Arnaud Z. Dragicevic - RILK - Bienvenue sur RILK - La société
Arnaud Z. Dragicevic - RILK - Bienvenue sur RILK - La société
Arnaud Z. Dragicevic - RILK - Bienvenue sur RILK - La société
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
THÈSE<br />
Pour l‟obtention du grade de<br />
Docteur de l‟École Polytechnique<br />
Spécialité : Sciences économiques<br />
Présentée et soutenue publiquement par<br />
<strong>Arnaud</strong> Z. <strong>Dragicevic</strong><br />
Le 4 Décembre 2009<br />
MARKET MECHANISMS AND VALUATION<br />
OF ENVIRONMENTAL PUBLIC GOODS<br />
MÉCANISMES DE MARCHÉ ET ÉVALUATION<br />
DES BIENS PUBLICS ENVIRONNEMENTAUX<br />
Directeur de thèse : Bernard Sinclair–Desgagné<br />
Membres du jury<br />
Pr. Bureau, D., École Polytechnique et Conseil économique pour le développement durable (président)<br />
Pr. Shogren, J., Université du Wyoming et Académie royale des sciences de Suède (rapporteur)<br />
Pr. Sinclair-Desgagné, B., École Polytechnique, HEC Montréal et CIRANO (directeur)<br />
Pr. Willinger, M., Université de Montpellier 1, Institut universitaire de France et LAMETA (rapporteur)
L'École Polytechnique n’entend donner aucune approbation,<br />
ni improbation, aux opinions émises dans les thèses.<br />
Ces opinions doivent être considérées comme propres à leur auteur.
Pour ma mère.
Remerciements<br />
Tout d‟abord, j‟aimerais remercier mon directeur de thèse, le Pr. Bernard<br />
Sinclair-Desgagné, pour la confiance qu‟il m‟a accordée et le goût de<br />
l‟exploration qu‟il m‟a transmis ; je le remercie ensuite pour son honnêteté<br />
intellectuelle et son grand enthousiasme communicatif ; pour m‟avoir apporté de<br />
précieux conseils et pour m‟avoir orienté sans obstruer ma liberté de penser, enfin.<br />
Ses méthodes de travail ont été une véritable inspiration pour moi. Sans lui, je<br />
n‟en serai pas là. Je lui suis très reconnaissant.<br />
Je sais également gré au Pr. Bertrand Munier pour avoir su animer la fibre<br />
de chercheur en moi, sans qui la notion de recherche aurait aujourd‟hui un tout<br />
autre visage. Mes pensées vont également à l‟ancienne équipe du GRID pour leurs<br />
débats animés et leurs appuis qui m‟ont donné l‟énergie nécessaire pour ce projet<br />
de thèse : Mohammed Abdellaoui, Marie-<strong>La</strong>ure Cabon-Dhersin, Nicolas Drouhin,<br />
Nathalie Etchart-Vincent et Marc <strong>La</strong>ssagne. Le soutien et la compréhension du Pr.<br />
Gérard Coffignal ont été d‟une importance primordiale. Un grand merci à mes<br />
complices doctoraux Aurélien Baillon, Thierno Diaw, <strong>La</strong>ëtitia Placido et Céline<br />
Tea pour leurs remarques, leur soutènement et leur gaîté.<br />
Je tiens à remercier l‟École Polytechnique ParisTech pour son accueil et la<br />
stimulation intellectuelle de tous les jours. Mes remerciements vont d‟abord au Pr.<br />
Jean-Pierre Ponssard dont l‟intervention a été capitale. Merci au Pr. Michel Rosso<br />
pour m‟avoir autorisé à poursuivre et finir mes recherches au sein du département<br />
d‟économie. Par ailleurs, un très grand merci aux Prs. Francis Bloch et Jean-<br />
François <strong>La</strong>slier pour nos conversations et leurs suggestions toujours avisées qui<br />
ont contribué à l‟amélioration de ce travail. Je tiens également à remercier Marie-<br />
<strong>La</strong>ure Allain, Claire Chambolle, Nicolas Houy, Yukio Koriyama et Ingmar<br />
Schumacher pour leur écoute et leur aide. Les Prs. Pierre Cahuc, Patricia Crifo et<br />
Jérôme Renaud m‟ont aussi beaucoup apporté. Une petite pensée pour Éliane
Nitiga-Madelaine, Lyza Racon et Chantal Poujouly qui ont fait preuve d‟une<br />
grande générosité à mon égard. Merci à Christine <strong>La</strong>vaur pour son appui ferme à<br />
plusieurs occasions. En dernier lieu, j‟aimerais remercier les doctorants de l‟EDX<br />
pour leur accueil chaleureux, pour nos dialogues en tous lieux, toutes heures et<br />
tous états, pour leurs décryptages et leurs conseils, ainsi que pour leur amitié :<br />
Ozlem Bedre, Clémence Berson, Damien Bosc, Clémence Christin, Julien<br />
Hardelin, Sabine Lemoyne de Forges, Fabienne Llense, Guy Meunier, Thuriane<br />
Mahé, Matias Nunez, François Perrot, Claudia Saavedra, Nicolas Schutz, Idrissa<br />
Sibally et Xavier Venel. Ils ont ma gratitude. Certains ont été témoins de l‟unique<br />
fois où l‟esprit d‟Éric Cantona a bien voulu habiter mon corps, bien que je sois<br />
très doué pour mal jouer au football.<br />
De <strong>sur</strong>croît, mes remerciements s‟adressent aux membres du jury pour<br />
m‟avoir grandement honoré en acceptant d‟évaluer cette thèse et pour m‟avoir<br />
recommandé, à l‟occasion de la pré-soutenance, des éléments essentiels pour la<br />
suite de mon travail. Je leur en suis obligé. Également, je tiens à remercier David<br />
Ettinger dont la relecture en rapporteur improvisé a été précieuse.<br />
Pour nourrir des idées, il faut des financeurs. C‟est pourquoi je tiens à<br />
remercier l‟École des Arts et Métiers ParisTech ainsi que le Ministère de la<br />
Recherche à qui je dois l‟entier financement de cette thèse. Merci également à la<br />
Chaire Développement Durable EDF–École Polytechnique pour avoir cru en mon<br />
projet d‟expérience et pour l‟avoir financé.<br />
Un grand merci au Pr. Dominique Namur ainsi qu‟à l‟École Supérieure<br />
d‟Électricité, plus singulièrement Stéphane Font, non seulement pour avoir cru en<br />
moi mais aussi pour m‟avoir appris à enseigner. L‟expérience professorale que j‟ai<br />
acquise durant les quatre années leur est due.<br />
Pour nourrir des idées, il faut aussi de l‟art qui agit à brûle-pourpoint. C‟est<br />
pourquoi cette thèse n‟aurait pu se faire sans les tableaux de Jérôme Bosch, sans<br />
les compositions du groupe Radiohead, sans les longs-métrages de Gus Van Sant<br />
ni les romans de Martin Page.<br />
Permettez-moi de louanger sans réserves mon compagnon de fortune,<br />
l‟ordinateur portable, pour ne pas m‟avoir lâché en cours de route, pour avoir tenu
on jusqu‟à la dernière équation, et ce malgré son âge très avancé : il est déjà une<br />
antiquité. Cet exploit vaut bien la transgression du rationnel par l‟animiste.<br />
Enfin, des remerciements inconditionnels à mon père pour son soutien et<br />
ses encouragements dans les moments délicats. J‟espère le rendre fier aujourd‟hui.<br />
Cette plage de remerciements ne peut s‟achever sans une pensée<br />
particulière pour mes amis qui ont accepté d‟être les souffre-douleurs de ce<br />
doctorat. Leur présence, leur intelligence, leur assistance, leur financement de la<br />
recherche, leurs relectures, leurs critiques contestables comme fondées, leurs<br />
protestations, leurs contradictions, leurs sourires et leurs larmes sont les<br />
fondements de ce que je suis. Ils savent pour sûr que je leur dois tant, ce qui<br />
m‟évitera de languir dans la mièvrerie. Par ordre alphabétique, la ligue des héros<br />
ordinaires qui, chacun à sa manière, ont sauvé mon monde : Karim Amyuni, Julie<br />
Aubry, Romain Aubry, Benjamin Baeckeroot, Anne Boring, Nicolas Coutel, Jasna<br />
<strong>Dragicevic</strong>, Claire Floride, Sophie Guebsi, Frank Helerard, Olivier Jay, Sophie<br />
<strong>La</strong>bbé, Pierre-Antoine <strong>La</strong>loë, Pierre-Yves <strong>La</strong>nfrey, <strong>La</strong>ëtitia Lefouin, Séverine<br />
Michelot, <strong>La</strong>urie Monné-Dao, Mathieu Monnoir, Jacques Rotrou, Karine Rubiol,<br />
Vincent Ruer, Nicolas Simon, Carole Van Honacker et Sergio Visinoni.
Table des Matières<br />
Chapitre 0 ........................................................ 11<br />
Introduction Générale<br />
0.1. Le préambule .................................................................................................. 12<br />
0.2. L‟approche économique .................................................................................. 13<br />
0.3. Les méthodes d‟élicitation .............................................................................. 17<br />
0.4. Les enchères expérimentales ........................................................................... 18<br />
0.5. Le résumé de la thèse ...................................................................................... 20<br />
0.6. Les recommandations de politique publique .................................................. 23<br />
0.7. Références ....................................................................................................... 23<br />
Chapter 1 ......................................................... 27<br />
Imperfect Substitutability in Standard<br />
and Reference-Dependence Models<br />
1.1. Introduction ..................................................................................................... 28<br />
1.2. The standard model ......................................................................................... 29<br />
1.3. The substitution effect..................................................................................... 38<br />
1.4. Imperfect substitutability and the endowment effect ...................................... 40<br />
1.5. Imperfect substitutability and loss aversion .................................................... 44<br />
1.6. Imperfect substitutability and boundedness .................................................... 47<br />
1.7. Concluding remarks ........................................................................................ 51<br />
1.8. References ....................................................................................................... 52<br />
1.9. Appendix ......................................................................................................... 55
Chapter 2 ........................................................ 61<br />
Private Valuation of a Public Good<br />
in Three Auction Mechanisms<br />
2.1. Introduction .................................................................................................... 62<br />
2.2. The experimental design ................................................................................ 66<br />
2.3. The results ...................................................................................................... 69<br />
2.4. Discussion ...................................................................................................... 75<br />
2.5. Concluding remarks ....................................................................................... 78<br />
2.6. References ...................................................................................................... 79<br />
2.7. Appendix ........................................................................................................ 83<br />
Chapter 3 ........................................................ 89<br />
Endogenous Market-Clearing Prices<br />
and Reference Point Adaptation<br />
3.1. Introduction .................................................................................................... 90<br />
3.2. Auctions and incentive-compatibility ............................................................ 93<br />
3.3. Interactive incentive-compatibility ................................................................ 98<br />
3.4. The behavioral model ................................................................................... 103<br />
3.5. The empirical study ...................................................................................... 113<br />
3.6. Concluding remarks ..................................................................................... 121<br />
3.7. References .................................................................................................... 122<br />
3.8. Appendix ...................................................................................................... 127<br />
Chapter 4 ...................................................... 133<br />
Competitive Private Supply of Public Goods<br />
4.1. Introduction .................................................................................................. 134<br />
4.2. The public good game .................................................................................. 136<br />
4.3. The explicit logarithmic model .................................................................... 143<br />
4.4. Concluding remarks ..................................................................................... 153<br />
4.5. References .................................................................................................... 154<br />
4.6. Appendix ...................................................................................................... 156
Table des Figures<br />
Fig. 1.1. A change in q and imperfect substitution with the x ‟s .............................. 39<br />
Fig. 1.2. Reference-dependent preferences ................................................................. 43<br />
Fig. 1.3. Loss aversion in welfare mea<strong>sur</strong>es ............................................................... 44<br />
Fig. 1.4. Unboundedness of the compensation demanded .......................................... 48<br />
Fig. 1.5. Comparison between reference-dependent indifference curves ................... 51<br />
Fig. 2.1. WTA / WTP disparity from trial 1 to trial 10 ............................................. 72<br />
Fig. 2.2. Exponential regression of WTA / WTP disparity ....................................... 73<br />
Fig. 2.3. Exponential regression of WTA / WTP disparity ....................................... 73<br />
Fig. 3.1. The sequential price weighting function ..................................................... 111<br />
Fig. 4.1. Agents i‟s and j‟s best-response functions .................................................. 147<br />
Fig. 4.2. Income transfer with strategic substitutes ................................................... 151<br />
Fig. 4.3. Income transfer with strategic complements .............................................. 151<br />
Fig. 4.4. The aggregate level of provisions ............................................................... 152
Table des Tableaux<br />
Table 1.1. Four welfare indices ................................................................................... 34<br />
Table 1.2. Welfare indices and context-dependence ................................................... 41<br />
Table 1.3. Welfare indices in a gain and loss perspective ........................................... 45<br />
Table 2.1. Summary statistics of the BDM, SPA and NPA mechanisms ................... 70<br />
Table 2.2. Exponential regression statistics ................................................................ 74<br />
Table 3.1. Unitary sequential weight coefficients ..................................................... 114<br />
Table 3.2. Summary statistics of the uniform and s-shaped theoretical estimates .... 115<br />
Table 3.3. � -factors statistics ................................................................................... 118<br />
Table 3.4. Comparison between extra expected and real winners from deviation .... 120<br />
Table 3.5. Comparison between extra expected and real gains from deviation ........ 121
11<br />
Chapitre 0<br />
Introduction Générale
0.1. Le préambule<br />
12<br />
"<strong>La</strong> nature n‟est ni morale ni immorale,<br />
elle est radieusement, glorieusement,<br />
amorale." Théodore Monod<br />
L‟évaluation économique des biens et services publics environnementaux<br />
répond à un double objectif : en premier lieu, produire un ordre de grandeur, en<br />
des termes monétaires, des services rendus par l‟environnement afin qu‟ils soient<br />
incorporés dans les décisions publiques à leur juste valeur ; en second lieu,<br />
apporter des éléments qui permettent de bâtir des politiques de l‟environnement<br />
tout en prenant en compte les préférences des agents économiques.<br />
<strong>La</strong> thèse publique en évaluation économique développée dans ce manuscrit<br />
se compose de quatre essais. Le premier interroge la nature des préférences des<br />
agents économiques pour les biens publics <strong>sur</strong> un marché hypothétique. Le<br />
deuxième examine le bien-fondé des mécanismes d‟enchères pour révéler les<br />
préférences environnementales. Le troisième considère la question de la sincérité<br />
des valeurs révélées en enchères répétées. Enfin, le quatrième appréhende ce qui<br />
motive les agents à financer un bien public, le financement et la valeur qu‟ils<br />
attribuent au bien publique étant des corrélatives, en dépit de l‟intérêt rationnel à<br />
se comporter en passager clandestin.<br />
<strong>La</strong> démarche scientifique transdisciplinaire qui consiste à mettre des<br />
concepts d‟horizons divers en relation les uns avec les autres – démarche que nous<br />
nous sommes efforcés d‟entreprendre tout au long de cette recherche – apporte des<br />
propositions à des questions soulevées en économie de l‟environnement et plus<br />
généralement celle des biens publics. Les essais n‟édifient pas de lois de la nature<br />
(quitte à y divertir d‟éventuels détracteurs des sciences économiques) et ouvrent<br />
autant de débats qu‟ils n‟en closent. Toutefois, nous espérons qu‟ils donnent une<br />
plus grande compréhension du comportement individuel vis-à-vis d‟un bien public
en contexte d‟échange marchand, et portent à la connaissance ce qui constitue la<br />
valeur économique de l‟environnement 1 .<br />
0.2. L’approche économique<br />
L‟environnement et les ressources naturelles fournissent aux agents des<br />
services essentiels tous les jours. Les pouvoirs publics ont nécessité de les évaluer<br />
pour budgéter les politiques environnementales. Si l‟environnement a une valeur,<br />
il n‟a pas de prix. Dès lors, comment justifier les montants d‟investissements<br />
inhérents à sa gestion ainsi que les dépenses pour la mise à disposition des biens<br />
publics ? L‟analyse économique permet de comparer les coûts et les bénéfices<br />
d‟actions envers l‟environnement, ce qui en fait un outil de décision robuste pour<br />
évaluer les politiques et mieux légiférer. Appliquée à l‟environnement, l‟approche<br />
économique se divise en régulation et évaluation. Elle observe et modélise les<br />
préférences des agents (eux-mêmes présumés conscients de leurs préférences) par<br />
rapport à leur cadre de vie, le milieu naturel dans le cas présent.<br />
<strong>La</strong> régulation représente l‟ensemble des règles qui ont pour but de<br />
maintenir l‟équilibre du marché. L‟absence de marché des biens publics implique<br />
l‟intervention de l‟État. <strong>La</strong> régulation devient alors la mise en place de règles de<br />
conduite qui permettent de maximiser le bien-être social. Les politiques s‟appuient<br />
généralement <strong>sur</strong> la régulation, à travers la taxation des pollueurs imaginée par<br />
Pigou en 1929 ainsi que les compensations monétaires fixées par le droit commun.<br />
