References

14

This implies that

∂ρ ∂α

≤ 0. Finally, diﬀerentiating the semi-reactive organization’s proﬁt B.3 with respect to α: ∂V SR ∂V SR ∂ρ dV SR = + ≥0 dα ∂α ∂ρ ∂α | {z } | {z } |{z}
≥0 ≤0 ≤0

So that we proved that V SR is an increasing function of α. ¯ b) We now turn to the second bullet point of proposition 4. Therefore, consider the case where: α0 BF (α0¯ < (1 − α0 )BF (αb). This means that for α = α0 , β2 < 0, so that a reactive organization fails to exist for this level of signal’s precision. b/2)R, and α1 such that: α1 BF (α1¯ = (1 − α1 )BF (b/2). ˆ ˆ ˆ ¯ Deﬁne α1 such that: V SR (α1 , β = 0) = V N R (β = 1) = 1 F (¯ ˜ ˆ b) 2 Finally, deﬁne α1 and α2 such that: @ α1 = max{α1 , α1 } ˜ ˆ ¯ b) α2 BF (α2¯ = (1 − α2 )BF (α2 b) First, for α ≥ α1 , we know that (1) a semi-reactive equilibrium exists, i.e. β2 ≥ 0 and (2) the semi-reactive organization with β = 0, if it exists, is more proﬁtable than the non-reactive organization. For α ≤ α2 , we know that the reactive organization does not exist, or equivalently that β = 0 supports a semi-reactive equilibrium. Let us prove that α2 ≥ α1 . Assume it is not the case so that α1 > α2 . First, if α1 = α1 , then there is a contradiction as for ˆ α ∈]α2 , α1 [, β2 < 0 but β2 > 0 which is not possible as β2 > β2 . Second, assume that α1 = α1 . Then, by deﬁnition of α1 , we have ˜ ˜ formally: V N R = V SR (α1 , β = 0) ≥ V SR (α2 , β = 0) = V R (α2 ) as α2 is such that β2 = 0, i.e. that a reactive equilibrium exists for β = 0. But then, this must implies that α2 ≤ α0 as the value of a reactive organization for α = α2 is inferior to the value of a non-reactive equilibrium. As β2 (α2 ) = 0, α2 < α0 and β2 is increasing with α, this must imply that β2 (α0 ) > 0 which is a contradiction. Thus: α2 ≥ α1 . Moreover, we have that α1 ≥ α0 (of course, α0 ≤ α2 otherwise this would contradict directly the deﬁnition of α0 (i.e. α0 would be such that a reactive equilibrium exists for β = 0)). By contradiction, assume that α0 > α1 . Then, α0 is such that there is a semi-reactive equilibrium for β = 0 (as α0 ≤ α2 ). Thus: V SR (α0 , β = 0) > V SR (α1 , β = 0) = V N R . But we know that: V SR (α0 , β = 0) ≤ α0 (F (α0¯ b) 2

+ F (α0 b)) so that: α0 (F (α0¯ + F (α0 b)) > V N R b) 2

which contradict the deﬁnition of α0 . Therefore, we have α0 ≤ α1 ≤ α2 . For α ≤ α1 , a pooling equilibrium always exist (optimal with β = 1) and is optimal compared to semi-reactive equilibrium which are the only other equilibrium that exist in this range of signal’s precision. For α ∈ [α1 , α2 ], semireactive organizations are optimal (with β = 0 as their value is decreasing with β and β = 0 always support a semi-reactive equilibrium in this range of α) over non-reactive equilibrium, because of the deﬁnition of α1 and α2 and because of the monotonicity in α of the semi-reactive organization’s value. Finally, for α ≥ α2 , the optimal organization is reactive (because of the monotonicity in α of the reactive organization’s value, because semi-reactive organizations are less proﬁtable than reactive organizations when such organizations exist, and because of the deﬁnition of α2 ) and can be supported by any β ∈ [β1 , β2 ]. QED.