Cependant, l‟absence de marché induit l‟absence de prix, lequel est un vecteur<br />
d‟information <strong>sur</strong> la valeur du bien. Il en résulte distorsions de valeur, coûts de<br />
transactions et asymétries d‟information très coûteuses en efficacité. En effet, peu<br />
de politiques environnementales se basent <strong>sur</strong> le critère d‟efficacité, notamment<br />
parce que les décideurs publics ont d‟autres objectifs que l‟efficacité économique,<br />
tels que l‟équité ou bien la soutenabilité des systèmes de ressources (Freeman<br />
1 S‟agissant d‟une thèse publique, malgré des efforts de vulgarisation, les quatre chapitres qui<br />
composent cette thèse comportent quelques passages techniques difficiles. Nous sollicitons<br />
l‟indulgence du lecteur intéressé mais non-initié.<br />
13
2003). Pourtant, l‟analyse économique se propose justement d‟éclairer le décideur<br />
<strong>sur</strong> les critères a minima lesquels permettent le développement durable ; et<br />
développement durable ne signifie pas développement souhaitable (Sinclair-<br />
Desgagné 2007a) qui relève d‟autres grilles de lecture sociétales.<br />
L‟évaluation économique consiste à montrer que l‟environnement a une<br />
valeur d‟usage. Préserver cet usage intact revient à s‟exprimer <strong>sur</strong> des projets qui<br />
impactent, positivement et négativement, le niveau de qualité environnementale,<br />
puis à arbitrer entre coûts et bénéfices. Il s‟agit de l‟analyse coût-bénéfice, basée<br />
<strong>sur</strong> la prise en compte des équivalents monétaires que les individus considèrent<br />
pertinents pour refléter leurs préférences (Gatzweiler et Volkmann 2007). Par<br />
exemple, cela signifie que les individus sont capables d‟associer des valeurs<br />
monétaires à des niveaux de préservation d‟un milieu naturel. Cette capacité<br />
d‟association est la pierre angulaire de l‟analyse économique <strong>sur</strong> les questions<br />
environnementales. Sans celle-ci, il apparaît impossible d‟appliquer des principes<br />
économiques développés en théorie du bien-être. L‟environnement naturel a donc<br />
une valeur économique, mais il n‟y a toujours pas de consensus <strong>sur</strong> la nature de<br />
cette valeur ou <strong>sur</strong> les meilleurs outils pour la me<strong>sur</strong>er.<br />
D‟un côté, les économistes néoclassiques lient la valeur d‟un bien ou<br />
service à l‟utilité, ou la satisfaction des préférences, qu‟il procure. Selon ce mode<br />
de pensée qu‟on peut définir comme anthropocentrique, l‟environnement a une<br />
valeur instrumentale, laquelle dépend des préférences des agents qui le<br />
considèrent comme un moyen et non comme une fin en soi (même un parc naturel<br />
est un moyen qui rend possibles la contemplation de la vie sauvage et la<br />
randonnée en milieu naturel). En effet, le socle de l‟analyse coût-bénéfice repose<br />
<strong>sur</strong> la logique instrumentale. <strong>La</strong> somme qu‟un individu est disposé à dépenser<br />
pour satisfaire ses préférences reflète la valeur qu‟il accorde au bien. Il est donc<br />
possible de révéler la valeur du bien à travers sa demande (Bateman et al. 2002).<br />
Les économistes calculent ensuite le taux auquel un agent est prêt à substituer ce<br />
bien pour un autre (en l‟occurrence, cet autre bien est le numéraire dans lequel<br />
sont me<strong>sur</strong>és les prix). Ce taux est capté par les indices de consentement-à-payer<br />
maximal (le CAP) et de consentement-à-recevoir minimal (le CAR). Les valeurs<br />
14
économiques sont d‟ordinaire révélées dans le cadre d‟une institution fondée <strong>sur</strong><br />
l‟échange. Le principe est que tous les agents possèdent la même quantité<br />
d‟informations <strong>sur</strong> le bien à valoriser, soit l‟absence d‟asymétrie d‟information.<br />
De l‟autre côté, les environnementalistes accordent au milieu naturel une<br />
valeur de non-usage, c‟est-à-dire une valeur intrinsèque ou per se. Or la valeur de<br />
non-usage est indépendante des prix du marché, si bien qu‟elle ne peut pas être<br />
approximée autrement que par l‟évaluation hors-marché. Sachant que les<br />
différentes natures de la valeur sont imbriquées dans ce qui serait la vraie valeur<br />
d‟un bien ou service, O‟Neill (1993) considère simplificateur d‟utiliser un outil<br />
d‟évaluation basé <strong>sur</strong> la commen<strong>sur</strong>abilité et la représentation monétaire.<br />
Diamond et Hausman (1994) affirment même que les agents n‟ont pas de<br />
préférences dites environnementales. Toutefois, les individus sont d‟expérience<br />
disposés à payer ou recevoir une valeur monétaire pour un bien ou service<br />
environnemental, prouvant ainsi qu‟ils sont prêts à substituer des biens entre eux,<br />
et donc à rendre comparables des biens privés avec des biens publics. Si la<br />
conversion monétaire était irrecevable, son refus serait observable quel que soit le<br />
contexte, ce qui n‟est pas le cas. C‟est pourquoi ont été introduits les marchés<br />
hypothétiques tels que l‟évaluation contingente initiée par Ciriacy-Wantrup<br />
(1947).<br />
L‟autre problème concerne la nature publique des biens et services<br />
environnementaux. En effet, ils sont des biens publics, donc par définition non<br />
exclusifs, c‟est-à-dire qu‟aucun agent ne peut être exclu de leur consommation, et<br />
non rivaux, à savoir que l‟usage d‟un agent n‟entrave pas celle d‟un autre agent.<br />
Comme les biens publics ne s‟échangent pas <strong>sur</strong> un marché, il en résulte absence<br />
du taux de substitution et du prix d‟échange. Néanmoins, grâce aux marchés<br />
hypothétiques, l‟agrégation des valeurs privées permet de construire une courbe<br />
de demande pour le bien public. Il est donc raisonnablement possible de baser les<br />
politiques environnementales <strong>sur</strong> les évaluations privées issues des enquêtes<br />
montées à cet effet.<br />
Il est argué que les problèmes liés à l‟environnement sont dus à l‟absence<br />
de définition adéquate des droits de propriété. Le prétexte juridique a souvent<br />
15
déplacé le débat des biens publics en dehors du sentier économique, par le fait que<br />
le CAP place l‟agent en position d‟acquéreur tandis que le CAR place l‟agent en<br />
situation de propriétaire ; alors que le bien public correspond au statut<br />
intermédiaire de copropriété. De fait ou par défaut, les normes juridiques sont<br />
devenues l‟instrument utilisé par les pouvoirs publics. Pourtant, les autorités<br />
régulatrices pourraient rétablir la logique du marché en régulant via des prix.<br />
L‟idée que l‟évaluation économique ne peut pas résoudre les questions de biens<br />
publics en raison de la logique marchande – qui serait inapte à traduire leur valeur<br />
sociale –, et que seul l‟aménagement juridique des droits individuels à l‟usage des<br />
biens publics en est capable, est sans véritable fondement. D‟abord, si les prix<br />
sont incomplets, comme le montrent les externalités négatives souvent citées en<br />
exemple pour justifier l‟échec des marchés, la juridiction l‟est autant. Créer des<br />
marchés hypothétiques pour l‟environnement, c‟est ni plus ni moins prendre en<br />
compte ces externalités, et la <strong>sur</strong>veillance des parties prenantes peut se substituer à<br />
l‟autorité publique. Ensuite, la mise en place d‟un arsenal juridique est onéreuse,<br />
et il appartient aux autorités régulatrices de minimiser les coûts d‟administration,<br />
parce que d‟autres politiques publiques peuvent être initiées et rétribuées par la<br />
réalisation de ces économies.<br />
L‟ère est à la rationalisation des dépenses publiques qui ont trop longtemps<br />
manqué dans les finances publiques, entraînant des gaspillages dont les coûts sont<br />
supportés par la <strong>société</strong> civile. Ainsi, Montgomery (1972) a démontré que le coût<br />
d‟implémentation d‟une politique environnementale par les instruments de marché<br />
tels que les droits d‟émission était minimisé à l‟équilibre. Également, d‟après<br />
Sinclair-Desgagné (2007b), "il incombe à l‟État de veiller au bon fonctionnement<br />
du mécanisme des prix [e]n réduisant le nombre de biens collectifs par<br />
l‟instauration de conditions propices à la naissance et au fonctionnement de<br />
marchés efficaces." Rappelons qu‟en situation de copropriété, de nombreuses<br />
décisions sont prises à la majorité, évitant le piège de l‟unanimité qui ne peut<br />
exister en analyse économique compte tenu de l‟hétérogénéité des préférences.<br />
Enfin, la démarche qui consiste à aller directement interroger les citoyens <strong>sur</strong> les<br />
questions environnementales n‟est-elle pas la plus démocratique qui soit ?<br />
16
0.3. Les méthodes d’élicitation<br />
<strong>La</strong> méthode des préférences révélées déduit la valeur de l‟environnement à<br />
partir des décisions prises par les agents économiques. Son ambition est<br />
d‟observer le comportement effectif de l‟agent, sensé traduire ses préférences et la<br />
valeur qu‟il accorde à l‟environnement. Cette méthode utilise les données du<br />
marché existantes pour extraire la valeur implicite d‟un bien. De la sorte,<br />
Hotelling (1949) a proposé la méthode indirecte des coûts de transport pour<br />
évaluer la demande pour les loisirs dans les milieux naturels. Cependant, les<br />
préférences révélées ne fonctionnent que si on dispose de données du marché.<br />
Il est souvent difficile d‟obtenir des données du marché relatives aux<br />
questions environnementales, aussi une part importante des études repose-t-elle<br />
<strong>sur</strong> les préférences déclarées, à l‟égal de l‟évaluation contingente. L‟évaluation<br />
contingente prend la forme d‟une enquête d‟opinion dans laquelle on demande<br />
aux individus de déclarer combien ils sont disposés à payer pour éviter une<br />
dégradation de l‟environnement ou bien combien ils sont disposés à recevoir en<br />
compensation pour laisser faire cette dégradation. Les valeurs – assimilées aux<br />
prix du marché hypothétique – sont ensuite agrégées pour calculer la valeur<br />
monétaire globale. Le but de l‟évaluation contingente a d‟abord été de me<strong>sur</strong>er la<br />
disposition à payer pour as<strong>sur</strong>er la disponibilité d‟un service environnemental.<br />
Mais, la dégradation accrue de l‟environnement a fait basculer cette littérature<br />
vers des études portant <strong>sur</strong> des dommages subis par le milieu naturel (voir Carson<br />
et al. 1992).<br />
Même si la méthode permet de prendre en compte la valeur de non-usage<br />
(Walsh et al. 1984) défendue par les environnementalistes, sa limite réside dans le<br />
fait qu‟elle est source de nombreux biais : risque de questions mal formulées qui<br />
orienteraient les réponses ; mauvaise perception du bien à évaluer ; réponse<br />
stratégique plutôt que sincère ; apparition de biais cognitifs incompatibles avec la<br />
rationalité. En effet, les individus valorisent un scénario hypothétique. L‟absence<br />
des incitations du marché, qui prennent la forme des contraintes budgétaires et de<br />
mise en disponibilité des substituts, produit donc des données contestables. Par<br />
17
exemple, les agents peuvent promettre des sommes destinées à la protection de<br />
l‟environnement largement supérieures à celles qu‟ils sont réellement prêts à<br />
payer (Diamond et Hausmann 1994, Hanemann 1994, Neill et al. 1994). Rien<br />
n‟incite donc l‟individu à donner sa vraie valeur lors d‟une déclaration. Les<br />
préférences déclarées ont ainsi été accueillies avec pyrrhonisme, voire hostilité.<br />
Lorsque les agents considèrent leurs déclarations inconséquentialistes,<br />
toutes les réponses se valent. Même en vertu de la sincérité des agents (ce qui<br />
demeure hypothétique), ceux-ci n‟ont pas incitation à engager d‟efforts cognitifs<br />
importants lorsqu‟ils doivent formuler une déclaration, ce qui rend les valeurs<br />
déclarées potentiellement bruyantes ou biaisées. Dans le cas où les agents<br />
considèrent leurs déclarations conséquentialistes, ils sont incités à donner des<br />
réponses fictives, comme minimiser leurs CAP s‟ils s‟aperçoivent que le projet<br />
porte <strong>sur</strong> la création d‟une nouvelle taxe, afin d‟influencer les décideurs publics<br />
qui peuvent être dans la projection d‟une réélection et donc dans l‟opportunisme.<br />
0.4. Les enchères expérimentales<br />
Puisque les économistes doivent en tout état de cause éliciter des valeurs<br />
pour mener à bien des analyses coût-bénéfice et estimer les effets d‟une politique<br />
publique <strong>sur</strong> le bien-être des agents (Boardman et al. 2005) pourquoi ne pas<br />
utiliser les mécanismes d‟enchères ? En effet, les économistes s‟intéressent aux<br />
enchères expérimentales depuis un certain temps déjà : Bohm (1972), Brookshire<br />
et Coursey (1987), Hoffman et al. (1993), Shogren et al. (1994), Shogren et al.<br />
(2001), Rozan et al. (2004), Lusk et al. (2007). <strong>La</strong> seule méthode capable à ce<br />
jour de combiner les avantages des préférences révélées avec la possibilité de<br />
construire un marché simulé est le mécanisme de ventes aux enchères. Simuler un<br />
marché en laboratoire, c‟est créer un marché qui n‟existe pas, pour quelques<br />
heures et avec quelques individus recrutés à cette fin. Cette création temporaire<br />
n‟a pas d‟autre finalité que d‟observer le comportement des agents <strong>sur</strong> le marché,<br />
seul capable de révéler les CAP (Robin et al. 2007).<br />
18
<strong>La</strong> valeur ajoutée des enchères expérimentales réside dans le fait qu‟elles<br />
peuvent s‟appliquer à n‟importe quel type de bien non-marchand, ou évaluer les<br />
programmes sociaux enclins aux divergences d‟intérêts (Heckman 2001). Bien<br />
que les mécanismes de marché de type ventes aux enchères aient initialement été<br />
conçus pour éliciter la valeur des loteries et tester la validité de l‟utilité espérée<br />
(Becker et al. 1964), ils ont depuis été largement repris pour des biens réels,<br />
notamment la protection de l‟environnement (Cummings et al. 1986).<br />
Les enchères expérimentales mettent les individus en situation d‟échange<br />
actif. Quand bien même ils prendraient en compte les données du marché et<br />
réviseraient leurs préférences en fonction de celles-ci, la compatibilité avec les<br />
incitations des mécanismes d‟enchères induit un coût désincitatif à dévier des<br />
préférences sincères ; rappelons que toutes les conséquences monétaires issues des<br />
décisions sont réelles. Par ailleurs, les chercheurs peuvent y observer la manière<br />
dont les agents réagissent aux signaux publics tels que les prix de compensation.<br />
Ils ont à disposition des données directes – par opposition aux données indirectes<br />
à l‟exemple des coûts de transport – afin de révéler la valeur économique d‟un<br />
bien. Les problématiques résolues par des expériences d‟évaluation sont<br />
nombreuses (Willinger 2001) mais nous nous contenterons de citer la différence<br />
entre le CAP et le CAR (Knetsch et Sinden 1984, Brookshire et Coursey 1987,<br />
Shogren et al. 1994, Shogren et al. 2001, Horowitz et McConnell 2003) ou encore<br />
l‟effet de dotation (Samuelson et Zeckhauser 1988, Kahneman et al. 1990,<br />
Horowitz et al. 2005, Bischoff 2008).<br />
Néanmoins, la validité externe des données de laboratoire est souvent<br />
remise en question. On accuse les expériences de simplisme ; on leur reproche<br />
l‟effet de contexte éloigné de la réalité, c‟est-à-dire un manque de reproduction<br />
fidèle des comportements des individus, comme dans une épicerie par exemple.<br />
Pour autant, le décideur public doit s‟accommoder de l‟absence du marché de<br />
référence. Il est inutile d‟essayer de répliquer le marché réel en laboratoire, car la<br />
simplicité permet d‟isoler de nombreux paramètres noyés dans la complexité du<br />
monde réel, ce qui améliore le contrôle de l‟étude (Friedman et Sunder 1994). En<br />
effet, le marché simulé en laboratoire permet de contrôler les variables<br />
19
décisionnelles qui pèsent <strong>sur</strong> le CAP et d‟étudier l‟impact d‟une variation à la<br />
marge de l‟une de ces variables décisionnelles, toutes choses étant égales par<br />
ailleurs (Robin et al. 2007). L‟expérimentation en laboratoire doit donc être jugée<br />
<strong>sur</strong> la qualité de la compréhension des préférences qu‟elle produit, non <strong>sur</strong> la<br />
qualité du facsimilé.<br />
0.5. Le résumé de la thèse<br />
Après ce bref chapitre introductif, nous aborderons dans un premier<br />
chapitre la question de l‟équivalence entre le CAP et le CAR. <strong>La</strong> disparité entre<br />
les deux indices a de profondes conséquences <strong>sur</strong> les prises de décision<br />
environnementales. Brown et Gregory (1999) mentionnent la formation des<br />
politiques de développement durable et l‟allocation des droits. Tout autant, on<br />
peut se demander comment baser les décisions publiques si les valeurs sont<br />
qualifiées d‟inconsistantes par rapport au choix rationnel ? Si la disparité était au<br />
départ associée aux carences de la méthode de mise en œuvre des enquêtes, les<br />
racines du problème s‟avèrent être sensiblement plus profondes. Eu égard à<br />
l‟évaluation des biens publics, nous pensons que la disparité est due à la<br />
substituabilité imparfaite entre les biens privés et publiques, ainsi qu‟en raison de<br />
perceptions différenciées des agents économiques entre gains et pertes. C‟est à<br />
cette problématique que le premier chapitre se consacre.<br />
Ainsi, le Chapitre 1 traite de la disparité entre les indices CAP et CAR<br />
dans l‟évaluation hors-marché. Dans la littérature, l‟effet de substitution et l‟effet<br />
de dotation sont tenus responsables de l‟existence des disparités. Nous montrons<br />
que la substituabilité imparfaite dans la fonction d‟utilité indirecte peut provoquer<br />
la disparité soit entre le CAP et le CAR – en raison du coût d‟opportunité –, soit<br />
entre les gains et les pertes, où il s‟agit d‟évaluer une perte sèche. <strong>La</strong> me<strong>sur</strong>e en<br />
termes relatifs accentue la substituabilité imparfaite, mais l‟effet de substitution<br />
est borné dans le modèle d‟aversion aux pertes.<br />
Ce premier chapitre prépare le terrain pour le Chapitre 2, où nous évaluons<br />
un vrai bien public dans un contexte d‟enchères expérimentales. Les offres d‟achat<br />
20
et de vente reflètent le CAP et le CAR, d‟où leur importance. L‟effet de dotation<br />
et le choix du meilleur mécanisme d‟enchères y sont examinés. Les études en<br />
enchères expérimentales jusqu‟ici menées ont porté <strong>sur</strong> des biens privés non<br />
marchands ; elles sont supposées divulguer ce qui se passerait en présence de<br />
biens publics, car il est a priori difficile d‟envisager une expérience où le bien<br />
public est échangé (Robin et al. 2007). Nous y parvenons. Nous n‟employons pas<br />
de valeurs induites mais laissons libre cours aux valeurs autoproduites par les<br />
sujets d‟étude recrutés pour l‟occasion. L‟étude nous permet de vérifier si, <strong>sur</strong> des<br />
marchés simulés, bien privé non marchand et bien public sont évalués de manière<br />
identique.<br />
Ainsi, nous évaluons l‟impact de trois mécanismes d‟enchère – le<br />
mécanisme Becker-DeGroot-Marschak (BDM), l‟enchère au deuxième prix, et<br />
l‟enchère aléatoire au nième prix – dans l‟évaluation des CAP et CAR privés d‟un<br />
bien public pur. Nos résultats montrent que l‟effet de dotation peut être éliminé en<br />
répétant le mécanisme BDM. Néanmoins, à l‟échelle logarithmique, l‟enchère<br />
aléatoire au nième prix donne la vitesse de convergence vers l‟égalité des indices<br />
de bien-être la plus élevée. Plus généralement, nous observons que les sujets<br />
d‟étude évaluent les biens publics en se référant à l‟avantage privé et subjectif qui<br />
résulte du financement du bien public.<br />
Par la suite, le Chapitre 3 discute de la sincérité des préférences en<br />
enchères expérimentales répétées et traite des propriétés incitatives des<br />
mécanismes BDM et l‟enchère aléatoire au nième prix. Une propriété des<br />
mécanismes d‟enchères est la compatibilité avec les incitations, dans laquelle un<br />
offreur a une stratégie faiblement dominante de soumettre une offre égale à sa<br />
valeur. Il a été prouvé que les deux mécanismes sont compatibles avec les<br />
incitations. En évaluation, on répète des sessions d‟enchères pour donner aux<br />
offreurs l‟opportunité d‟apprendre le mécanisme de marché : leur donner du temps<br />
pour révéler leurs préférences. Or, ce procédé les contre-incite à adapter leurs<br />
préférences en fonction des prix publiquement signalés, si bien qu‟il crée un<br />
risque de licitation stratégique (par opposition aux offres sincères). Si les offreurs<br />
s‟engagent dans des stratégies déviantes pour faire face à l‟incertitude <strong>sur</strong> la<br />
21
valeur du bien public, les mécanismes d‟enchères perdent leur propriété de<br />
compatibilité avec les incitations et révèlent de fausses préférences.<br />
Lorsque les prix dépendent des offres soumises, c‟est-à-dire en présence de<br />
mécanismes de marché répétés avec prix de compensation endogènes, l‟hypothèse<br />
de l‟indépendance des valeurs privées – sous-jacente à la compatibilité avec les<br />
incitations – est remise en question ; même si ce type de mécanismes fournit une<br />
participation active et un apprentissage du marché. Dans sa vision orthodoxe, le<br />
comportement marchand d‟adaptation met en péril la compatibilité avec les<br />
incitations. Nous introduisons un modèle qui montre que les enchérisseurs licitent<br />
suivant l‟heuristique d‟ancrage et d‟ajustement, dépendante d‟une fonction de<br />
pondération séquentielle, laquelle prend en compte les contraintes de compatibilité<br />
avec les incitations sans rejeter les prix signalés issus des autres offres. En déviant<br />
de leur ancrage dans le sens du signal public, les enchérisseurs opèrent dans un<br />
équilibre corrélé.<br />
En dernier lieu, Vatn (2005) estime que les préférences environnementales<br />
dépendent des normes sociales intériorisées : elles sont socialement contingentes.<br />
Comme le prouve l‟expérience du Chapitre 2, les contributions privées aux biens<br />
publics sont issues d‟une démarche d‟évaluation. Elles sont conduites aussi bien<br />
par des incitations asociales que sociales. Si l‟offre privée du bien public est<br />
stimulée à la fois par une rationalité qui dicte de ne pas contribuer au bien public<br />
et de profiter de l‟effort fourni par la collectivité, et par l‟appétit pour la<br />
reconnaissance sociale qui incite à se faire publiquement connaître en tant que<br />
généreux donateur, laquelle des deux motivations domine ?<br />
Le Chapitre 4 fait ainsi la comparaison entre déculpabilisation et<br />
compétition pour le statut social dans la provision privée des biens publics.<br />
Lorsque les agents sont intrinsèquement impulsés, c‟est-à-dire qu‟ils contribuent<br />
essentiellement aux biens publics dans le but de soulager leur culpabilité d‟avoir<br />
indirectement participé à leur dégradation, ils tendent à se comporter en passagers<br />
clandestins. En revanche, lorsque les agents sont extrinsèquement impulsés et se<br />
mettent en compétition pour atteindre du statut social qu‟ils visent par le<br />
financement des biens publics à titre privé, leurs contributions deviennent des<br />
22
compléments stratégiques. Dans ce cas, le niveau agrégé des biens publics croît<br />
avec la réduction des écarts de revenus entre les agents. Injecter de la compétition<br />
pour le statut social dans des fonctions d‟utilité augmente les contributions aux<br />
biens publics, et donc leur niveau global, faisant de la concurrence une incitation<br />
féconde pour résoudre le problème du passager clandestin.<br />
0.6. Les recommandations de politique publique<br />
Quatre recommandations découlent de ce travail de recherche, à savoir que<br />
nous suggérons de : (1) conduire des expériences de marchés simulés et répéter<br />
des sessions de marché pour évaluer les préférences environnementales ; évaluer<br />
à la fois les deux indices de bien-être ; (2) privilégier les mécanismes d’enchères<br />
tels que BDM et l’enchère aléatoire au nième prix, pour la raison qu’ils sont<br />
capables de réduire, voire supprimer, l’écart initial entre les indices en sessions<br />
répétées ; si l’écart persiste, considérer les valeurs comme une fourchette révélée<br />
par l’ensemble des individus ; (3) tolérer l’influence des prix de compensation<br />
signalés <strong>sur</strong> la licitation, celle-ci révélant la rationalité limitée des individus<br />
plutôt que leur imposture ; (4) inciter à la provision privée des biens publics, et<br />
encourager ce type d’actions par leur mise en valeur sociale, tout en s’as<strong>sur</strong>ant<br />
de transferts de revenu des agents économiques à haut revenu vers des agents<br />
économiques à bas revenu, afin que la compétition accroît le niveau des biens<br />
publics.<br />
0.7. Références<br />
Bateman, I., Carson, R.T., Day, B., Hanemann, M., Hanley, N., Hett, T., Jones-<br />
Lee, M., Loomes, G., Mourato, S., Ozdemiroglu, E., Pearce, D.W., Sugden,<br />
R., et Swanson, J. (ed.) (2002), Economic Valuation with Stated Preference<br />
Techniques: A Manual, Edward Elgar, Cheltenham.<br />
Becker, G., DeGroot, M., et Marschak, J. (1964), “Mea<strong>sur</strong>ing Utility by a Single<br />
Response Sequential Method”, Behavioral Science, 9: 226-232.<br />
23
Bischoff, I. (2008), “Endowment effect theory, prediction bias and publicly<br />
provided goods: an experimental study”, Environmental and Resource<br />
Economics, 39: 283–296<br />
Boardman, A., Greenberg, A., et Weimer, D. (2005), “Cost Benefit Analysis:<br />
Concepts and Practice”, 3rd ed. Prentice Hall.<br />
Bohm, P. (1972) “Estimating Demand for Public Goods: An Experiment”,<br />
European Economic Review, 3(2), pp. 111-30.<br />
Brookshire, D., et Coursey, D. (1987) "Mea<strong>sur</strong>ing the Value of a Public Good: An<br />
Empirical Comparison of Elicitation Procedures." American Economic<br />
Review, 77(4), pp. 554-66.<br />
Brown, T., et Gregory, R. (1999), “Why the WTP–WTA disparity matters”,<br />
Ecological Economics, 28: 323–335.<br />
Carson, R., Mitchell, R., Hanemann, W., Kopp, R., Presser, S., et Ruud, P. (1992),<br />
“A Contingent Valuation Study of Lost Passive Use Values Resulting From<br />
the Exxon Valdez Oil Spill”, Unpublished.<br />
Ciriacy-Wantrup S. (1947), “Capital returns from soil conservation practices”,<br />
Journal of Farm Economics, 29: 1181–1186.<br />
Cummings, R., Brookshire, D., et Schulze, W., editors (1986), Valuing<br />
Environmental Goods - An Assessment of the Contingent Valuation Method,<br />
Rowman and Allanheld, Totowa, New Jersey.<br />
Diamond, P., et Hausman, J. (1994), “Contingent valuation: is some number better<br />
than no number?”, Journal of Economic Perspectives, 8: 45– 64.<br />
Gatzweiler, F., et Volkmann, J., (2007), “Beyond Economic Efficiency in<br />
Biodiversity Conservation”, ICAR Discussion Papers 1807, Humboldt<br />
University Berlin.<br />
Freeman, A. (2003), “The Mea<strong>sur</strong>ement of Environmental and Resource Values:<br />
Theory and Methods”, Edition 2, Resources for the Future (Washington, D.C.)<br />
Friedman, D. et Sunder, S. (1994) “Experimental Methods: A Primer for<br />
Economists”, Cambridge, Cambridge University Press.<br />
Hanemann, M. (1994), “Valuing the Environment Through Contigent Valuation”,<br />
Journal of Economic Perspectives, 8: 19–43.<br />
Heckman, J. (2001), “Accounting for Heterogeneity, Diversity, and General<br />
Equilibrium in Evaluating Social Programs”, Economic Journal, 111: 654–<br />
699.<br />
24
Hoffman, E., Menkhaus, D., Chakravarti, D., Field, R., et Whipple, G. (1993),<br />
“Using <strong>La</strong>boratory Experimental Auctions in Marketing Research: A Case<br />
Study of New Packaging for Fresh Beef”, Marketing Science, 12: 318-338.<br />
Horowitz, J., et McConnell, K. (2003) “Willingness to Accept, Willingness to Pay<br />
and the Income Effect”, Journal of Economic Behavior and Organization, 51:<br />
537-545.<br />
Lusk, J., Alexander, C., et Rousu, M. (2007), “Designing Experimental Auctions<br />
For Marketing Research: Effect Of Values, Distributions, And Mechanisms<br />
On Incentives For Truthful Bidding”, Review of Marketing Science, 5: 1-30<br />
Montgomery, D. (1972), “Markets in Licenses and Efficient Pollution Control<br />
Programs”, Journal of Economic Theory, 5: 395–418.<br />
Neill, H., Cummings, R., Ganderton, P., Harrison, G., et McGuckin, T. (1994),<br />
“Hypothetical Surveys and Real Economic Commitments”, <strong>La</strong>nd Economics,<br />
70: 145–154.<br />
O‟Neill, J. (1993), “Ecology, Policy and Politics”, Human Well-being and the<br />
Natural World, Routledge, London.<br />
Robin, S., Rozan, A., et Ruffieux, B. (2007), “Me<strong>sur</strong>er les préférences du<br />
consommateur pour orienter les décisions des pouvoirs publics : l‟apport de la<br />
méthode expérimentale”, Working paper, GATE WP 07-23.<br />
Rozan, A., Stenger, A., et Willinger, M. (2004), “Willingness-to-Pay for Food<br />
Safety: an Experimental Investigation of Quality Certification on Bidding<br />
Behaviour”, European Review of Agricultural Economics, 31: 409-425.<br />
Samuelson, W., et Zeckhauser, R. (1988), “Status Quo Bias in Decision Making”,<br />
Journal of Risk and Uncertainty, 1: 7-60.<br />
Sinclair-Desgagné, B. (2007a), “Le Développement Durable: Vers un Nouvel<br />
Esprit d‟Entreprise”, Revue Internationale de Gestion, 32 : 46-53.<br />
Sinclair-Desgagné, B. (2007b), “A Propos du Coût de l‟Ecologisme”,<br />
Présentation lors du 32ème Congrès de l’ASDEQ, Disponible à l‟adresse:<br />
http://www.asdeq.org/congres/pdf/32ieme_congres/Sinclair-Desgagne-<br />
ASDEQ-2007.pdf<br />
Shogren, J., Shin, S., Hayes, D., et Kliebenstein, J. (1994), “Resolving Differences<br />
in Willingness to Pay and Willingness to Accept”, American Economic<br />
Review, 84: 255-270.<br />
25
Shogren, J., Cho, S., Koo, C., List, J., Park, C., Polo, P., et Wilhelmi, R. (2001),<br />
“Auction Mechanisms and the Mea<strong>sur</strong>ement of WTP and WTA”, Resource<br />
and Energy Economics, 23: 97-109.<br />
Vatn, A. (2005), “Rationality, institutions and environmental policy”, Ecological<br />
Economics, 55: 203-217.<br />
Walsh R., Loomis J., et Gillman R. (1984), “Valuing option, existence and<br />
bequest demands for wilderness”, <strong>La</strong>nd Economics, 60: 14–29.<br />
Willinger, M. (2001), “Environmental Quality, Health and the Value of Life”,<br />
EVE Policy Research Brief, Number 7.<br />
26
27<br />
Chapter 1<br />
Imperfect Substitutability in Standard<br />
and Reference-Dependence Models<br />
Abstract<br />
This chapter focuses on the disparity between willingness-to-pay and willingness-toaccept<br />
indices in nonmarket valuation. In the literature, the substitution effect and the<br />
endowment effect are presumed to cause the disparities. We show that imperfect<br />
substitutability in the indirect utility function can lead to disparity either between<br />
WTA and WTP – due to the opportunity loss – or between gains and losses, which<br />
reflects a net loss. Context-dependent valuation accentuates the imperfect<br />
substitutability, but the substitution effect is bounded inside the behavioral model of<br />
loss aversion.<br />
Keywords: contingent valuation, WTP-WTA disparity, substitution effect, loss<br />
aversion<br />
JEL Classification: D61, D81, Q51
1.1. Introduction<br />
"A thing does not have value because it costs,<br />
as people suppose; instead it costs because it<br />
has a value." Étienne Bonnot de Condillac<br />
The debate on how policy-makers compare the benefits derived from one<br />
public plan against another has been led by the cost-benefit analysis. In this debate,<br />
the quest for revealing nonmarket values induced the direct contingent valuation<br />
method based on the Hicksian C and E , i.e. the individual‟s maximum willingness-<br />
to-pay (or best offer) to guarantee the change and the minimum willingness-to-accept<br />
(or reservation price) to sacrifice the change.<br />
Empirical and experimental studies have given evidence of large disparity<br />
between WTP and WTA, which makes impractical the use of values estimates<br />
derived from the contingent valuation. Experimental laboratory markets confirmed<br />
persistency in disparities (Knetsch and Sinden 1984; Brookshire and Coursey 1987).<br />
To justify the disparity, theorists invoked the substitution effect or the context-<br />
dependent endowment effect, and oriented the effects in rivalry. The substitution<br />
effect results from the agent‟s imperfect trade-off between private goods and public<br />
goods. The loss aversion output, that is, the endowment effect, makes agents value<br />
losses higher than equivalent gains. Morrison (1997) asserts that the endowment<br />
effect and the substitution effect play a combined role in the disparity.<br />
To be loss averse, an agent has to consider herself an owner of the public<br />
good. In general, dealing with substitution rather than endowment allows to study the<br />
consumers‟ behavior without the constraint of the initial allocation of property rights.<br />
As Sinclair-Desgagné (2005) emphasizes, the property rights remain difficult to<br />
establish, guarantee, or to legitimate in public policies, whereas in a market, the price<br />
of a good or service signals the value of the resources; agents adjust their preferences<br />
and make necessary substitutions. We consider a gain in the environmental level as a<br />
non-essential right. In reverse, a compensation for a loss of the environmental level is<br />
an essential right that agents express by means of high valuation statements. This<br />
28
distinction explains the difference between the standard disparity and the gain and<br />
loss disparity in terms of the property rights.<br />
This chapter brings out three elements. First, through convex preferences or<br />
quasi-concave utility functions, where agents prefer mixed over extreme consumption<br />
bundles of private and public goods, we show that, akin to the standard disparity<br />
between WTP and WTA, the gain and loss disparity is prone to imperfect<br />
substitutability. Second, the nature of the disparities is different, simply because<br />
agents do not tolerate a loss in the same way they bear a foregone gain. Inside the<br />
neoclassical paradigm, the substitution effect works as an opportunity loss 2 : the lower<br />
the substitutability, the higher the opportunity loss. But the utility of an agent does not<br />
change along the indifference curve. At worst, an agent faces the status quo. On the<br />
contrary, when an agent is asked to value the loss of the public good and to weigh this<br />
loss against an equivalent gain, the opportunity loss becomes a net loss. The net loss<br />
is a critical change, for agents attach a high value to the goods or services they cannot<br />
regain. They use the status quo as a reference point to switch to a steeper indifference<br />
curve. This would be the endowment effect, or what Kahneman and Tversky refer to<br />
as loss aversion in their behavioral model. Finally, we emphasize that the substitution<br />
effect – proved to be infinite in the Hicksian context – is bounded inside the<br />
behavioral model of loss aversion.<br />
We recall the basic account of the neoclassical model in Section 1.2, we<br />
provide clarifications of the substitution effect in Section 1.3 and the endowment<br />
effect in Section 1.4. We scrutinize loss aversion through imperfect substitutability in<br />
Section 1.5, and we study boundedness within imperfect substitutability in Section<br />
1.6. Concluding comments are given in Section 1.7.<br />
1.2. The standard model<br />
According to Hicksian theory, an agent has preferences over nonnegative<br />
quantities of goods and her preference ordering is transitive, continuous,<br />
2 By analogy to finance, consider the foregone opportunity to improve the level of the public as an<br />
opportunity cost of an interest in a bank account.<br />
29
nondecreasing and convex 3 . Assume an agent has convex preferences for market<br />
private goods x and some public good such as the environmental quality q . She can<br />
vary the quantity of consumption of the x ‟s, whereas the quantity of q is taken to be<br />
fixed to her. Her preferences are quasi-concave in utility 4 of the x ‟s and represented<br />
by a continuous and nondecreasing utility function u u �x, q�<br />
30<br />
� which is twice<br />
differentiable 5 . The agent faces a budget constraint based on her income y and the<br />
prices of the private goods p . She maximizes her utility subject to a budget<br />
constraint:<br />
u� x q� � p x � y<br />
[1]<br />
max , subject to i i<br />
x<br />
According to [1], the program yields the Marshallian ordinary direct demand<br />
functions x i . Substituting them as functions of � � , py gives the indirect utility<br />
functions which represent agent‟s preference ordering. v�� � is continuous, decreasing,<br />
and quasi-convex:<br />
� , , �<br />
� for i� 1,..., n<br />
[1a]<br />
i<br />
xi h p q y<br />
� � , , �,<br />
� � , , �<br />
u h p q y q � v p q y<br />
[1b]<br />
3 The completeness of preferences implies that utility is complete. When preferences satisfy<br />
completeness and transitivity, preferences are considered to be rational. In addition, when they satisfy<br />
continuity, the utility function is continuous. At last, when preferences are monotonic, utility is<br />