B.1.5

Proof of Proposition 5

As we already proved in the main text, the optimal reactive organization for the Decision Maker is the one with β = β2 . Moreover, as soon as a reactive equilibrium is feasible (i.e. as soon as β2 > 0), then this organization with β = β2 can be put in place. Assume that a reactive organization is feasible (β2 > 0). In that case, a semi-reactive organization with the Decision Maker randomizing only after having received signal 2 is also feasible. Using the indiﬀerence condition 3.12, one can easily compute the Decision Maker’s ex ante utility from such a semi-reactive organization (at least for any ρ ∈]0, 1[ such that the semi-reactive equilibrium exists): ∀β ∈]β2 , β2 [, V SR = ¡ 1 α B βF (αB) + (1 − β)F (α¯ b) 21−α

Therefore, the value of a semi-reactive organization is decreasing with β. It is thus maximized for β = β2 , i.e. for a reactive organization. Thus, as long as a reactive organization is feasible, the Decision Maker will always prefer a reactive organization over a semi-reactive one. The Decision Maker, provided that reactive and non-reactive organizations are feasible (i.e. that β2 < 1 and β2 > 0), must therefore choose between a reactive and a non reactive organization from which he gets utility given by expression 3.18 and 3.19. This trade-oﬀ corresponds to the trade-oﬀ we exhibited in expression 3.20. In particular, because β2 is an increasing function of α (see proof of proposition 1), we see that the Decision Maker’s utility from a reactive organization is an increasing function of α while it is independent of α for non-reactive organizations. Deﬁne now α1 ∈]1/2, 1[ the unique value of α such that the value of a reactive organization equals that of a non reactive organization: ¯ ¯ ¯ ¡ B B b/2) β2 (α1 )F (α1¯ + (1 − β2 (α1 )F (α1 b)) = F (¯ ¯ ¯ b) ¯ ¯ 2 2 First note that this α1 is indeed unique and in ]1/2, 1[: ¯

References

15

¡ ¯ b) Proof if ψ(α) = B β2 (α)F (α¯ + (1 − β2 (α)F (αb)) − 2 of α from proposition 1.

¯ B F (¯ b/2), 2

then ψ is strictly increasing with α as β2 () is an increasing function

Proof β2 (1) = 1 so that ψ(α = 1) > 0 Proof β1 (1/2) < 1 so that ψ(1/2) < 0, so that ﬁnally, α is uniquely deﬁned and belongs to ]1/2, 1[. ¯ Deﬁne also α2 as the unique α ∈]1/2, 1[ such that a non-reactive organization with the Decision Maker always selecting project 2 ¯ fails to exist: (1 − α2 )F (¯ B = α2 BF (α2 b) ¯ b/2) ¯ ¯ ¯ The proof that α2 is uniquely deﬁned in the interval ]1/2, 1[ is similar to the proof for α1 and is thus skipped. ¯ ¯ Deﬁned α = min α1 , α2 . ¯ ¯ ¯ We are going to prove that under condition 3.21, α > α0 , so that for α ≥ α a reactive equilibrium exists and is preferred by the ¯ ¯ Decision Maker while for α ≤ α a non-reactive organization exists and is preferred by the Decision Maker. ¯ First consider any α such that α ≥ α1 , then by deﬁnition of α1 (using expression 3.17 for the Decision Maker’s utility in reactive ¯ ¯ organizations): U R (β) = ¯ £ α¢ B α ¯ [β.F (α¯ + (1 − β).F (αb)]B + b) b/2) = U N R (1 − β).F (α¯ + β.F (αb) B ≥ F (¯ b) 2 2 2
£ α¢ 1 F (α¯ + F (αb) > F (¯ b) b/2) 2 2

¯ Therefore, because B > B:

Which proves that α1 > α0 . ¯ This implies that β2 (α1 ) > 0 so that a reactive equilibrium exists for α ≥ α1 and using the deﬁnition of α1 , this implies that ¯ ¯ ¯ reactivity is the optimal organizational form for this range of α. Second, as we saw in the proof of proposition 4, for α = α0 , we have β2 < 1, so that, as β2 (α2 ) = 1 and β2 is an increasing ¯ function of α (see proposition 2, this must imply that: α2 > α0 . ¯ Therefore: α > α0 . For α > α, the Decision Maker’s optimal organization is a reactive one (with any β ∈ [β1 , β2 ]) either because ¯ ¯ such an organization delivers a higher utility (when α ≥ α1 ) or because a non-reactive organization is not feasible and a reactive ¯ organization delivers a higher utility than a semi-reactive organization (for α ≥ α2 ) In this range of α, thanks to condition 3.21, a ¯ ¯ reactive organization does indeed exist. For α ≤ alpha, the Decision Maker would select a non-reactive organization with a perfectly congruent Implementer (i.e. β = 1), and by deﬁnition of α, we are sure that such an organization exists in this range of α. ¯ Therefore, over the range ]α0 , α[, which we proved is non empty, the Decision Maker would select a non-reactive organization while ¯ the Owner would select a reactive organization: organizational design in these cases should not be delegated to the Decision Maker. For all other values of α, the Decision Maker would design an optimal organization from the Owner’s point of view. QED.