nondecreasing.<br />
4 A quasi-concave utility function means that preferences are convex, that is: for all x and q , and any<br />
� � , u��x�1��q�min �u � x� , u� q��<br />
� , 0 � 1<br />
� � � . It en<strong>sur</strong>es the preference ordering.<br />
5 This assumption eliminates kinks in the indifference curves.
The agent‟s consumption of �x, q�<br />
can also be obtained through a program<br />
that minimizes her expenditures � i i�<br />
�x q�<br />
u , :<br />
� based on her utility level constraint<br />
i px<br />
� pixi u � u� x q�<br />
[2]<br />
min subject to ,<br />
x<br />
Its resolution gives the expenditure or cost function 6 or the minimum amount<br />
of income necessary to achieve an attainable utility level at least as high as u , given<br />
the price vector p :<br />
i<br />
� , , � � , , �<br />
e p q u � � p g p q u for i� 1,..., n<br />
[2a]<br />
i<br />
The expenditure function is jointly continuous in � p, q, u � , strictly increasing<br />
in u , positively linear homogenous, and concave in � pq , �.<br />
Its derivative with respect<br />
to y gives the cost-minimizing demand function or the Hicksian compensated<br />
demand function that delivers optimal quantities at various prices. Moreover, the<br />
income is compensated in such a way as to leave utility unchanged:<br />
� , , �<br />
i<br />
xi g p q u<br />
� for i� 1,..., n<br />
[2b]<br />
So far, preferences are just as well represented by both the indirect utility<br />
function and the expenditure function:<br />
� �<br />
u � v ��p, q, e p, q, u ��<br />
[3]<br />
� �<br />
y � e ��p, q, v p, q, y ��<br />
[4]<br />
6 The expenditure function is twice differentiable due to the assumption that the utility function is<br />
differentiable.<br />
31
to<br />
The standard Hicksian welfare mea<strong>sur</strong>es deal with changes in prices (from<br />
1<br />
p ) while q and y are left unchanged. These changes have an impact on the<br />
indirect utility functions. The compensating variation C is the maximum amount of<br />
income that could be taken from an agent who gains from a particular change while<br />
leaving her no worse-off than before the change. The equivalent variation E is the<br />
minimum amount that an agent who gains from a particular change would be willing<br />
to accept to forego the change after it has taken place.<br />
Definition 1.1.: C and E are implicitly and explicitly defined by:<br />
0 1<br />
0<br />
� , , � � � , , � � and C � y � e� p, q, u �<br />
v p q y v p q y C<br />
1 0<br />
1<br />
� , , � � � , , � � and � , , �<br />
v p q y v p q y E<br />
changes from<br />
E �epqu� y<br />
Now assume changes occur in the levels of the environmental quantity 7 . If q<br />
� , , �<br />
0 0<br />
u � v p q y<br />
� , , �<br />
1 1<br />
u � v p q y<br />
0<br />
q to<br />
1<br />
q , the agent‟s utility changes from<br />
32<br />
0<br />
u to<br />
These changes will also have an impact on the expenditure functions. The<br />
welfare mea<strong>sur</strong>e is the change in expenditure necessary to hold the utility constant, at<br />
the two quantity sets. We can write C and E as the difference between the minimal<br />
expenditure before the change and minimal expenditure after the change given utility<br />
levels<br />
0<br />
u and<br />
1<br />
u :<br />
7 This model does not consider irreversible environmental damages. Therefore, the individual can<br />
always increase the level of the environmental quality and then recover some utility level.<br />
1<br />
u :<br />
0<br />
p
33<br />
0 � , , �<br />
�e<br />
p q u<br />
0 1 0 0 1 0<br />
C � C �q , q , p, y� � e� p, q , u � � e� p, q , u � � �<br />
dq<br />
q �q<br />
1<br />
q<br />
� [3a]<br />
0<br />
1 � , , �<br />
�e<br />
p q u<br />
0 1 0 1 1 1<br />
E � E �q , q , p, y� � e� p, q , u � �e � p, q , u � � �<br />
dq<br />
q �q<br />
1<br />
q<br />
� [3b]<br />
0<br />
The superscripts 0 and 1 indicate the situation before and after the change. C<br />
equals the maximum amount of money an agent could give up in situation 1 without<br />
being worse-off than in situation 0. E equals the minimum amount of money an agent<br />
would require in situation 0 to attain the utility in situation 1. C and E depend on the<br />
starting and ending values of q , and the value of � py , � at which the change takes<br />
places.<br />
In terms of the indirect utility function, C and E are plugged as follows:<br />
� , , � � , , �<br />
0 0 1<br />
u v p q y v p q y C<br />
� � � [4a]<br />
� , , � � , , �<br />
1 1 0<br />
u v p q y v p q y E<br />
� � � [4b]<br />
Be it with indirect utility function or expenditure function, the concepts of<br />
WTP and WTA can be derived from the Hicksian paradigm. Depending on the<br />
direction of the change, C and E may be positive or negative. When the change<br />
improves utility or �u� 0,<br />
C is the agent‟s maximum willingness-to-pay to<br />
guarantee the improvement and E is the agent‟s minimum willingness-to-accept to<br />
forego the improvement. When the change deteriorates utility or �u� 0,<br />
� C is the<br />
agent‟s minimum willingness-to-accept to tolerate the deterioration and � E is the<br />
agent‟s maximum willingness-to-pay to avoid it. This property is obtained by<br />
reversing the initial and final levels (see Table 1.1.) Indeed, Ebert (1984) proves that<br />
the welfare mea<strong>sur</strong>es possess the property of circularity. Therefore, C and E are<br />
0 1 1 0<br />
symmetric or C �q , q � E �q , q �<br />
�� . From [4a] and [4b], we get:
0 1 1 0<br />
� , , , � �� � , , , �<br />
C q q p y E q q p y<br />
0 1 1 0<br />
� , , , � � , , , �<br />
E q q p y �� C q q p y .<br />
Table 1.1. Four welfare indices<br />
1 0<br />
1 0<br />
decline ( u �u � 0 ) growth ( u �u � 0 )<br />
�C� WTA<br />
C � WTP<br />
�E� WTP<br />
E � WTA<br />
As pioneered by Mäler (1974) and taken over by Hanemann (1991), suppose<br />
now an agent can pay for the environmental quality as if it were marketed. She thus<br />
pays for q in this hypothetical market at some implicit price � . The standard price<br />
flexibility of income can be interpreted as the income elasticity of demand for the<br />
environmental quality. We then fix the following programs:<br />
u� x q� � p x ��q� y<br />
[5]<br />
max , subject to i i<br />
x<br />
xq ,<br />
� pixi ��qu�u�xq� [6]<br />
min subject to ,<br />
function:<br />
From which we obtain the following indirect utility function and expenditure<br />
� , , � ˆ � , ˆ ��, , �, � ˆ ��,<br />
, �<br />
v p q y � v �ppquu� � p q u �q<br />
[5a]<br />
i q<br />
eˆ � p, �, u� �� p ˆ � , , � ˆ<br />
ig<br />
p � u ��g�p,<br />
�,<br />
u�<br />
[6a]<br />
The derivative of the Marshallian demand function with respect to � p, q, y �<br />
gives the indirect utility function. Inverted, it gives � , i.e. the inverse demand<br />
34
function to obtain q supplied at ˆ � �� � . In this case, the agent‟s income must be<br />
supplemented so she can both afford q and the x ‟s:<br />
� � �<br />
ˆ q<br />
x �hp, q, y � q<br />
[5b]<br />
i<br />
� p q y�<br />
� � ˆ � , ,<br />
[5c]<br />
The derivative of the expenditure function with respect to � p, q, u � gives the<br />
Hicksian compensated demand function x i . Inverted, it gives the inverse<br />
compensated demand price ˆ � �� � , that is, the price that would induce her to purchase<br />
q units if her income were increased:<br />
� � �<br />
q<br />
x � gˆ p, , u<br />
[6b]<br />
i<br />
� p q u�<br />
� � ˆ � , ,<br />
[6c]<br />
The two inverse demand functions are:<br />
� p, q, y� p, q, v� p, q, y�<br />
ˆ � � ˆ � �� ��<br />
[5'c]<br />
� p, q, u� p, q, e� p, q, u�<br />
ˆ � � ˆ � �� ��<br />
[6'c]<br />
From [5'c] and [6'c] it follows that:<br />
� p, q , u � � p, q , y�<br />
0 0 0 0<br />
� � ˆ � � ˆ �<br />
[7a]<br />
1<br />
� ˆ � p, q , u � � ˆ � p, q , y�<br />
� �<br />
1 1 1<br />
� [7b]<br />
[5] differs with [1] on ˆ � � , , �<br />
� p q u � q . The expenditure function and the<br />
compensated demand function are equal, thus the inverse compensated demand<br />
function for q becomes:<br />
35
�e<br />
e � , , � ˆ<br />
q p q u � � �� �p, q, e� p, q, u��<br />
�q<br />
� �<br />
The inverse demand ˆ � �� � mea<strong>sur</strong>es shadow or virtual prices, or marginal<br />
valuation, or marginal WTP or WTA to pay for a unit of q by the agent, i.e. the<br />
marginal rate of substitution between the x ‟s and q . As the inverse of indirect utility<br />
functions yields the expenditures functions, the inverse of direct utility functions<br />
gives indirect expenditure functions. Combining [5'c] and [6'c] with [4a] and [4b] and<br />
using Shepard‟s Lemma yields the following:<br />
� �<br />
0<br />
1 1<br />
q �e�<br />
p, q, u � q<br />
� ˆ<br />
0 � � 0 � �<br />
[9a]<br />
0 1 0<br />
C � C q , q , p, y � � dq � p, q, u dq<br />
q q<br />
� �<br />
�q<br />
1<br />
1 1<br />
q �e�<br />
p, q, u � q<br />
� ˆ<br />
0 � � 0 � �<br />
[9b]<br />
0 1 1<br />
E � E q , q , p, y � � dq � p, q, u dq<br />
q q<br />
�q<br />
Thus WTP and WTA can be expressed by way of the integral of inverse<br />
compensated demand curves for a change in quantities from<br />
36<br />
0<br />
q to<br />
[8]<br />
1<br />
q . The distinction<br />
between WTP and WTA is the level of utility the compensation is designed to reach:<br />
0<br />
u and<br />
1<br />
u respectively.<br />
Welfare mea<strong>sur</strong>es can also be defined by a distance function (Ebert 1984).<br />
The distance function is a utility function normalized by monetary income, i.e. a<br />
monotonic transformation of the direct utility function for fixed quantities:<br />
d � d( x, q, u)<br />
[10a]<br />
d �� � is continuous, decreasing in u , increasing and positively linear<br />
homogenous, and concave in x . The Shephard‟s input distance function has been<br />
introduced to consumer theory and defined in terms of the utility function (Deaton
1979). The derivatives of d �� � with respect to � , �<br />
compensated demand functions:<br />
� , , � , , ��<br />
i<br />
xi a x q d x q u<br />
37<br />
xq give a set of inverse<br />
� for i� 1,..., n<br />
[10b]<br />
i<br />
a is the normalized price of q with respect to income. The distance function<br />
can be interpreted as an indirect expenditure function. Indeed, duality results show<br />
that the expenditure (cost) function is a distance function derived from the indirect<br />
utility function (Blackorby et al. 1978). Apart from the monotonicity and definition<br />
over different arguments, the expenditure function and the distance function share the<br />
same properties. If we consider the distance function for quantity changes, it is a dual<br />
to the expenditure function for fixed quantities and can be used to examine the<br />
welfare effects of quantity changes. To recover (non-normalized) monetary mea<strong>sur</strong>es,<br />
the welfare mea<strong>sur</strong>es must be multiplied by income. Thus, C and E are defined by:<br />
� � ,<br />
0<br />
,<br />
0 � � ,<br />
1<br />
,<br />
0 ��<br />
� � ,<br />
0<br />
,<br />
1� � ,<br />
1<br />
,<br />
1��<br />
C � y d x q u � d x q u<br />
[11a]<br />
E � y d x q u � d x q u<br />
[11b]<br />
q<br />
q<br />
0<br />
1<br />
Using Shepard‟s Lemma, the latter reduces to:<br />
0<br />
� , , � , , ��<br />
C � � a x q yd x q u dq<br />
[12a]<br />
q<br />
q<br />
0<br />
1<br />
1<br />
� , , � , , ��<br />
E � � a x q yd x q u dq<br />
[12b]<br />
curves from<br />
C and E are mea<strong>sur</strong>ed by the area under the compensated inverse demand<br />
0<br />
q to<br />
1<br />
q with the old and new utility levels, respectively.<br />
Let us now compare those areas in order to see whether positive and negative<br />
changes induce the same consumer behavior.
1.3. The substitution effect<br />
When goods are available in a market at no cost, there is a regular<br />
intermediate monetary exchange of commodities, which involves a linear indifference<br />
curve for the x ‟s and q . If there is disparity, it depends on the constant price<br />
flexibility of income, i.e. the elasticity of the marginal valuation of q with respect to<br />
the x ‟s (or y that buys the x ‟s):<br />
1<br />
�C<br />
q<br />
0 0<br />
C � , , � ˆ<br />
y � �vypq<br />
y � 0 u � p, q, u �dq<br />
�y � [13a]<br />
q<br />
1<br />
�E<br />
q<br />
1 1<br />
E � , , � ˆ<br />
y � �vypq<br />
y � 0 u � p, q, u �dq<br />
�y � [13b]<br />
q<br />
2<br />
� � and if equ � e q u�<br />
If C E 0<br />
y y<br />
� � � � � 0 , E� C.<br />
Indeed, the second cross-<br />
partial derivative e qu reflects the substitution effect. A null substitution effect<br />
involves linear indifference curves and null opportunity loss. Due to perfect<br />
substitutability, agents are indifferent to the variations of the public good, because<br />
they can always adjust the level of the x ‟s to maintain their utility constant. One<br />
interpretation could be that they feel unconcerned by the changes in the level of the<br />
public good. Another interpretation could be that they are unconditionally ready to<br />
substitute the public good with some private good. The usual proposition results from<br />
the above.<br />
Proposition 1.1.: When the welfare change is induced by q , due to imperfect<br />
substitutability or low elasticity of substitution between q and the x ‟s, there is values<br />
disparity. It can be infinite in the limit.<br />
Proof: In the appendix.<br />
38
The substitution effect reflects the convex curvature of the indifference curve<br />
between q and the x ‟s and the convexity of the expenditure function in q . Ebert<br />
(1993) claims that quasi-concavity of the indirect utility function v�� � , jointly with<br />
the normality of the public good, is necessary and sufficient to obtain WTA superior<br />
to WTP. If the combinations of �xq , � lead to the same level of utility, it is in the<br />
interest of an agent to have a convex mixture of goods, for it never decreases utility.<br />
Fig. 1.1. illustrates a diminishing marginal rate of substitution between the<br />
x ‟s and q with quasi-concave utility functions. As q rises from<br />
increases from<br />
0<br />
u to<br />
39<br />
0<br />
q to<br />
1<br />
q , utility<br />
1<br />
u . Displacements of the indifference curves reflect unitary<br />
income elasticity. As can be seen, WTA � WTP . The trade-off between<br />
environmental quality and the private good turns out to be less and less attractive: the<br />
marginal utility from environmental quality upgrading is diminishing. Vice versa, it<br />
means that the marginal loss of environmental quality is increasing. In any case, the<br />
lower the elasticity of substitution between q and the x ‟s – or to be more accurate,<br />
between q and the x ‟s that y buys – the broader the disparity.<br />
WTA<br />
y<br />
WTP<br />
y<br />
0<br />
q 0<br />
Fig. 1.1. A change in q and imperfect substitution with the x ’s<br />
q 1<br />
� , , �<br />
1 1<br />
u � v p q y<br />
� , , �<br />
0 0<br />
u � v p q y<br />
q
Regarding the distance function, in presence of a normal good, the inverse<br />
0<br />
compensated demand curve � , , �<br />
1<br />
curve � , , �<br />
a x q u lies below the inverse compensated demand<br />
a x q u for the reason that scale effects depend on the elasticity of<br />
substitution between q and the x ‟s. In the presence of two goods, Park (1997) finds<br />
that the Hicks elasticity of substitution equals the Allen-Uzawa elasticity of<br />
substitution. The difference between WTP and WTA thus arises whenever<br />
substitutability is imperfect.<br />
What happens to the agent‟s behavior if we now distinguish foregone gains<br />
from real losses?<br />
1.4. Imperfect substitutability and the endowment effect<br />
Hanemann (1991) points out in his footnote 25 that Kahneman and Tversky‟s<br />
(1979) loss aversion, observed from some reference point, differs from the standard<br />
disparity. Indeed, in the Hicksian framework, preferences over consumption bundles<br />
are independent of initial endowments. In reference to the gain and loss perspective,<br />
Thaler (1980) proposed the term endowment effect. When an agent is endowed with a<br />
good, her reference point changes, she shifts her position on the map, and the shape of<br />
her indifference curve is altered.<br />
If we adapt the standard framework to the loss aversion idea, a gain or a loss<br />
in q can be written q � q � � and q � q � � , with 0 � � . Assume agent‟s utility is<br />
1 0<br />
�<br />
1 0<br />
�<br />
affected by variations of the environmental quality level q . In this case, her utility<br />
� , , �<br />
0 0<br />
u v p q y<br />
� , which now involves a single indifference curve, changes either to<br />
u � in a case of a gain or to u � in case of a loss:<br />
0 � , , �<br />
�<br />
u � v p q � � y<br />
[14a]<br />
0 � , , �<br />
�<br />
u � v p q � � y<br />
[14b]<br />
40
Bateman et al. (1997) define two additional mea<strong>sur</strong>es, identified with some<br />
reference point. Regarding the first mea<strong>sur</strong>e, the question is what additional amount<br />
of private consumption is as preferable as an increase in the environmental quality.<br />
This is the equivalent gain, equal to WTA. Regarding the second mea<strong>sur</strong>e, the<br />
question is what loss of private consumption would be just as preferable as a decrease<br />
in the environmental quality. This is the equivalent loss, equal to WTP.<br />
When the agent is endowed, fixing a gain and a loss in [3a] and [3b] or [4a]<br />
and [4b] gives the following relationships: C � or compensating gain is the maximum<br />
amount she would pay to secure the gain; E � or equivalent gain is the minimum<br />
amount she would accept to sacrifice the gain; E �<br />
� or equivalent loss is the<br />
maximum amount she would give up to avoid the loss; C �<br />
� or compensating loss is<br />
the minimum amount she would accept to tolerate the loss. The summary is<br />
recapitulated in Table 1.2.<br />
Table 1.2. Welfare indices and context-dependence<br />
1 0<br />
1 0<br />
loss ( q � q � � ) gain ( q � q � � )<br />
�<br />
� �<br />
�C�WTA�� 0�<br />
C WTP<br />
41<br />
�<br />
� �� 0�<br />
� �<br />
� �<br />
�E�WTP�� 0�<br />
E WTA<br />
� �� 0�<br />
� �<br />
� �<br />
� �<br />
Unlike the standard disparity alias �WTA � WTP � or �WTA WTP �<br />
� , where<br />
changes go in the same direction, a gain and loss disparity is computed differently,<br />
simply because we observe changes that depart in opposite directions from some<br />
reference point. Here, we subtract WTA to tolerate the loss and WTP to guarantee the<br />
� �<br />
gain or �WTA WTP �<br />
� . From [3a] and [3b], it follows:<br />
1 0 0 0 0 0 1 0<br />
� , �, � � , , � � , , � � , �,<br />
�<br />
� �<br />
�C �C � �e p q u � e p q u � � �e p q u �e<br />
p q u �<br />
� � � � [15a]
Since the utility function u �x, q�<br />
is quasi-concave in � q�<br />
0<br />
the expenditure function � , , �<br />
42<br />
x, , when q increases<br />
e p q u decreases, i.e. is convex in q , as less income is<br />
necessary to attain the fixed utility level. The second reaction is that the indirect<br />
utility function v �p, q,<br />
y�<br />
is quasi-concave in � y�<br />
q, , which means that the cross-<br />
2 2<br />
partial derivative only implies the substitution effect eqq � e q �<br />
� � � . As a matter of<br />
fact, the income effect – the spacing of the indifference curves – does not count, for<br />
gain and loss perspective involves a single indifference curve<br />
positive and negative change, thus:<br />
0<br />
u observed from some<br />
� �<br />
WTA � WTP � 0<br />
[15b]<br />
Proposition 1.2.: Imperfect substitutability between q and the x ‟s in the indirect<br />
utility function causes disparity either between WTP and WTA or between gain and<br />
loss, independently from the reference.<br />
Proof: In the appendix.<br />
Fig. 1.2. illustrates the four mea<strong>sur</strong>es, observed from some reference<br />
0<br />
coordinates �q , y �.<br />
Grey curves depict the same pre-endowed utility<br />
0<br />
u observed<br />
from two reference points. Through the incursion of context-dependence, the utility<br />
changes from<br />
0<br />
u to either u � (a gain in utility) or u � (a loss in utility). The reference<br />
point for WTA � and WTP � is G. Viewed from G, the distance from<br />
0<br />
q to<br />
q � q � �<br />
1 0<br />
�<br />
is a gain in level of the environmental quality. For WTA � and WTP � the reference<br />
point is L. Viewed from L, the distance from<br />
0<br />
q to<br />
q � q � � is perceived as a loss<br />
1 0<br />
�<br />
in level of the environmental quality. The endowment effect induces the pivoting of<br />
the indifference curve from the reference point, which illustrates the discontinuity in<br />
the slope from<br />
0<br />
u to u � or u � . The steeper the indifference curve, the less the<br />
substitutability between q and the x ‟s that y buys.
The difference between WTA � and WTA � , which is essential to distinguish<br />
the standard disparity from the gain and loss disparity, lies in the way the loss is<br />
perceived. In the first case, the agent is asked to state her value to give up a gain from<br />
an increase in the environmental quality. This is an opportunity loss. This cannot be a<br />
right. In the second case, the agent is asked to state her value to suffer a loss from a<br />
decrease in environmental quality. This is a net loss and it differs from the former.<br />
.<br />
Fig. 1.2. Reference-dependent preferences<br />
The difference is due to imperfect substitutability, for agents take more account of the<br />
goods they can not regain. When agents are asked to value their losses in monetary<br />
terms, the behavioral effect of loss aversion arises and they shift their indifference<br />
curves. They know they have the right to be compensated for the loss and claim this<br />
right in form of high WTA � statements.<br />
� � � �<br />
Transitivity implies that whenever WTA > WTA and WTA > WTP ,<br />
� �<br />
WTA > WTP .<br />
WTA –<br />
WTA +<br />
y<br />
WTP +<br />
WTP –<br />
y<br />
0<br />
G<br />
1<br />
q� 43<br />
1<br />
q� L<br />
q
1.5. Imperfect substitutability and loss aversion<br />
In their 1991 article, Tversky and Kahneman propose the behavioral<br />
reference-dependent theory as an alternative to the Hicksian theory of preferences.<br />
Outcomes are now valued using a value (utility) function where agents have<br />
preferences over goods relative to some reference level � x, q�<br />
44<br />
r r seen as the status quo.<br />
According to them, (i) all is perceived as a gain or a loss; (ii) losses are weighted<br />
more heavily than gains or agents are loss averse; and (iii) the marginal value of gains<br />
or losses exhibits diminishing sensitivity. They assume that preferences are transitive,<br />
continuous and nondecreasing (but not convex).<br />
r r stands for the reference points for consuming � q�<br />
If � x, q�<br />
function changes to u u� x, q, rx, rq�<br />
x, , the utility<br />
� ; the demand functions take the form of<br />
i<br />
i<br />
� � , , , , � and xi h � p, q, u, rx , rq<br />
�<br />
x h p q y r r<br />
i x q<br />
� ; the indirect utility is now<br />
v� p, q, y, rx, r q�<br />
just as is the expenditure function � , , , x, q�<br />
e p q y r r . These new<br />
functions are discontinuous at the reference point (Putler 1992). Fig. 1.3. shows a<br />
typical loss aversion curve observed within the context of welfare mea<strong>sur</strong>ement.<br />
WTA –<br />
y�rx WTP +<br />
y<br />
0<br />
rq �<br />
q�rq rq �<br />
Fig. 1.3. Loss aversion in welfare mea<strong>sur</strong>es<br />
q
The additive formulation of the constant loss aversion model used by Tversky<br />
and Kahneman gives the following indirect utility function:<br />
q y<br />
� , � R �i�q��i�y� u � v q y � � q � r � y � r �<br />
� � [16]<br />
with R' � 0.<br />
Parameters<br />
q<br />
� i and<br />
y<br />
� i , for 1,2<br />
i � and � � 1,<br />
are defined as coefficients<br />
of loss aversion for dimensions r q and r y . They magnify the disutility of losing some<br />
environmental quality. When the agent perceives a change as a gain, this coefficient<br />
amounts to 1 1 � � , which means that the agent has neoclassical utility. When she<br />
perceives a loss, this coefficient amounts to 2 >1<br />
y y<br />
2 1<br />
� �<br />
� > � � 1 .<br />
45<br />
i<br />
q q<br />
� . We can see that � > � �� 1�and<br />
2 1<br />
Definition 1.2.: The change from the reference level r q to either a gain rq � or a loss<br />
rq � , while ry� y,<br />
gives the following:<br />
q q y y<br />
�� �i � �1 if q � rq � 0 and �i � �1<br />
if y � ry<br />
� 0<br />
� q q y y<br />
�� �i � �2 if q � rq < 0 and �i � �2<br />
if y � ry<br />
� 0<br />
In terms of coefficients of loss aversion, the welfare mea<strong>sur</strong>es matrix becomes<br />
what is shown in Table 1.3.<br />
Table 1.3. Welfare indices in a gain and loss perspective<br />
� �<br />
loss ( rq � rq<br />
� � ) gain ( rq � rq<br />
� � )<br />
� � � � 1 2 WTP<br />
q y<br />
� � � �<br />
2 1 WTA<br />
q y<br />
�<br />
� � � � 1 1 WTA<br />
q y<br />
� � � �<br />
2 2 WTP<br />
q y<br />
�<br />
�<br />
�
q y<br />
Since �1 �1<br />
1<br />
� � , < 2<br />
q<br />
y �<br />
�<br />
� or<br />
�<br />
R �� � , i.e. y � e�q, u�<br />
is the inverse of u v�q, y�<br />
2<br />
+<br />
WTP < WTA � . If we invert the function<br />
46<br />
� with �R � y<br />
�<br />
differentiate it with respect to u , the following disparity arises:<br />
q<br />
�u� �u� i �q rq�<br />
�� ' if � ��� �<br />
eu �q, u� � ��<br />
'�u<br />
�<br />
� if � < �<br />
y<br />
� �2<br />
�1<br />
where � � � R � �<br />
q �u� i �q � rq�<br />
� � � and � '� 0.<br />
y� � r , and<br />
i y<br />
[17]<br />
The indirect utility function is quasi-concave because of the monotone<br />
transformation � �<br />
q<br />
R � in [16]. Moreover, i �q rq�<br />
� � is a concave function of q , which<br />
illustrates the gain and loss disparity with decreasing sensitivity to losses. Since<br />
2 >1<br />
y<br />
� , when q increases u<br />
e decreases, which implies the negativity of the derivative<br />
e qu from changes in q . As a result, we get back to the standard disparity between<br />
WTP and WTA.<br />
Recall that the curvature of the indifference curves shows diminishing<br />
marginal utility between the consumption bundles, and thus the standard WTA-WTP<br />
disparity. Furthermore, it generates disparity between gain and loss because of the<br />
imperfect substitution in the indirect utility function between y and some function of<br />
the environmental quality q . Through the discontinuity in the slope at the reference<br />
point, loss aversion theory implies convex indifference curves. On that subject,<br />
Hanemann (1999) argues that the assumption of quasi-concave utility function<br />
suffices to observe convexity. Quasi-concavity with inversely proportional disparity<br />
to the substitution effect can explain the disparity between gain and loss. The<br />
endowment effect within loss aversion can be explained through less than perfect<br />
substitutability.
The behavioral theory of loss aversion also works with distance effects. In<br />
terms of the distance function, it is a matter of distance between coordinates of some<br />
level of y or q and the agent‟s reference point. In this case, the function becomes<br />
� , , , x, q�<br />
d x q u r r . Adding coefficients of loss aversion into the distance function yields<br />
now a weighted distance function of the form:<br />
p p<br />
� �<br />
q y<br />
� , , � R �i�q��i�y� d � d x q u � q � r � y � r<br />
�� ��<br />
[18]<br />
where p � 1 denotes the metric. When p � 1,<br />
the distance is mea<strong>sur</strong>ed as the sum of<br />
weighted absolute differences. We then fall on [16]. The distance function recovers<br />
from the expenditure function. Therefore, imperfect substitutability can once again<br />
explain the gain and loss disparity.<br />
1.6. Imperfect substitutability and boundedness<br />
Randall and Stoll (1980) demonstrate that the disparity between WTP and<br />
WTA is bounded by the ratio between the price flexibility of income and endowment.<br />
Cook and Graham (1977) assert that the compensation demanded for irreplaceable<br />
commodities, which we can assume to be imperfectly substitutable, depends on the<br />
initial level of wealth or endowment. As the probability of loss p � 1,<br />
WTA,<br />
dependent on the income that buys the x‟s, tends to infinity as the indifference curve<br />
is asymptotic to the vertical line at p � 1.<br />
This is what Amiran and Hagen (2003) also<br />
suggest in a slightly different manner: in presence of asymptomatically bounded<br />
utility functions, there exists an initial level of wealth sufficiently high to produce an<br />
infinite WTA – . Nevertheless, the substitution effect still plays a capital role, for it<br />
induces frictional trade-off between public and private goods. In terms of elasticity,<br />
the authors show that the income elasticity of the inverse compensated demand is<br />
bounded above and below by positive values independent of the amounts of public<br />
goods.<br />
47
In case of reference-dependent preferences, we believe that imperfect<br />
substitutability accentuates the pivoting of the indifference curve, which in turn can<br />
produce infinite compensation demanded. We replace the nonsatiation assumption<br />
from Amiran and Hagen (2003) by the assumption that for each level of income y, the<br />
status quo q r is strictly preferred to some net loss of the public good rq �� with<br />
� >0<br />
or that u� rq , y� < u� rq, y�<br />
�� with r �� < r . A double outcome arises. The<br />
48<br />
q q<br />
first outcome lies in the convex curvature of the indifference curve. In point of fact,<br />
imperfect substitutability induces a steeper slope for higher opportunity losses (see<br />
Fig. 1.4.: grey segment and arrow 1). The second outcome results from the<br />
enlargement of the substitution effect due to aversion of net losses, yielding<br />
clockwise rotation and, accordingly, a steeper slope of the initial indifference curve<br />
(see Fig. 1.4.: black segment and arrow 2).<br />
Fig. 1.4. Unboundedness of the compensation demanded<br />
Beyond some level of loss of the public good �� 0 in view of their reference point,<br />
i.e.<br />
0<br />
q rq<br />
WTA –<br />
WTA +<br />
� � � , standard agents ask for an infinite monetary compensation. Formally,<br />
this yields a level of monetary compensation s – analogue to WTA – strictly inferior<br />
to the disutility of the loss:<br />
y<br />
y<br />
0<br />
2<br />
q 0<br />
1<br />
rq<br />
q
0<br />
� q, �> � q � > � , �<br />
u r y z r � � � s u q y<br />
[19]<br />
Proposition 1.3.: In case of reference-dependent preferences and imperfect<br />
substitutability between q and the x‟s that y buys, large net losses of the public good<br />
can be infinitely uncompensated.<br />
Proof: In the appendix.<br />
Hanemann (1999) points out that the wealth effect in Cook and Graham is not<br />
the income effect typically considered in consumer demand theory. While being true,<br />
let us recall that the income effect does not count within context-dependence. We<br />
therefore explain the infinite limit of WTA – by the pivoting of the indifference curve<br />
from the reference (endowed) level of the public good. Again, this is a net loss<br />
perception magnifying the substitution effect. Contrary to Cook and Graham who find<br />
infinite WTA – as the probability of losses moves towards one, our indifference curve<br />
is asymptotic to the vertical line at<br />
severe and approach<br />
0<br />
q , which shows infinite WTA – when losses are<br />
0<br />
q . Unlike the previous models – which unquestionably<br />
consider substitutability as the mainspring for infinite monetary compensation – our<br />
design neither depends on the initial level of wealth or the initial endowment in<br />
market goods nor on the boundedness of the utility function. It rather depends on the<br />
severity of loss of the public goods combined with their unfeasibility to be perfectly<br />
substitutable.<br />
In the behavioral loss aversion model, when an agent stands at � q, x�<br />
49<br />
r r , that is,<br />
at the kink point, q and y are perfect substitutes, for she is equidistant to both<br />
references points and indifferent between the level of environmental quality and her<br />
income. Except these coordinates, any other point along the curve exhibits imperfect<br />
substitutability. As can be noticed in terms of distance minimization, above the kink<br />
point she substitutes the loss of the environmental quality with monetary<br />
compensation. Below is the opposite. Because of loss aversion, as r � � � r
have �r r � v r 0<br />
� � � � � � . When rq �� goes farther from r q , additional decreases<br />
q q q<br />
in q lead to smaller changes in utility, which implies<br />
50<br />
2 2<br />
� v �q � 0 . In other terms,<br />
diminishing sensitivity implies a lower substitution effect in both gains and losses.<br />
Conversely to the standard model, the marginal disutility from environmental quality<br />
downgrading is decreasing as the agent moves from the reference point, implying a<br />
bounded value for compensation 8 .<br />
Proposition 1.4.: Inside the behavioral theory of loss aversion, given constant<br />
diminishing sensitivity towards losses, the substitution effect is bounded.<br />
Proof: In the appendix.<br />
Hence, there is a difference between the Hicksian standard paradigm and loss<br />
aversion in the representation of context-dependence. If we superpose the<br />
indifference curves illustrating the willingness-to-accept to tolerate a loss, their<br />
respective curvatures reveal two types of behavior on the subject of losses. The grey<br />
segment represents the standard theory context-dependence. The black segment<br />
stands for the behavioral model of loss aversion (see Fig. 1.5.).<br />
Inside the standard model, agents show increasing marginal disutility as<br />
rq �� tends towards<br />
0<br />
q . Inside the behavioral model, agents exhibit high loss<br />
aversion with small changes as regards their reference point, but they turn out to be<br />
less and less sensitive as<br />
r q<br />
0<br />
q � � � . Diminishing sensitivity of the marginal value of<br />
losses clarifies this phenomenon. The farther something moves from a reference<br />
point, the less additional changes should matter, which in our case <strong>sur</strong>prisingly means<br />
increasing substitutability. This counter effect appears because agents are myopic,<br />
which makes them feel unconcerned by changes out of their visual field. As a<br />
consequence, they end up asking for a bounded amount of compensation, no matter<br />
the additional degradation of the environment.<br />
8 The shape of the losses is represented in a positive space because we deal with positive values of<br />
welfare indices.
WTA –<br />
y�rx Fig. 1.5. Comparison between reference-dependent indifference curves<br />
The significance of it is non-negligible. In case of irreversible damages to the<br />
environment or high losses of public goods with regards to their initial level –<br />
commodities that we know to be imperfectly substitutable –, standard agents which<br />
turn out to be far-sighted will ask for an infinite monetary compensation, whereas loss<br />
aversion agents will ask for a bounded amount of compensation, and neither can<br />
adapt their reference points. While economists have long considered loss aversion to<br />
degenerate agents‟ rational preferences, we see that past some level of changes in q, it<br />
limits their proclivity towards abnormal valuation.<br />
1.7. Concluding remarks<br />
Applied to market valuation of the public goods, this chapter dealt with<br />
imperfect substitutability in both standard welfare and reference-dependence theories.<br />
Imperfect substitutability in the indirect utility function can provoke disparity either<br />
between WTA and WTP or between gain and loss. Further, the same quasi-concave<br />
utility functions can explain the endowment effect.<br />
y<br />
0<br />
0<br />
q<br />
51<br />
rq<br />
q
What is the point of finding that imperfect substitutability plays a role in both<br />
the WTA and WTP disparity and the gain and loss disparity? According to the above,<br />
it basically means that agents‟ unwillingness to substitute an environmental good or<br />
service increases with its defective substitutability. When agents substitute a public<br />
good for a private good, an opportunity loss appears and induces the standard<br />
disparity. In case the scenario to price is a loss instead of a foregone gain, loss<br />
aversion transforms the opportunity loss in a net loss, which enlarges the initial<br />
disparity, for people heavily value things they cannot regain. Experimental findings<br />
from Boyce et al. (1992) and Chapman (1998) support this conclusion. At last, the<br />
substitution effect observed from some reference point has a bound inside the<br />
behavioral model of loss aversion. This could be tested in a laboratory. Whether<br />
agents should have infinite values for severe or irreversible losses might be the topic<br />
that decides which model better values environmental preferences.<br />
Yet, these common findings must be toned down. Valuing environmental<br />
goods or services calls for an understanding of the public and private benefits derived<br />
from the public good. This is partially en<strong>sur</strong>ed, as environmental commodities are<br />
unfamiliar to agents and their benefits for utility obscure in most cases. The risk of<br />
having naïve valuations is existent. Only an interactive market-like setting permits to<br />
<strong>sur</strong>mount these limits, and hypothetical markets remain devoid of market interactions.<br />
Experimental markets are thus essential in the contingent valuation. In<br />
experimentation, the early disparity between welfare mea<strong>sur</strong>es is redundant,<br />
supporting either of the two effects. But their confrontation occults the market<br />
efficiency which rules the economic valuation. Indeed, markets bound anomalies by<br />
means of ad hoc incentives, for they aid agents to correct their untruthful or naïve<br />
valuations. The next step consists in identifying, by probing into auction mechanisms,<br />
why some of them reduce the disparity better than others. As well, studying agents‟<br />
context-dependent behavior faced with irreversible environmental damages and<br />
ambiguity – when they can adapt their reference points – is a matter of future<br />
research.<br />
1.8. References<br />
52
Amiran, E. and Hagen, D. (2003), “Willingness To Pay and Willingness To Accept:<br />
How Much Can They Differ? Comment”, American Economic Review, 93: 458–<br />
463.<br />
Bateman, I., Munro, A., Rhodes, B., Starmer, C. and Sugden, R. (1997), “A Test of<br />
the Theory of Reference-Dependent Preferences”, Quarterly Journal of<br />
Economics, 112: 479–505.<br />
Boyce, R., Brown, T., McClelland, G., Peterson, G. and Schulze, W. (1992), “An<br />
Experimental Examination of Intrinsic Values as a Source of the WTA-WTP<br />
Disparity”, American Economic Review, 82: 1366–1373.<br />
Blackorby, C., Primont, D. and Russell, R. (1978), “Duality, Separability and<br />
Functional Structure”, Theory and Economic Applications, North-Holland, New<br />
York.<br />
Brookshire, D. and Coursey, D. (1987), “Mea<strong>sur</strong>ing the Value of a Public Good: An<br />
Empirical Comparison of Elicitation Procedures”, American Economic Review,<br />
77: 554–556.<br />
Cook, P. and Graham, D. (1977), “The Demand for In<strong>sur</strong>ance and Protection: The<br />
Case of Irreplaceable Commodities”, Quarterly Journal of Economics, 91: 143–<br />
156.<br />
Chapman, G. (1998), “Similarity and Reluctance to Trade”, Journal of Behavioral<br />
Decision Making, 11: 47–58.<br />
Deaton, A. (1979), “The Distance Function in Consumer Behavior with Applications<br />
to Index Numbers and Optimal Taxation”, Review of Economic Studies, 46: 391–<br />
405.<br />
Ebert, U. (1984), “Exact Welfare Mea<strong>sur</strong>es and Economic Index Numbers”, Journal<br />
of Economics, 44: 27–38.<br />
Ebert, U. (1993), “A Note on Willingness to Pay and Willingness to Accept”, Social<br />
Choice and Welfare, 10: 363–370.<br />
Hanemann, W. (1991), “Willingness to Pay and Willingness to Accept: How Much<br />
Can they Differ?”, American Economic Review, 81: 635–647.<br />
Hanemann, W. (1999), “The Economic Theory of WTP and WTA”, In Valuing the<br />
Environment Preferences: Theory and Practice of the Contingent Valuation<br />
Method in the US, EC and Developing Countries, edited by Ian Bateman and Ken<br />
Willis, Oxford: Oxford University Press.<br />
53
Hicks, J. (1943), “The Four Consumer‟s Surpluses”, Review of Economic Studies, 11:<br />
31–41.<br />
Kahneman, D. and Tversky, A. (1979), “Loss aversion in Riskless Choice: A<br />
Reference-Dependent Model”, Quarterly Journal of Economics, 106: 1039–1061.<br />
Kahneman, D., Knetsch, J. and Thaler, R. (1991) “Anomalies: The endowment effect,<br />
Loss Aversion, and Status Quo Bias”, Journal of Economic Perspectives, 5: 193–<br />
206.<br />
Knetsch, J. and Sinden, J. (1984), “Willingness to Pay and Compensation Demanded:<br />
Experimental Evidence of an Unexpected Disparity in Mea<strong>sur</strong>es of Value”,<br />
Quarterly Journal of Economics, 99: 507–521.<br />
Mäler, K. (1974), Environmental Economics: A Theoretical Inquiry, John Hopkins<br />
University Press, Baltimore.<br />
Morrisson, G. (1997) “Resoving Differences in Willingness to Pay and Willingness to<br />
Accept: Comment”, American Economic Review, 87: 236–240.<br />
Park, H. (1997), “Randall and Stoll‟s Bound in an Inverse Demand System”,<br />
Economics Letters, 56: 281–286.<br />
Putler, D. (1992), “Incorporating Reference Point Effects into a Theory of Consumer<br />
Choice”, Marketing Science, 11: 287–309.<br />
Randall, A. and Stoll, J. (1980), “Consumer‟s Surplus in Commodity Space”,<br />
American Economic Review, 70: 449–455.<br />
Sinclair-Desgagné, B. (2005), “Calcul économique et développement durable”, Esprit<br />
critique, Vol. 07–N.01, available at http://www.espritcritique.fr.<br />
Thaler, R. (1980), “Toward a positive theory of consumer choice”, Journal of<br />