B.1.6

¡ We thus assume that F is uniform over [0,X], where X ≥ ¯ + R so that F is always deﬁned over the relevant range. Call wi,j , the b Decision Maker’s wage when project i is successful and the Implementer’s preferred project is project j. Note zi,j , the Implementer’s wage when project i is implemented and successful and the Implementer’s preferred project is project j. When the organization is designed to be reactive, the Owner’s program can be written as:
& ¡ α ¯ max wi,j ,zi,j ,β 2 βα(b + z11 ) (R − z11 − w11 ) + (1 − β)α(b + z12 ) (R − z12 − w12 ) ' ¡ α + βα(b + z21 ) (R − z21 − w21 ) + (1 − β)α(¯ + z22 ) (R − z22 − w22 ) b 2 α βα(¯ + z11 ) B + w11 ¡ + (1 − β)α(B + z12 ) B + w12 ¡¡ ≥ (1 − α) βα(¯ + z22 ) (B + w22 ) + (1 − β)α(b + z21 ) (B + w12 )¡ ¯ ¯ ¯ b b α βα(¯ + z22 ) (B + w22 ) + (1 − β)α(b + z21 ) (B + w12 )¡ ≥ (1 − α) βα(¯ + z11 ) B + w11 ¡ + (1 − β)α(B + z12 ) (R − z12 − w12 )¡ ¯ ¯ b b

Proof of Proposition 6

First of all, let us omit the two incentive constraints. Then, the unconstrained solution of the previous problem clearly features wi,j = 0, i.e. there is no need to give any rent to the Decision Maker once there is the incentive for decision making are omitted. Solving for the optimal zi,j is then very easy and independent from the chosen β : the Owner must give proper incentives to the Implementer for each case (i.e. each project selected and each Implementer’s preferred project). These four diﬀerent maximization problems in zi,j

are concave and therefore admit a unique solution given by:
V b bz = z = z = R −¯ b ` 11 22 b b Xz = z = z = R − b ¯ 12 21
2 2

In this case, the organization proﬁt can be rewritten as:
V
R

=

α2 2

¯ 2 R b α2 R b 2 + + + 2 2 2 2 2

Note ﬁrst that, even in the presence of monetary contracts, this proﬁt is independent of β : this is so because the optimal Implementer’s compensation is symmetric (i.e. provides the same reward to an Implementer implementing his preferred project, whatever this preferred project). Second, note that this (unconstrained) proﬁt is, when feasible (i.e. when β2 > 0), clearly the solution to the Owner’s problem. Thus, we need to check when this unconstrained solution is a solution to the constrained problem: we must check when the Decision Maker’s incentive constraints are veriﬁed with this solution.

References

16

We ﬁnd that reactivity is feasible with a strict level of dissent as long as: α2 = ˆ B

¯ B
R 2

+

¯ b 2

R 2

+

b 2

R 2

¯ +B

+

b 2

¯ Because B > B , a little computation gives: α2 > α1 ˆ ˆ
2 1 2 R 2

+

¯ b 2

2

R 2

+ +

¯ b 2

2
+
b 2 2 ˆ 2 = α1 ∈]1/2, 1[

R 2

Therefore, for all α ∈ [min α1 , α2 , α2 ], the optimal organizational is a reactive organization with strict dissent i.e. β < 1. Thus, ˆ ˆ ˆ complete contracting is not in general a limit to the use of dissent in organizations. QED.