Economic Behavior and Organization, 1: 39–60.<br />
Tversky, A. and Kahneman, D. (1991), “Loss Aversion in Riskless Choice: A<br />
Reference-Dependent Model”, Quarterly Journal of Economics, 106: 1039–1061.<br />
54
1.9. Appendix<br />
Proof of Proposition 1.1.<br />
The demonstration is as follows. After Randall and Stoll (1980), WTP and WTA for<br />
changes in public goods should not differ with small income effects. They bound<br />
�E� C�via<br />
the income elasticity of demand (or income elasticity of willingness-to-<br />
pay) of the public good. For example, when the price of a certain good one changes,<br />
the disparity amounts:<br />
�<br />
1<br />
1<br />
1 0<br />
� , , � � , , �<br />
p<br />
� �<br />
p � p p �<br />
E �C � e p q u � e p q u dp<br />
or<br />
0 1 1<br />
1<br />
1<br />
1 1<br />
� � � � � �<br />
1<br />
� � � �<br />
E �C � e p, q, u � g p, q, u � g p, q, e p, q, u � e p, q, u<br />
p u u y u<br />
The income effect associated with good one – the second cross-partial derivative<br />
2<br />
epu e p1 u<br />
1<br />
� � � � – establishes the size of the disparity, the limit being the<br />
individual‟s income. The bounding method carries over to welfare mea<strong>sur</strong>es of the<br />
quantity changes. The analogous result for a change in q gives the cross-partial<br />
derivative<br />
market goods:<br />
�<br />
q<br />
2<br />
equ e q u<br />
1<br />
� � � � , i.e. the substitutability of the nonmarket good by means of<br />
0 1<br />
� , , � � , , � � , , �<br />
E �C � �e 0<br />
q<br />
q p q u eq p q u �<br />
�<br />
�<br />
�<br />
dq � �equ<br />
p q u<br />
For a change in the public good‟s level, Hanemann (1991) demonstrates that the<br />
second cross-partial derivative e qu reflects the substitution effect. Indeed, from [8]<br />
and the differentiation of the compensating demand function for q , we hold the<br />
derivative involved in changes in q that impact on u :<br />
55
� , , �<br />
e p q u<br />
g � p, �, e p, �, u �� e� p, �,<br />
u�<br />
� , �, , � , � y � , �,<br />
�<br />
� p, q, u�<br />
q<br />
y � �<br />
� �<br />
2<br />
� e ��<br />
ˆ ˆ<br />
� � �<br />
�q�u �u gˆ p eˆ p u �gˆpu�q qu q<br />
�<br />
q<br />
By the Hicks decomposition, the precedent becomes:<br />
� � �<br />
� � �<br />
q<br />
gˆ u p, , u<br />
equ � p, q, u�<br />
= q<br />
gˆ p, , u<br />
�<br />
This difference between WTP and WTA depends on the price flexibility of income<br />
and thus the ratio of the income elasticity of the ordinary demand function for q to<br />
the elasticity of substitution between q and the x ‟s 9 . The numerator represents the<br />
income effect of q in the hypothetical market, established from the derivative of the<br />
demand function with respect to income. The denominator is the own-price derivative<br />
of the compensated demand function for q and gives the aggregate Allen-Uzawa<br />
elasticity of substitution between q and the private goods weighted by the budget<br />
share of the same private goods.<br />
Changes in prices and changes in q both vary with income and depend on a cross-<br />
partial derivative of the expenditure function. And when e qu
A necessary and sufficient condition for the disparity between gain and loss to occur,<br />
i.e. WTA > WTP<br />
� �<br />
, is that v �p, q,<br />
y�<br />
is quasi-concave in � y�<br />
57<br />
0<br />
p, or that � , , �<br />
e p q u is a<br />
convex function of q . In this case, the second partial derivative e qq must be strictly<br />
positive. Let us look at the expenditure function.<br />
The disparity arises because of the convexity of the initial indifference curve<br />
follows from [3a] and [3b] that:<br />
1 0 0 0 0 0 1 0<br />
� , �, � � , , � � , , � � , �,<br />
�<br />
� �<br />
�C �C � �e p q u � e p q u � � �e p q u �e<br />
p q u �<br />
� � � �<br />
Which gives:<br />
1 0 1 0 0 0<br />
0 � e� p, q�, u � � e� p, q�, u � � 2 e� p, q , u �<br />
� 0 0 � � � 0<br />
� �<br />
0 � � � 0<br />
� �<br />
0 �<br />
2 e p, q , u e p, q , u e p, q , u<br />
� �<br />
1<br />
e p q u e p q u e p q u<br />
2<br />
0 0 0 0 0 0<br />
� , , � � � , � �, � � � , � �,<br />
�<br />
0 0<br />
In parallel, � , , �<br />
e p q u can be rewritten as:<br />
�� � �� � � � ���<br />
� 1 0 0 0 �<br />
e�p, q q , u �<br />
� 2<br />
�<br />
Substituting the precedent into the general inequality gives:<br />
0 0 0 0 0 0 0<br />
�� � �� � � � ��� � � � � � � � � � �<br />
� �<br />
� 1 � 1<br />
e� p, q q , u � e p, q , u e p, q , u<br />
� 2 � 2<br />
� 1 1 � 1 1<br />
e� p, q �1 � q , u � e p, q , u �1 �e<br />
p, q , u<br />
� 2 � 2 � �<br />
2 � 2 �<br />
0 � � 0 0 0 0 � � 0 0<br />
� � �� � � � � �� � � � � � � � � � � �<br />
0<br />
u . It
From which directly follows the convexity of the expenditure function of q . The<br />
expenditure function being convex, we have e qq >0.<br />
There is a disparity between gain and loss. �<br />
Proof of Proposition 1.3.<br />
For<br />
0<br />
q rq rq<br />
� � � � with >0<br />
�<br />
us set z �q� �sup y u �q, y� �u�rq,<br />
y�<br />
�<br />
0<br />
� , we have u0 � u �rq , y�<br />
and u �q, y� u �q , y�<br />
58<br />
� . Let<br />
� �<br />
� �<br />
for a level of monetary compensation such<br />
that the utility remains constant. For each 0<br />
q � we have z �q� u� q, y�<br />
� . The<br />
supremum z�q � is increasing in q. This says that for each level of income y and for<br />
0<br />
q � we have u� rq, y� u� rq , y�<br />
r q<br />
the net loss of the public good � .<br />
Let us set � � � �=<br />
q q<br />
� � � because the status quo is always preferred to<br />
z r � z r � � s where s > � is the compensation analogue to WTA.<br />
With z�r q � being the supremum for u �rq , y � and � q �<br />
z r �� being the supremum for<br />
0 � q �� , � is there y that gives u� q , y� > z �rq �� � or � q, �> � q �<br />
u r y<br />
We know that<br />
0<br />
q rq rq<br />
u r y z r � � � s ?<br />
0<br />
� � � � so for all y we have z �q � z �rq �<br />
0 � � � � q � � � . By definition we know that z �rq � s z �rq� z q z r<br />
0<br />
0 0 � � � � � q � holds. Moreover, z �q � u� q , y�<br />
z q s z r<br />
0 0 0 � � � � � , � or z �rq � s u �q , y�<br />
z q s u q y<br />
As<br />
� � � � .<br />
0<br />
0<br />
q � we have z �rq � z �q �<br />
r q<br />
0<br />
� q,<br />
� � �<br />
u r y � z q � s .<br />
� and<br />
� � � � such that<br />
� because z �q� u� q, y�<br />
0<br />
� � � or z �rq � s z �q �<br />
� thus<br />
� � hence
0<br />
From the above we see that u� q , y� < z �rq � s u �rq, y�<br />
Proof of Proposition 1.4.<br />
59<br />
� � � � �<br />
Let us now prove that WTA is bounded within the behavioral model of loss aversion.<br />
One way comes directly from the construction of the model: according to diminishing<br />
sensitivity, smaller changes in � should be accompanied by smaller increases in y ,<br />
the utility being constant, thus<br />
2 2<br />
� u �q � 0.<br />
��rq � ��� �0, r � q�<br />
and some value function *<br />
v C�R�� � on �0, r � q � which is concave<br />
and nonincreasing, one has: v�r q � v'� rq ��rq rq � v�r q �<br />
� � � � � � � . The right-hand<br />
expression of the weak inequality is the tangent of v at r q . It gives<br />
� q � '�<br />
q �� q � � q �<br />
v r � � � v r �r � v r when rq � � � 0 which is independent of rq ��.<br />
Hence, the losses‟ side of the value function is bounded. �
61<br />
Chapter 2<br />
Private Valuation of a Public Good<br />
in Three Auction Mechanisms<br />
Abstract<br />
We evaluate the impact of three auction mechanisms – the Becker–DeGroot–<br />
Marschak (BDM) mechanism, the second-price auction, and the random nth-price<br />
auction – in the mea<strong>sur</strong>ement of private willingness-to-pay and willingness-to-accept<br />
for a pure public good. Our results show that the endowment effect can be eliminated<br />
with repetitions of the BDM mechanism. Yet, on a logarithmic scale, the random nthprice<br />
auction yields the highest speed of convergence to welfare indices‟ equality.<br />
Overall, we observe that subjects value public goods in reference to their private<br />
subjective benefit derived from the public good funding.<br />
Keywords: contingent valuation, WTP-WTA gap, auctions, public good private<br />
provision<br />
JEL classification: C91, D44, Q53
2.1. Introduction<br />
62<br />
"I have never known much good done<br />
by those who affected to trade for the<br />
public good." Adam Smith<br />
The experimental private provision of public goods based on the contingent<br />
valuation method is often used to value public goods such as health, safety or<br />
environment. Estimating preferences for public goods is however laborious, for<br />
individuals reveal behavioral biases during their valuation process.<br />
In accordance with the Coase theorem (Coase 1960), neoclassical theory<br />
postulates that with null income effect and close substitutes, the willingness-to-pay<br />
(WTP), which is the price at which an individual is ready to buy a commodity, and<br />
the willingness-to-accept (WTA), which is the price at which an individual is ready to<br />
sell the same commodity, should be equal (Randall and Stoll 1980, Hanemann 1991).<br />
If the good is available in an active market at the market price, an individual‟s WTP<br />
and WTA should be similar. And if people face similar transaction costs, WTP and<br />
WTA should be similar among people as well. Yet, experimental research that<br />
stemmed from contingent valuation studies has found large disparities between the<br />
WTP and WTA. The endowment effect, or loss aversion, as a behavioral feature is<br />
often invoked to explain the disparity. It occurs when people offer to sell a commonly<br />
available good in their possession at a substantially higher rate than they will pay for<br />
the identical good not in their possession. The other effect, promoted to explain the<br />
disparity, is imperfect substitutability.<br />
Two remedies help remove the initial disparity. The first corresponds to<br />
market settings. Market institutions serve as social tools that induce and reinforce<br />
individual rationality (Smith 1991). Gode and Sunder (1993) assert that an auction<br />
market exerts a powerful constraining force on individual behavior. Cherry et al.<br />
(2003) suggest that a dynamic market environment with repeated expo<strong>sur</strong>e to<br />
discipline is necessary to achieve rationality. When they act rationally, individuals<br />
refine their statements of value. List (2003a) provides evidence consistent with the<br />
notion that experience in bidding with an incentive-compatible auction can remove
the WTA/WTP gap. The second corresponds to market repetition. The motive for<br />
repeating auctions that are incentive-compatible is that individuals require experience<br />
to understand that sincere bidding is the dominant strategy (Coppinger et al. 1980)<br />
and to realize their true valuation of unfamiliar products (Shogren et al. 2000). Plott<br />
(1996) advances a discovered preference hypothesis argument, positing that<br />
responses reflect a type of internal search process in which subjects use practice<br />
rounds to discover their preferences. The experience they gain is reflected in their<br />
bidding behavior. Hence, the imperfect substitutability effect disappears when the<br />
value of the unfamiliar good is perfectly revealed.<br />
Market-based mechanisms such as auctions are widely studied as a means of<br />
buying and selling resources. Auctions took part in the environmental valuation to<br />
answer two questions: (1) which effect counts the most in the WTP and WTA<br />
disparity? and (2) which of the auction mechanisms best removes this disparity?<br />
At first, Kahneman et al. (1990) report experimental evidence of the<br />
endowment effect. They perform an experiment on WTP and WTA by way of<br />
hypothetical telephone inquiry, trading environmental improvements and<br />
preparedness for disasters. They find that randomly assigned owners of an item<br />
require more money to separate from their possession than random buyers are willing<br />
to pay to acquire it. To elicit individuals‟ estimates, they use a Becker-DeGroot-<br />
Marschak mechanism (BDM) – described later on – with random exogenous price<br />
feedback. According to their results, preferences are dependent on endowments, even<br />
in market settings.<br />
Shogren et al. (1994) assert that Kahneman et al.‟s experiment creates<br />
artificial scarcity. They find no evidence of the endowment effect on trading candy<br />
bars, for the values converge over time. But, in the experiment with contaminated<br />
food – a good with imperfect substitutes that can be considered as nonmarketed –<br />
they show that the discrepancy remains significant after iteration.<br />
<strong>La</strong>ter on, Shogren et al. (2001) test three auction mechanisms to trade candy<br />
bars and mugs and suggest that the auction mechanism can itself account for the<br />
conflicting observations in experiments. In their experiments, they show that the<br />
common early disparity between WTP and WTA in auctions is not to be called into<br />
63
question. However, the gap ebbs away under the Vickrey‟s second price auction<br />
(SPA) and random nth-price auction (NPA) – see Section 2 for further details – while<br />
it lasts under the BDM mechanism, implying that the endowment effect can be<br />
eliminated with repetitions of some market mechanisms.<br />
Horowitz (2006a) states that the BDM framework could be used to assess<br />
public WTP for public projects, with the distribution of costs equal to the project<br />
costs; and other valuation mechanism should be used if the behavioral evidence<br />
shows that mechanisms are equivalent. Lusk and Rousu (2006) suggest that NPA is<br />
preferable to BDM if the researcher is looking for true valuation above all. Lusk et al.<br />
(2007) conclude in their study of payoff functions that BDM and NPA "provide<br />
relatively strong incentives for truthful bidding for all individuals regardless of the<br />
magnitude of their true WTP".<br />
Seeing that findings suggest that the auction mechanism per se accounts for<br />
the conflicting observations across market settings, Plott and Zeiler‟s (2005)<br />
conclusion that the results differ from unsound experimental procedures is<br />
incomplete.<br />
This chapter builds on Shogren et al.‟s (2001) results. Which auction<br />
mechanism is the best and fastest at reducing the gap? Which mechanism should be<br />
preferred over another? While Shogren et al. (1994) support Hanneman‟s results,<br />
assuming that the low substitution elasticity for the nonmarket good explains the<br />
WTA/WTP gap, they do not advocate the institution capable of properly valuing<br />
nonmarket goods. Likewise, Shogren et al. (2001) use only private goods to compare<br />
the influence of auction mechanisms. Only List (2003b) gives credit to the use of the<br />
random nth-price auction in valuing nonmarket private goods, but he does not state<br />
whether his results carry over to public goods.<br />
We aim at studying private valuation of public goods without direct<br />
substitutes, so we put realistic public goods such as the carbon offset, which can be<br />
attained via tree planting, into auctioning. Public goods have two defining<br />
characteristics: non-excludability and non-rivalry. Offsetting carbon emission helps<br />
prevent the effects of climate change; it is considered as a public good because, once<br />
provided, everyone can enjoy the benefits without adversely affecting anyone else‟s<br />
64
ability to do the same 10 . Rather then compulsory carbon trade, we institute voluntary<br />
trade to approach truthful valuation on both the bidder‟s (buyer‟s) and the offerer‟s<br />
(seller‟s) sides. On account of the common bias of nescience 11 in valuing unfamiliar<br />
or public goods, we remind the subjects that they are part of the milieu, which makes<br />
them indirectly and partly accountable for the current level of greenhouse gases, as<br />
they solicit industries to produce goods they are willing to consume at some<br />
environmental cost; in our case, it is the paper and energy used by students to achieve<br />
their education 12 .<br />
By means of repetitive auction mechanisms, the initial disparity between WTP<br />
and WTA can be removed. Nevertheless, we obtain different results from preceding<br />
studies, in a sense that only the BDM mechanism is able to remove the gap in later<br />
bidding rounds. SPA and NPA, which are also incentive-compatible, do not succeed<br />
in removing the disparity between bids and offers. Still, when we submit our<br />
experimental results to the exponential regression, we notice that in spite of a large<br />
early gap, NPA yields the highest speed of convergence to welfare indices‟ equality,<br />
suggesting that it contains strong incentives for rational behavior. In addition, we<br />
observe that subjects are strongly motivated by the subjective private benefit from<br />
funding the public good (either due to warm-glow 13 or to a concern for being<br />
formally identified as a contributor of the public good).<br />
The remainder of this chapter proceeds as follows. Section 2.2 describes the<br />
experimental design. Section 2.3 presents results and the analysis of data through<br />
standard and novel statistical tools. Section 2.4 provides discussion on how our<br />
results relate with existing work and present a new line of reasoning. We give some<br />
concluding remarks in Section 2.5.<br />
10 We in<strong>sur</strong>ed the public good characteristic by providing to every subject, after couple of weeks, an<br />
email feedback on the aggregate offset achievement.<br />
11 It reflects the absence of knowledge or the consideration that things are unknowable.<br />
12 The money released from trading (buying and non-selling) was sent to a non-governmental<br />
organization that launched a plantation of 1,404 Mangrove trees in Sumatra, Indonesia.<br />
13 Utility derived from the warm-glow (see Andreoni 1990) arises when the act itself of giving<br />
generates utility. It contrasts with the usual case where the individual only cares about the total amount<br />
of the carbon offset.<br />
65
2.2. The experimental design<br />
We want to evaluate the impact of three incentive-compatible auction<br />
mechanisms in the mea<strong>sur</strong>ement of WTP and WTA for a public good without<br />
substitutes. Our experiments were conducted during three sessions at the École<br />
Polytechnique. Different subjects took part in each of the three sessions (three types<br />
of auction mechanism) for a total of 102 participants, divided in three groups of<br />
subjects, which in turn were arbitrarily divided into two subgroups of buyers and<br />
sellers. Each subject received an identification number she filled in on each bid or<br />
offer, enabling her to be tracked whilst preserving her anonymity. The initial<br />
endowment distributed to the buyers was put forward to fund tree planting. On the<br />
WTP market-side, each buyer received EUR 15 and was asked to state her bid for a<br />
certificate of one ton of carbon offset (≤ EUR 15). If she won the bid, trees were<br />
planted in her name (this was acknowledged by a certificate). On the WTA market-<br />
side, each seller was given a certificate of one ton of carbon offset she could keep, in<br />
which case trees were planted in her name, or sell. If she decided to sell the certificate<br />
on the offer she stated (≤ EUR 15), no trees were planted. Subjects ignored that the<br />
cost of offsetting one ton of carbon in a five-year period was EUR 15, which enabled<br />
to plant 36 trees 14 .<br />
The parameters – recapitulated in the table below – of the experiments are the<br />
following: (i) 31 to 37 subjects participated per experiment; (ii) subjects were<br />
recruited among the voluntary students from the École Polytechnique 15 ; (iii) the good<br />
put in auctioning was a certificate of one ton of carbon offset; (iv) none information<br />
on price was provided; (v) subjects received an initial balance of EUR 15 or a<br />
certificate of one ton of carbon offset as an endowment; (vi) ten trials per experiment<br />
were unfolded, one of which was randomly selected as the binding trial; and (vii)<br />
BDM, SPA and NPA auction mechanisms were tested.<br />
14 In accordance with the system of reference applied by the non-governmental organization.<br />
15 Multi-cultural elite undergraduate students in science and engineering, salaried by the French<br />
Government. Their curriculum includes economics courses.<br />
66
Comments on the experimental protocol: our goal is to question auction<br />
mechanisms‟ influence on the gap between WTP and WTA, and not to divulge the<br />
gap itself, for we consider it as an established fact, so we decide to put an upper-<br />
bound on the sellers‟ choices in order to monitor which of the three market settings<br />
best replies to the early disparity. The bounds and endowments definitely create an<br />
anchoring effect, but there is no reason that it affects differently the three incentive-<br />
compatible mechanisms. Then, we publicly suggested to the subjects that revealing<br />
truthful preferences is a neutral strategy which will not penalize them. At last, we<br />
pooled all performed rounds in the mea<strong>sur</strong>ement of the gap.<br />
Market environment BDM SPA NPA<br />
Auctioned goods CO 2 offset certificate CO 2 offset certificate CO 2 offset certificate<br />
Initially endowment EUR 15 EUR 15 EUR 15<br />