B.2

Proof of Proposition 7

Consistently with the analysis of Section 3.3, it is straightforward to show - proof omitted - that: (1) for intermediate values of β, the reactive equilibrium is feasible and (2) for extreme values of β , the non-reactive equilibrium is feasible. In between, there are semi-reactive equilibria where the Decision Maker only partially reacts to the signal. We omit here the analysis pertaining to very low β s, where the Decision Maker always follows the Implementer’s bias, as, for a broad range of parameters such equilibria are not even feasible. We also omit the analysis of semi-reactive equilibria, because (1) it is similar to part 3.3 and (2) such arrangements are never optimal from the Owner’s perspective. In this new, asymmetric, model the equivalent to proposition 4 is given by:
Lemma When the Decision Maker intrinsically prefers the“status quo”project:
∗∗ 1. A non-reactive equilibrium where the “status quo” is always selected occurs for all β ∈ [ξ2 (θ); 1]. 2. A reactive equilibrium where the Decision Maker chooses the project indicated by her private signal occurs for all ∗ ∗ ∗ ∗∗ β ∈ [ξ1 (θ); ξ2 (θ)], with ξ2 (θ) < ξ2 (θ)

Proposition 8 1. When the Decision Maker intrinsically prefers the “change” project:
∗∗ 1. A non-reactive equilibrium where the “status quo” is always selected occurs for all β ∈ [β2 (θ), 1]. ∗ ∗ 2. A reactive equilibrium where the Decision Maker selects the project indicated by the signal occurs for all β ∈ [β1 (θ), β2 ](θ), ∗ ∗∗ with β2 (θ) < β2 (θ). Provided that β is not “too low,”a reactive equilibrium is easier to obtain with a pro-“change” Decision Maker, i.e.: ∗∗ ∗∗ ∗ ∗ ∗ ∗ ξ2 (θ) < β2 (θ), ξ2 (θ) < β2 (θ) and ξ1 (θ) < β1 (θ)

When θ increases (i.e., uncertainty decreases), reactivity becomes more diﬃcult to sustain for a given level of congruence, β, provided this congruence is not too high, i.e.:
∗∗ ∗∗ ∗ ∗ ∗ ∗ ξ2 (θ), β2 (θ) , ξ2 (θ), β2 (θ), ξ1 (θ) and β1 (θ)

θ

In this context, the net gain of a reactive organization is given by:
∗∗ ∆(θ) = V R (θ, β2 (θ)) − V R (θ)

where: V R (θ, β) =

! θα . βb + (1 − β)b θα + (1 − θ).(1 − α) ! (1 − θ).α +(1 − θ).α. . βb + (1 − β)b θ(1 − α) + (1 − θ).α

θα.

and: V R (θ) = θ2 .b

References

17

The marginal eﬀect of an increase in θ is written as:
d∆ ∂V db∗∗ ∂ R = . 2 + V − V NR dθ ∂β dθ ∂θ {z } |{z} | {z } |
>0 1/2. As a result, this second derivative is positive. Thus, ∂G/∂θ is an increasing function of θ, and: ∂G (β, 1) = 16α2 .(1 − α)2 .(b + b) − 2b < −b + b < 0 ∂θ

thus, ∂G/∂θ is negative, and G is decreasing. But: 1 = 2β. b − b α2 .(3 − 2α) − b < −b < 0 G β, 2 so G < 0 for all θ > 1/2. QED.

B.3

The True Role of Uncertainty

In order to disentangle the eﬀect of uncertainty about the state nature from the eﬀect of uncertainty about the implementer’s preferences, we provide here a simple extension of our main model, where:
• There are N > 2 possible projects, indexed by i ∈ [1; N ] • The decision maker has no bias, she is just interested in maximizing the probability of success • The implementer receives b > 0 if one particular project succeeds, and 0 else. • From the DM’s viewpoint, the implementer prefers project 1 with probability β , and all the others with probability (1 − β)/(N − 1) • The DM receives a private signal on the state of nature. α is the probability that the state is i if the signal says “i”. The state can be any other j with probability (1 − α)/(N − 1).