Sellers‟ bound EUR 15 EUR 15 EUR 15<br />
Number of trials 10 10 10<br />
Retail price information None provided None provided None provided<br />
Optimal responses explained Suggested Suggested Suggested<br />
Practice round performed Pooled Pooled Pooled<br />
Subject participation Voluntary Voluntary Voluntary<br />
Number of subjects 37 34 31<br />
The Becker–DeGroot–Marschak mechanism (BDM)<br />
Becker, DeGroot, and Marschak (1964) introduce a mechanism under which<br />
buyers (respectively sellers) simultaneously state the highest (respectively lowest)<br />
amount they are willing to pay (respectively accept) for the good. In our experiment,<br />
each buyer and seller was asked to give, for each of the ten trials, independently and<br />
privately, her WTP or WTA by marking an "x" on a recording sheet that listed price<br />
intervals, such as in the following illustration. The price intervals ranged from EUR<br />
1–15, in increments of EUR 0.5. After collecting recording sheets from buyers and<br />
sellers, the monitor randomly selected one price from the list. If a buyer was willing<br />
to pay at least the random price for the certificate of one ton of carbon offset, she<br />
bought the item at that price. Otherwise, she did not buy the item. If a seller was<br />
67
willing to accept a price lower than or equal to the random price for the certificate of<br />
one ton of carbon offset, she sold the item at that price. Otherwise, she did not sell the<br />
item.<br />
I will buy (sell) I will not buy (sell)<br />
If the price is EUR 0.0 -- --<br />
If the price is EUR 0.5 -- --<br />
If the price is EUR 1.0 -- --<br />
If the price is EUR 1.5 -- --<br />
…<br />
If the price is EUR 14.0 -- --<br />
If the price is EUR 14.5 -- --<br />
If the price is EUR 15.0 -- --<br />
The random price, all bids and offers, and the number of buyers and sellers<br />
willing to buy and sell at the random price were made public after each trial. At the<br />
end of the experiment, one of the trials was randomly selected as the binding trial for<br />
the take-home pay.<br />
The second-price auction mechanism (SPA)<br />
Under the Vickrey (1961) second-price auction, bidders and offerers operated<br />
simultaneously. Buyers were asked to record, for each of the ten trials, privately and<br />
independently, the maximum they were willing to pay for the certificate of one ton of<br />
carbon offset. In this case, buyers wrote a numerical value on the recording sheet. The<br />
monitor collected values and, after each trial, made all bids public, as well as the<br />
identification number of the highest bidder and the market-clearing price (second<br />
highest bid). The monitor gave each seller a certificate of one ton of carbon offset.<br />
For each trial, sellers wrote their minimum WTA to sell the certificate. After each<br />
trial, the monitor publicly diffused all offers, the identification number of the lowest<br />
offerer and the market-clearing price (second lowest offer). Like with BDM, after the<br />
tenth trial, the monitor randomly selected one of the trials as the binding trial for the<br />
take-home pay for both buyers and sellers.<br />
68
The random nth-price auction mechanism (NPA)<br />
The random nth-price auction is conducted as follows (bidders and offerers<br />
operate simultaneously): (i) for each trial, each bidder submits a bid (resp. an offer)<br />
on a recording sheet; (ii) all bids are ranked from lowest to highest, all offers are<br />
ranked from highest to lowest; (iii) the monitor selects a random number n� �2, N�<br />
with N the number of bidders; (iv) the n � 1 buyers who made the highest bids buy the<br />
certificate of one ton of carbon offset at the nth-price and the n � 1 sellers who made<br />
the lowest offers sell the certificate of one ton of carbon offset at the nth-price. The<br />
value of n, all bids and offers, the buying and selling price, and the number of buyers<br />
and sellers willing to buy and sell at the random price, are made public after each<br />
trial. Once again, after the tenth trial, the monitor randomly selects one of the trials as<br />
the binding trial for the take-home pay for both buyers and sellers.<br />
The BDM, SPA and NPA mechanisms are incentive-compatible. It is not in a<br />
buyer‟s interest to understate her WTP; if the random buying price falls between the<br />
stated WTP and the true WTP, the buyer foregoes a beneficial trade. It is also not in a<br />
buyer‟s interest to overstate true WTP; if the random buying price is greater than the<br />
true value but less than the stated value, the buyer is required to buy the good at a<br />
price greater than her true WTP. The reasoning is identical for the seller.<br />
A complementary remark on NPA can be made. Contrary to SPA, subjects<br />
have a nonnegative probability of winning the auction, which engages off-margin<br />
bidders and offerers who usually consider that they will be excluded from the market.<br />
As well, the endogenously determined market-clearing price prevents bidders and<br />
offerers from using the random market-clearing price as an indicator.<br />
2.3. The results<br />
Table 2.1. presents the summary statistics of the experimental results under<br />
BDM, SPA and NPA. In all experiments, bidding behavior in the initial trial does<br />
69
Auction Value mea<strong>sur</strong>e<br />
Table 2.1. Summary statistics of the BDM, SPA and NPA mechanisms<br />
H0: Mean WTP – Mean WTA = 0; H1: Mean WTP – Mean WTA < 0<br />
a t-test: reject H0 at the 5% level<br />
Trial<br />
1 2 3 4 5 6 7 8 9 10<br />
BDM WTP Mean 6.18 7.11 7.82 8.11 8.29 8.66 8.39 8.71 8.82 8.61<br />
N=19 Median 5.00 5.50 6.50 6.50 7.00 7.00 7.00 7.50 7.50 7.50<br />
Variance 12.51 15.52 15.39 15.43 15.09 15.86 15.27 14.62 14.37 17.74<br />
WTA Mean 10.53 9.47 9.56 8.42 8.92 8.69 9.53 9.19 8.67 8.06<br />
N=18 Median 10.00 10.00 10.00 8.75 9.50 9.75 10.00 10.00 9.75 8.25<br />
Variance 6.07 12.34 18.03 18.60 20.95 21.53 19.75 16.86 17.79 20.97<br />
Ratio of mean WTA/WTP 1.70 1.33 1.22 1.04 1.08 1.00 1.13 1.06 0.98 0.94<br />
t-test of means a –3.85 –1.46 –0.83 0.27 0.06 0.46 –0.39 0.09 0.58 0.91<br />
SPA WTP Mean 3.47 3.91 4.69 5.43 5.68 5.71 6.01 6.50 5.46 6.59<br />
N=17 Median 3.00 4.10 5.00 5.60 5.80 6.05 7.00 7.00 7.00 7.00<br />
Variance 9.64 6.68 5.52 5.42 6.15 7.71 8.86 14.50 12.56 10.04<br />
WTA Mean 10.66 8.74 8.47 9.07 8.59 9.82 9.40 8.32 9.52 9.23<br />
N=17 Median 10.00 9.00 8.00 9.00 7.00 10.00 8.00 8.00 8.00 8.00<br />
Variance 16.60 19.56 14.03 22.27 20.72 29.45 29.44 32.86 26.44 30.86<br />
Ratio of mean WTA/WTP 3.07 2.23 1.81 1.67 1.51 1.72 1.57 1.28 1.75 1.40<br />
t-test of means a –5.28 –3.41 –3.06 –2.35 –1.78 –2.30 –1.78 –0.59 –2.21 –1.20<br />
NPA WTP Mean 3.97 3.98 4.77 4.93 4.77 5.19 6.18 6.12 6.85 6.72<br />
N=15 Median 2.50 4.00 5.00 5.12 5.14 5.01 7.00 6.50 7.00 7.26<br />
Variance 12.67 6.92 4.83 4.30 5.40 6.33 5.81 6.54 7.77 10.03<br />
WTA Mean 10.75 10.52 10.29 10.22 9.86 9.05 9.17 9.14 9.23 9.37<br />
N=16 Median 10.50 10.00 9.74 9.65 8.77 8.50 8.49 8.35 8.09 8.50<br />
Variance 10.19 6.99 6.32 9.46 10.31 13.75 16.67 13.30 14.08 20.64<br />
Ratio of mean WTA/WTP 2.71 2.64 2.16 2.07 2.07 1.74 1.48 1.49 1.35 1.39<br />
t-test of means a –5.06 –6.45 –6.21 –5.17 –4.60 –2.87 –1.90 –2.10 –1.40 –1.33<br />
70
not contradict the endowment effect: the mean offer WTA 16 is significantly<br />
greater than the mean bid WTP 17 . Still, with experience gained through repetitive<br />
auctioning under the BDM mechanism, WTA offers decrease and WTP bids<br />
increase over time 18 . The WTA / WTP ratios thus decline throughout the ten<br />
trials, falling from 1.70 in trial 1 to 0.94 in trial 10 (see Fig 2.1.), which<br />
corresponds to WTP increase of 39% and WTA decrease of 23%. Concerning<br />
variances, we notice that the dispersion around the mean increases for both WTP<br />
(42%) and WTA (245%) from trial 1 to trial 10. In trials 4–10, a t-test shows that<br />
we cannot reject the null hypothesis that WTP and WTA come from the same<br />
distribution at the p
3,50<br />
3,00<br />
2,50<br />
2,00<br />
1,50<br />
1,00<br />
0,50<br />
1 2 3 4 5 6 7 8 9 10<br />
Fig. 2.1. WTA / WTP disparity from trial 1 to trial 10<br />
Let us now take a further insight in our results and those of the mug<br />
experiments from Shogren et al. (2001). At first sight, we obtain contradictory<br />
results. In our experiment, the gap disappears under BDM, whereas in theirs,<br />
BDM is the only mechanism unable to remove the early gap.<br />
Our findings show that repetitions under the BDM mechanism can remove<br />
the endowment effect, as long as it steers people‟s behavior. Likewise, they<br />
suggest that the auction mechanism per se can account for the conflicting<br />
observations, as we clearly observe different paths of equalization of WTP and<br />
WTA . We introduce an innovative tool to study the path of gap removal: the<br />
exponential regression on the WTA / WTP ratios.<br />
An exponential regression is of a form<br />
72<br />
BDM<br />
SPA<br />
NPA<br />
ax<br />
y � be with x the variable along<br />
the x-axis, y the regressed values of WTA / WTP , a the amplitude of the<br />
decrease (or speed of convergence to equality) and b the y-intercept of regression.<br />
The function is based on the function linear regression, with the y-axis<br />
logarithmically scaled. R-square gives information on the exponential relationship<br />
between ratios.<br />
We apply this method to Shogren et al.‟s (2001) mug experiments (see<br />
Fig. 2.2.) and to our experiments (see Fig. 2.3.). The exponential regression is
used for two reasons: first, it allows observing phenomena with rapid variations,<br />
such as in our experiments; second, it allows observing the decrease path to<br />
equality, that is, the way ratios tend to one. We try to unearth the mechanism able<br />
to remove the gap as quickly as possible, whatever the initial ratio. We can thus<br />
consider the highest coefficient of decrease as the highest speed of convergence to<br />
welfare indices‟ equality (see Table 2.2.).<br />
3,50<br />
3,00<br />
2,50<br />
2,00<br />
1,50<br />
1,00<br />
0,50<br />
3,5<br />
3<br />
2,5<br />
2<br />
1,5<br />
1<br />
0,5<br />
0 1 2 3 4 5 6 7 8 9 10<br />
Fig. 2.2. Exponential regression of WTA / WTP disparity<br />
from Shogren et al.’s (2001) mug experiments<br />
0 1 2 3 4 5 6 7 8 9 10<br />
Fig. 2.3. Exponential regression of WTA / WTP disparity<br />
73<br />
BDM<br />
SPA<br />
NP A<br />
Expon. (NP A)<br />
Expon. (S P A)<br />
Expon. (BDM)<br />
BDM<br />
SPA<br />
NP A<br />
Expon. (S P A)<br />
Expon. (BDM)<br />
Expon. (NP A)
Shogren et al.‟s (2001) data from BDM provides no exponential<br />
relationship between sequential ratios, but ours does. Although the y-intercept of<br />
regression starts with the same value (both 1.5), the gap disappears in our<br />
experiment (illustrated by the speed of convergence –0.04) but stays stationary in<br />
the mug experiment (null speed of convergence).<br />
We find in both data that NPA provides the best exponential relationship<br />
between ratios (respectively<br />
2<br />
R � 0.95 and<br />
74<br />
2<br />
R � 0.96 ) and the highest speed of<br />
convergence to equality (respectively –0.08 and –0.12) in time. Under SPA, the<br />
exponential relationship between ratios (respectively<br />
2<br />
R � 0.61 and<br />
2<br />
R � 0.63)<br />
and the speed of convergence to equality (respectively –0.06 and –0.09) are<br />
significant but lower. Sudden leaps of increase of the WTA / WTP ratio under<br />
SPA – due to off-margin bidders – explain the differences in<br />
2<br />
R with regard to<br />
NPA. It is worthwhile noticing that SPA comes out as the "worst" active market<br />
mechanism even though it is frequently used in experimental environments to<br />
reveal agents‟ preferences. Under BDM, our experiment and Shogren et al.‟s<br />
(2001) experiment both obtain the lowest results in terms of exponential<br />
relationship 19 and speed of convergence to equality. Therefore, the orderings of<br />
convergence in our experiments and those of Shogren et al.‟s (2001) are alike.<br />
Table 2.2. Exponential regression statistics<br />
Auction Regression statistics Our experiments<br />
Shogren et al.’s<br />
mug experiments<br />
BDM Speed of convergence (a) –0.04 –0.00<br />
y-intercept of regression (b) 1.5 1.5<br />
R-square 0.69 0.00<br />
SPA Speed of convergence (a) –0.06 –0.09<br />
y-intercept of regression (b) 2.5 1.9<br />
R-square 0.61 0.63<br />
NPA Speed of convergence (a) –0.08 –0.12<br />
y-intercept of regression (b) 2.9 2.8<br />
R-square 0.95 0.96<br />
19 The low exponential factor with the BDM is partially explained by the initial smaller difference<br />
between WTP and WTA.
If the initial gap is due to the choice of the market mechanism, then the<br />
choice of BDM is appropriate, for it produces the smallest initial gap. But, if we<br />
are to urge the auction mechanism able to rapidly deflate an excessive initial<br />
WTA / WTP gap in a market-clearing price setting, we suggest the use of NPA<br />
which involves most of the bidders in the auctioning. Indeed, as for the model of<br />
exponential regression, the BDM mechanism would not have equalized the<br />
welfare indices if the starting ratio were more of SPA or NPA‟s magnitude. This<br />
appears all the more sound, provided the BDM mechanism is a passive market-<br />
like setting with only minor adjustments in bidding behavior 20 . Indeed, NPA<br />
applies competitive pres<strong>sur</strong>e to the participating bidders. A bidder cannot avoid<br />
acting strategically since her best bid depends on the competing bids. By bidding<br />
more aggressively, the bidder improves her chances of winning the auction. As far<br />
as SPA is concerned, the unevenness in the decrease of the gap jeopardizes its<br />
robustness.<br />
2.4. Discussion<br />
Our results support the standard thesis that market mechanisms can<br />
remove or at least sturdily reduce the initial disparity between WTP and WTA.<br />
However, some points need to be clarified.<br />
Let us first focus on the specificity of the good in sale. Under NPA and<br />
SPA, the number of traded tons of carbon offset in a period is independent of the<br />
bids and offers submitted by the subjects. In any case, in SPA, one ton of carbon<br />
offset is bought and sold; in NPA, n � 1 tons of carbon offset are bought and sold.<br />
As a result, free-riding is likely to occur, since a subject‟s bid cannot affect the<br />
total public good provision while it affects her payment (buying a certificate is<br />
costly). On the contrary, under BDM, subjects‟ choices affect the total provision<br />
of public good. Indeed, if a seller chooses a minimum selling price higher than the<br />
randomly selected price, she will keep her certificate and one more ton of carbon<br />
20 See footnote 7 in Shogren et al. (2001).<br />
75
will be offset. The same reasoning applies for buyers. Put differently, subjects<br />
know they can influence the amount of carbon offset under BDM as their<br />
probability of winning the right to buy one certificate is independent of other<br />
bidders: the higher the private bid, the higher the chances that a ton of carbon is<br />
offset.<br />
This difference between BDM on the one side and NPA and SPA on the<br />
other side allows identifying two distinct motivations for funding the public good.<br />
First, there is "the public good motivation": a subject wants to buy or keep a<br />
certificate because it allows offsetting one ton of carbon for the community.<br />
Second, there is "the private good motivation": a subject wants to buy or keep a<br />
certificate because she wants to own a certificate and be associated to the<br />
offsetting even though this does not change the number of tons of carbon offset<br />
(she either wants to derive a warm-glow from altruism or wants to gain social<br />
status through the public good funding). Individuals often provide more public<br />
goods than traditional economic theory predicts. Public goods are then considered<br />
as impure public goods, which are products or services that combine both public<br />
and private benefits.<br />
In BDM, both motivations for funding the public good are present,<br />
whereas in NPA and SPA, only the private good motivation is present. Let us<br />
consider s – the mean value of all bids (WTP) and offers (WTA) – as the mean<br />
value of the public good. After computation, we observe that s over the ten<br />
rounds is strictly higher with BDM (8.57) than with SPA (7.26) or NPA (7.63).<br />
Locally, at the last period, the values are respectively 8.34, 7.91 and 8.09. These<br />
results indicate that the private good motivation is extreme compared to the public<br />
good motivation, i.e. subjects are mainly paying for enjoying warm-glow or being<br />
identified as contributors of the carbon offsetting. If we take s of BDM as a<br />
benchmark value of the public good, the <strong>sur</strong>plus of the BDM value compared to<br />
SPA and NPA values corresponds to the value of the public good motivation<br />
which lies in the interval �0.94, 1.31 �.<br />
Since the public benefit for an individual is<br />
negligible, individuals mostly derive some private benefit from the public good.<br />
76
These results are thus consistent with microeconomic analysis, where the private<br />
benefit governs the decisions of economic agents.<br />
Contrary to the observations where repeat-play public good games produce<br />
declining contribution over time (see Andreoni (1988) and Caldas et al. (2003)),<br />
s is increasing in our experiments. As a matter of fact, if we regress s with the<br />
number of periods, we obtain a small but strictly positive correlation coefficient<br />
(BDM: 0.18; SPA: 0.13; NPA: 0.15). In standard public good games, the fall is<br />
motivated by free-riding and discouragement of high type players to pursue alone<br />
the provision of public good. We propose two explanations for the rise we<br />
observed. First, the funded public good does not only concern the subjects but the<br />
population outside the experiment. Therefore, the free-riding attitude of some<br />
participants does not alter subjects‟ motivations since they do not specifically<br />
contribute for these free-riders (while they do in public good games). Second, as<br />
already mentioned, the private good motivation outperforms the public good<br />
motivation, which also explains the absence of the usual decline in subjects‟<br />
bids 21 .<br />
For all these reasons, we decide to focus on the private value dimension of<br />
the public good in the following discussion.<br />
Contrary to NPA or SPA, the initial gap under the BDM mechanism is<br />
closer to one in both Shogren et al.‟s (2001) and our experiments. As WTA is<br />
similar under the three auction mechanisms in the first trial, this observation<br />
comes from a high starting WTP under BDM, i.e. shorter distance to cover from<br />
bids to offers. Given that BDM and NPA both share the properties of incentive-<br />
compatibility and the possibility for every bidder to offset a ton of carbon, the<br />
explanation could come from the unambiguous distribution of prices and payoffs<br />
under the BDM mechanism, whereas under NPA there is ambiguity in view of the<br />
unknown bids of the opponents (see Sarin and Weber (1993)).<br />
Another explanation could come from the theory of disappointment<br />
aversion. In a recent article, Horowitz (2006b) relates that under BDM an<br />
21 One could argue that bids increased because of the house money effect. However, Clark (2002)<br />
finds no evidence of it in a public good experiment.<br />
77
individual may report a higher value than the true one, simply because she is more<br />
disappointed from not receiving the good than from receiving it at a relatively<br />
high price, which induces her to report a higher bid to increase the chance of<br />
winning the auction. This could explain why the bids under BDM started higher<br />
earlier than the bids under NPA and SPA; subjects knew from the very beginning<br />
that they were bidding against a market-clearing price issued from a known<br />
ceiling market-clearing price.<br />
Let us also mention a proposition from Milgrom and Weber (1982) that<br />
could be spoken for our results. The authors state that common uncertainty about<br />
the value of a good creates affiliated private values, especially in case of<br />
unfamiliar goods. This is because early trials send information from which high-<br />
bidders induce low-bidders to revise their preferences and increase their bids, the<br />
logic being that there are some common but unknown characteristics of the item<br />
released with bids. Our experimental protocol does not permit to validate or<br />
invalidate this hypothesis, but we can specify that all subjects received the same<br />
amount of information on the nature of the unfamiliar good before the auction<br />
took place 22 . Although the mimesis phenomenon could explain rising low bids<br />
under SPA and NPA just after the start-off, our BDM experiment shows higher<br />
early bids; therefore, the logic of common uncertainty could only relate to the<br />
latter bidding rounds. Moreover, the dispersion around the mean from trial 1 to<br />
trial 10 increases in all experiments, partly refuting the argument of affiliated<br />
private values. The only case of dispersion fall that could challenge independent<br />
values‟ validity deals with WTP in the NPA mechanism.<br />