In this model, β ≥ 1/N parameterizes uncertainty of the DM about the implementer’s type. N parameterizes fundamental uncertainty about the right course of action. We ﬁrst look for reactive equilibria. The binding condition for them to exist is that a DM receiving a signal that is not “1” follows it nonetheless. The cost of following such a signal is that the implementer is likely not to like it. This happens iﬀ:
α. 1−α 1−β .F (αb) > .β.F (αb) ⇐⇒ β < α N −1 N −1

where the DM faces the tradeoﬀ of following an informative signal and of betting on the likely preferences of the implementer. As a result, the reactive equilibrium is sustainable as long as the uncertainty on the implementer’s type is high enough compared to the signal precision. The analysis of pooling equilibria is also similar to that of our main model. For such an equilibrium to emerge, it has to be that the DM wants to resist the temptation to give an order consistent with signals that are not “1”. To simplify matters, we use the same reﬁnement as in proposition 4: beliefs held by an implementer that receives the out-of-equilibrium order j = 1 are that the DM has told the truth. Thus, pooling equilibria are sustainable iﬀ:
1−α .β.F N −1

b N

> α.

1−β .F (αb) ⇔ β > β > α N −1

thus, pooling equilibria are sustainable when β is large enough compared to α. The model makes formally the point of the main text: as uncertainty about the implementer’s type increases (β is reduced), reactivity becomes easier to sustain. Therefore, in cases where reactivity is valuable (α is large), total DM ignorance about the implementer’s type is more likely to be optimal (β = 1/N ).

References

18

Appendix C

The Corporate Wealth Eﬀect
C.1 Construction of the pt Landi Variable

In this section, we explain how we have constructed the pt Landi variable used in column 6 of table 4.4. To derive a simple quantitative assessment of our eﬀect, we want to regress investment on real estate capital gains made on initial land holding. The variable we have at our disposal in COMPUSTAT, item #260, provides us with the book value of initial land holdings. Calling pt0i the “composite” price of land purchased by ﬁrm i before the ﬁrst year of appearance in COMPUSTAT and Li the quantity (square feet) of land purchased, we have: item#260 = pt0i × Li The current market value of these initial land holdings is given by:

2 pt Li = pt0i × Li ×

pt pt0i

3
= item#260 ×

2

pt pt0i

3

We thus need to provide a proxy for this “composite price” of initial land holdings, as we do not observe the date of purchase of these real estate assets owned by ﬁrms initially. The most conservative proxy we can use assumes that ﬁrms purchased these initial real estate assets in their ﬁrst year of appearance in COMPUSTAT (and not necessarily in our sample). Given that real estate inﬂation is positive on average, we will tend to underestimate the market value of real estate assets purchased after the birth date of a ﬁrm, and therefore we will tend to underestimate the sensitivity of investment to collateral. Using such an assumption, we are able to compute a proxy for the ratio 1975. We need to deﬁne this ratio ps t pts 0i pt pt 0i

for

most ﬁrms in our sample. A last diﬃculty with computing this ratio stems from the fact that real estate price series are available only since for ﬁrms appearing in COMPUSTAT before 1975. We assume that before 1975, real estate inﬂation in

state s is equal to the average real estate inﬂation in this state over the 1975-2004 period. Using this assumption, we can construct backward real estate prices for all years before 1975. Note that this approximation concerns 900 ﬁrms over the 8,493 ﬁrms that constitute our sample. Using this approximation and the assumption that real estate assets are purchased in the ﬁrst year of appearance in COMPUSTAT, we are able to deﬁne, for each ﬁrm in the sample, the variable pt Li /Kt−1 Once we have deﬁned this proxy, the regression we estimate amounts to: s Pt Li CAP EXit ∼β× , P P Eit−1 P P Eit−1

which can be rewritten as:

∆ (CAP EXit ) ∼ β × ∆ Pt .Li

s

¡ }

|

{z

capital gain

Column 6 of table 4.4 reports that β is signiﬁcantly positive and equal to .60, implying that a $1 capital gain on land holdings translates in a $.60 increase in capital expenditure.

Essais En Finance D’entreprise

Similar Documents

Wfnx

Microfinance Et Pauvreté : Quantification de La Relation Sur La Population de Tlemcen

Fr Comm

Marly

Management

Etude Pays Norvège

Djcauxde Report

Marketing Viral

The Beginning of Capitalism During the Middle Age

Economie

Analyse de La Trajectoire Historique de La Monnaie Electronique

Steve Jobs Biography

Analysis of International Marketing

Popular Essays