2.5. Concluding remarks<br />
We examined three mechanisms that could rectify the initial gap between<br />
WTP and WTA in the trading of a public good. From simple observations of the<br />
disparity ratios, we observe different results from Shogren et al. (2001) and can<br />
22 The market price effect, implied by affiliated private values, disappears when bidders receive<br />
nonprice information about the good before the experience is conducted (List and Shogren 1999).<br />
78
conclude either that their findings – which suggest the validity of SPA and NPA<br />
in valuing private goods – are local, or that the public goods are subject to a<br />
different bidding behavior.<br />
We think that under a quasi-market setting such as the BDM mechanism<br />
subjects understood the fact that they could influence the level of the public good<br />
and behaved accordingly. In active markets with endogenous market-clearing<br />
prices such as NPA, no subject could influence the level of the public good which<br />
acted as a disincentive to augment the level of public good. Our results show a<br />
disparity dropped with repetition under the three mechanisms, suggesting that the<br />
economic theory of rationality within markets operates. And yet, the theory<br />
implies a perfect equality between WTP and WTA, which seems not to be<br />
guaranteed when funding a public good. Research must deal with this.<br />
Value mea<strong>sur</strong>es approached equality principally for the reason that bids<br />
considerably increased throughout trials. Since offers moderately decreased in<br />
time, signifying a modest remedy to loss aversion, we could think of markets as<br />
systems which lift the subjects‟ regret not to acquire the good. Two-sided market<br />
value would then be somewhere between the behavioral exaggerations of loss<br />
aversion and disappointment aversion. These unforeseen questions necessitate<br />
further research.<br />
In addition, more experimental research on private and public values of a<br />
public good should be conducted. For example, we could identify more accurately<br />
the private good and public good motivations by explicitly insisting on the fact<br />
that bids cannot affect the size of the provision of public goods in NPA and SPA.<br />
As well, we could conduct experiments where subjects would be purposely<br />
deprived from any proof of having financed the public good; that way, we could<br />
distinguish between the desire to finance the public good and the desire to be<br />
identified by others as a generous contributor to the public good.<br />
2.6. References<br />
Andreoni, J. (1988), “Why free-ride? Strategies and Learning in Public Goods<br />
Experiments”, Journal of Public Economics, 37: 291–304.<br />
79
Andreoni, J. (1990), “Impure Altruism and Donations to Public Goods: A Theory<br />
of Warm-Glow Giving?”, Economic Journal, Royal Economic Society, 100:<br />
464–477.<br />
Becker, G., DeGroot, M. and Marschak, J. (1964), “Mea<strong>sur</strong>ing Utility by a Single<br />
Response Sequential Method”, Behavioral Science, 9: 226–232.<br />
Bohm, P., Linden, J. and Sonnegard, J. (1997), “Eliciting Reservation Prices,<br />
Becker-DeGroot-Marschak Mechanisms vs. Markets”, Economic Journal,<br />
107: 1079–1089.<br />
Brouwer, R., Bateman, I., Saunders, C. and <strong>La</strong>ngford, I. (1999), “Perception and<br />
valuation of risk reduction as a public and private good: Investigating<br />
methodological issues in contingent valuation of UV risks in New Zealand”,<br />
CSERGE Working Paper GEC 99-06.<br />
Caldas, J., Rodgrigues, J. and Carvalho, L. (2003), Economics and social<br />
psychology on public goods: experiments and explorations, Dinâmia Working<br />
Paper, 2003/30.<br />
Cherry, T., Crocker, T. and Shogren, J. (2003), “Rationality Spillovers”, Journal<br />
of Environmental Economics and Management, 45: 63–84.<br />
Clark, J. (2002), “House Money Effects in Public Good Experiments”,<br />
Experimental Economics, 5: 223–231.<br />
Coase, R. (1960), “The Problem of Social Cost”, Journal of <strong>La</strong>w and Economics,<br />
3: 1–44.<br />
Coppinger, V., Smith, V. and Titus, J. (1980), “Incentives and Behavior in<br />
English, Dutch and Sealed-Bid Auctions”, Economic Inquiry, 18: 1–22<br />
Gode, D. and Sunder, S. (1993), “Allocative Efficiency of Markets with Zero-<br />
Intelligence Traders: Market as a Partial Substitute for Individual Rationality”,<br />
Journal of Political Economy, 101: 119–137.<br />
Hanemann, W. (1991), “Willingness to Pay and Willingness to Accept: How<br />
Much Can They Differ?”, American Economic Review, 81: 635–647.<br />
Horowitz, J. (2006a), “What Do We Know About the Performance of the Becker-<br />
DeGroot-Marschak Mechanism”, available at<br />
http://faculty.arec.umd.edu/jhorowitz/BDM-Empirics-1.doc.<br />
80
Horowitz, J. (2006b), “The Becker-DeGroot-Marschak Mechanism Is Not<br />
Necessarily Incentive Compatible, Even for Non-Random Goods”, Economics<br />
Letters, 93: 6–11.<br />
Kahneman, D., Knetsch, J. and Thaler, R. (1990), “Experimental Test of the<br />
Endowment Effect and the Coase Theorem”, Journal of Political Economy,<br />
98: 1325–1348.<br />
List, J. and Shogren, J. (1999), “Price Information and Bidding Behavior in<br />
Repeated Second-Price Auctions”, American Journal of Agricultural<br />
Economics, 81: 942–49.<br />
List, J. (2003a), “Does Market Experience Eliminate Market Anomalies?”,<br />
Quarterly Journal of Economics, 118: 41–71.<br />
List, J. (2003b), “Using random nth-Price Auctions to value Non-Market Goods<br />
and Services”, Journal of Regulatory Economics, 23: 193–205.<br />
Lusk, J. and Rousu, M. (2006), “Market Price Endogeneity and Accuracy of<br />
Value Elicitation Mechanisms”, in List, J. (ed.), Using Experimental Methods<br />
in Environmental and Resource Economics, Northhampton, MA: Edward<br />
Elgar Publishing.<br />
Lusk, J., Alexander, C. and Rousu, M. (2007), “Designing Experimental Auctions<br />
For Marketing Research: Effect Of Values, Distributions, And Mechanisms<br />
On Incentives For Truthful Bidding”, Review of Marketing Science, 5: 1–30.<br />
Milgrom, P. and Weber, R. (1982), “A theory of auctions and competitive<br />
bidding”, Econometrica, 50: 1089–1122.<br />
Plott, C. (1996), “Rational individual behavior in markets and social choice<br />
processes: the Discovered Preference Hypothesis”, in Arrow, K., Colombatto,<br />
E., Perleman, M. and Schmidt, C., (eds.), Rational Foundations of Economic<br />
Behavior, Macmillan and St. Martins London.<br />
Plott, C. and Zeiler, K. (2005), “The Willingness to Pay/Willingness to Accept<br />
Gap, the Endowment Effect, Subject Misconceptions and Experimental<br />
Procedures for Eliciting Valuations”, American Economic Review, 95: 530–<br />
545.<br />
Randall, A. and Stoll, J. (1980), “Consumer‟s Surplus in Commodity Space”,<br />
American Economic Review, 71: 449–457.<br />
Sarin, R. and Weber, M. (1993), “Effects of Ambiguity in Market Experiments”,<br />
Management Science, 39: 609–615.<br />
81
Shogren, J., Shin, S., Hayes, D. and Kliebenstein, J. (1994), “Resolving<br />
Differences in Willingness to Pay and Willingness to Accept”, American<br />
Economic Review, 84: 255–270.<br />
Shogren, J., List, J. and Hayes, D. (2000), “Preference Learning in Consecutive<br />
Experimental Auctions”, American Journal of Agricultural Economics, 83:<br />
1016–1021.<br />
Shogren, J., Cho, S., Koo, C., List, J., Park, C., Polo, P. and Wilhelmi, R. (2001),<br />
“Auction Mechanisms and the Mea<strong>sur</strong>ement of WTP and WTA”, Resource<br />
and Energy Economics, 23: 97–109.<br />
Smith, V. (1991), “Rational Choice: The Contrast between Economics and<br />
Psychology”, Journal of Political Economy, 99: 877–897.<br />
Vickrey, W. (1961), “Counterspeculation, Auctions, and Competitive Sealed<br />
Tenders”, Journal of Finance, 16: 8–37.<br />
82
2.7. Appendix<br />
GENERAL INSTRUCTIONS (translated from French)<br />
You are about to participate in an experiment about decision making. You are not<br />
allowed to speak to your neighbors during the experiment.<br />
All human activities release greenhouse gases, including CO2, that provoke the<br />
global warming. This warming endangers the planet, its inhabitants, its<br />
ecosystems and biodiversity. One way to fight against global warming is to plant<br />
trees. The key elements are the following: the forested <strong>sur</strong>faces are a carbon trap;<br />
young forests store much more carbon than old forests, for trees absorb CO2 as<br />
they grow; forests preserve plant and animal biodiversity.<br />
An NGO has launched a project of carbon offsetting by funding the reforestation<br />
projects. The purpose is to offset carbon emissions by buying off your own<br />
emissions. The compensation is acknowledged by a certificate of one ton of<br />
carbon offset.<br />
During your education at the École Polytechnique, you have received and printed,<br />
and will certainly do it over in the future, number of documents required for your<br />
schoolwork; it is also the case with your consumption of energy (such as light,<br />
heating, power supply for computers, etc.) Because you are contributing to the<br />
emissions through your consumption of paper and energy via your indirect<br />
demand for their manufacturing and distribution, we want to value your<br />
willingness to buy off your CO2 emissions.<br />
To this end, we will use a mechanism of purchasing and selling certificates of one<br />
ton of CO2 offset, such as the ones we currently hold in our hands.<br />
In couple of weeks, we will get in touch with you by email to inform you about<br />
the number of offset tons of CO2 according to your decisions.<br />
We will now conduct an experiment. As you came into the class, some of you<br />
were designated as sellers while others were designated as buyers. Indeed, each of<br />
you randomly drew a number which decided between buyer and seller. Please<br />
keep this number until the end of the experiment: it will serve us to track you on<br />
the information cards. In the end of the experiment, during the imbursement,<br />
please give us back your numbers.<br />
Only one trial will be binding. We will repeat the experiment ten times. After the<br />
tenth trial, the youngest person in the room will randomly draw a number between<br />
1 and 10, which will designate the binding trial.<br />
Please feel free to interrupt us and ask any question you might have in mind.<br />
83
Without further delay, we are going to read you the instructions concerning the<br />
conduct of the experiment. Let‟s start with those of you who are buyers.<br />
RANDOM NTH-PRICE AUCTION<br />
Buyers<br />
You own €15. You can now participate in an auction in order to buy a certificate<br />
of one ton of CO2 offset. If that is your wish, please submit a bid. The bid you<br />
submit can range between €0 and €15. If you decide to buy the certificate, trees<br />
which are planted on your behalf (acknowledged by your name on the certificate)<br />
will compensate one ton of CO2.<br />
To submit a bid, please specify on the information card the price at which<br />
you’re willing to buy the certificate.<br />
Rules: your bid is ranked among all bids. We randomly select a number between 2<br />
and n (n being the total number of offers). In other words, we randomly draw one<br />
of the bids and look at its rank. You buy a certificate, at the nth price, if your bid<br />
is contained in n–1 highest bids.<br />
Example: twenty bids are submitted. We randomly draw seven, that is, the<br />
seventh-highest bid in the increasing order. You buy a certificate at a displayed<br />
price (seventh-highest bid) if your bid is contained in the six highest bids.<br />
Nota bene: the higher your bid, the higher your chances of buying the certificate.<br />
If your bid is randomly drawn, your bid becomes the displayed price imposed to<br />
the n–1 highest bidders. Since you ignore the displayed price ex ante, giving your<br />
own value of one ton of CO2 offset enables you to buy the certificate if your value<br />
is higher than the displayed price, and prevents you from buying otherwise.<br />
Sellers<br />
You own a certificate of one ton of CO2 offset. You can now participate in an<br />
auction in order to sell your certificate. If that is your wish, please submit an offer.<br />
The offer you submit can range between €0 and €15. If you decide to sell the<br />
certificate with your name on, no ton of CO2 will be offset.<br />
To submit an offer, please specify on the information card the price at which<br />
you’re willing to sell the certificate.<br />
Rules: your offer is ranked among all offers. We randomly select a number<br />
between 2 and n (n being the total number of offers). In other words, we randomly<br />
draw one of the offers and look at its rank. You sell a certificate, at the nth price,<br />
if your offer is contained in n–1 lowest offers.<br />
84
Example: twenty offers are submitted. We randomly draw six, that is, the sixthlowest<br />
offer in the decreasing order. You sell your certificate at a displayed price<br />
(sixth-lowest offer) if your offer is contained in the five lowest offers.<br />
Nota bene: the lower your offer, the higher your chances of selling the certificate.<br />
If your offer is randomly drawn, your offer becomes the displayed price imposed<br />
on the n–1 lowest offers. Since you ignore the displayed price ex ante, giving your<br />
own value of one ton of CO2 offset enables you to sell the certificate if the price is<br />
higher than your value, and prevents you from selling otherwise.<br />
SECOND-PRICE AUCTION<br />
Buyers<br />
You own €15. You can now participate in an auction in order to buy a certificate<br />
of one ton of CO2 offset. If that is your wish, please submit a bid. The bid you<br />
submit can range between €0 and €15. If you decide to buy the certificate, trees<br />
which are planted on your behalf (acknowledged by your name on the certificate)<br />
will compensate one ton of CO2.<br />
To submit a bid, please specify on the information card the price at which<br />
you’re willing to buy the certificate.<br />
Rules: where you‟ve decided to participate in the auction, your offer to purchase is<br />
ranked among all offerings purchase. Offerings are classified in ascending order.<br />
You take the bid if your offer is highest. However, you only pay for the certificate<br />
that the amount of the second offers the highest.<br />
Example: ten bids are submitted. The highest bid is €13. The second highest bid is<br />
€11. The bidder who proposed €13 buys the certificate and pays €11.<br />
Nota bene: the higher your bid, the higher your chances of buying the certificate.<br />
Since you ignore the displayed price ex ante, giving your own value of one ton of<br />
CO2 offset enables you to buy the certificate if your value is higher than the<br />
displayed price, and prevents you from buying otherwise.<br />
Sellers<br />
You own a certificate of one ton of CO2 offset. You can now participate in an<br />
auction in order to sell your certificate. If that is your wish, please submit an offer.<br />
The offer you submit can range between €0 and €15. If you decide to sell the<br />
certificate with your name on, no ton of CO2 will be offset.<br />
To submit an offer, please specify on the information card the price at which<br />
you’re willing to sell the certificate.<br />
85
Rules: your offer to sell is ranked among all offers. Offers are ranked in a<br />
descending order. You sell a certificate if your offer is the lowest, and you sell it<br />
at a displayed price, that is, the second-lowest offer price.<br />
Example: ten offers are submitted. The lowest offer is €5. The second lowest offer<br />
is €7. The seller who proposes €5 sells her certificate and earns €7.<br />
Nota bene: the lower your offer, the higher your chances of selling the certificate.<br />
Since you ignore the displayed price ex ante, giving your own value of one ton of<br />
CO2 offset enables you to sell the certificate if the price is higher than your value,<br />
and prevents you from selling otherwise.<br />
BDM MECHANISM<br />
Buyers<br />
You own 15 €. You can now participate in an auction in order to buy a certificate<br />
of one ton of CO2 offset. If that is your wish, please submit a bid. The bid you<br />
submit can range between €0 and €15. If you decide to buy the certificate, trees<br />
which are planted on your behalf (acknowledged by your name on the certificate)<br />
will compensate one ton of CO2.<br />
To submit a bid, please fill in the following table and mark an “X” for each<br />
price at which you’re (and are not) willing to buy the certificate.<br />
Rules: your maximum bid is ranked among all bids. We randomly select one price<br />
from the price list, which becomes the displayed price. You buy a certificate if<br />
your bid is higher than or equal to the displayed price.<br />
Example: We randomly draw €6. Since your bid is higher than or equal to €6, you<br />
buy the certificate and pay €6.<br />
I will buy I will not buy<br />
If the price is €0 X<br />
If the price is €0.5 X<br />
If the price is €1.0 X<br />
… X<br />
If the price is €8.5 X<br />
If the price is €9 X<br />
If the price is €9.5 X<br />
… X<br />
If the price is €14.0 X<br />
If the price is €14.5 X<br />
If the price is €15.0 X<br />
86
Nota bene: the higher your bid, the higher your chances of buying the certificate.<br />
Since you ignore the displayed price ex ante, giving your own value of one ton of<br />
CO2 offset enables you to buy the certificate if your value is higher than the<br />
displayed price, and prevents you from buying otherwise.<br />
Sellers<br />
You own a certificate of one ton of CO2 offset. You can now participate in an<br />
auction in order to sell your certificate. If that is your wish, please submit an offer.<br />
The offer you submit can range between €0 and €15. If you decide to sell the<br />
certificate with your name on, no ton of CO2 will be offset.<br />
To submit an offer, please fill in the following table and mark an “X” for<br />
each price at which you’re (and are not) willing to sell the certificate.<br />
Rules: your minimum offer is ranked among all offers. We randomly select one<br />
price from the price list, which becomes the displayed price. You sell a certificate<br />
if your offer is lower than or equal to the displayed price.<br />
Example: We randomly draw €10. Since your offer is lower than or equal to €10,<br />
you sell the certificate and earn €10.<br />
I will sell I will not sell<br />
If the price is €15.0 X<br />
If the price is €14.5 X<br />
If the price is €14.0 X<br />
… X<br />
If the price is €5.0 X<br />
If the price is €4.5 X<br />
If the price is €4.0 X<br />
… X<br />
If the price is €1.0 X<br />
If the price is €0.5 X<br />
If the price is €0.0 X<br />
Nota bene: the lower your offer, the higher your chances of selling the certificate.<br />
Since you ignore the displayed price ex ante, giving your own value of one ton of<br />
CO2 offset enables you to sell the certificate if the price is higher than your value,<br />
and prevents you from selling otherwise.<br />
87
89<br />
Chapter 3<br />
Endogenous Market-Clearing Prices<br />
and Reference Point Adaptation<br />
Abstract<br />
When prices depend on the submitted bids, i.e. with endogenous market-clearing<br />
prices in repeated-round auction mechanisms, the assumption of independent private<br />
values that underlines the property of incentive-compatibility is to be brought into<br />
question; even if these mechanisms provide active involvement and market learning.<br />
In its orthodox view, adaptive bidding behavior imperils incentive-compatibility. We<br />
introduce a model which shows that bidders bid according to the anchoring-andadjustment<br />
heuristic, contingent on a sequential weighting function, which neither<br />
ignores the incentive-compatibility constraints nor rejects the posted prices issued<br />
from others‟ bids. By deviating from their anchor in the direction of the public signal,<br />
bidders operate in a correlated equilibrium.<br />
Keywords: auctions, incentive-compatibility, rank-dependence, reference point,<br />
heuristic, bounded rationality, correlated equilibrium<br />
JEL classification: C73, D44, D81, D83
3.1. Introduction<br />
90<br />
"Verum esse ipsum factum 23 ."<br />
Giovanni Battista Vico.<br />
To know how much an individual is willing to pay for some item or for the<br />
provision of public services, and to assess how individuals would behave in the<br />
real world, economists now learn from experiments of repeated-round auctions. In<br />
this way, experimental auctions have been used to examine economic issues such<br />
as the disparity between willingness-to-pay and willingness-to-accept (Kahneman<br />
et al. 1990, Shogren et al. 1994, Shogren et al. 2001a) or preference reversals<br />
(Cherry et al. 2003, Cox and Grether 1996).<br />
In the presence of an active market, rational behavior ensues from<br />
repetition. In experimental repeated-round auctions, individuals repeatedly bid for<br />
the same good. One of the arguments supportive of repeating auctions is that<br />
practice allows bidders to learn about the auction format and form values in a<br />
market-like setting, which improves the accuracy of value estimates (Alfnes and<br />
Rickertsen 2003, Hayes et al. 1995, Lusk et al. 2001). Plott (1996) formulated the<br />
discovered preference hypothesis which says that preferences converge to the<br />
same underlying preferences – respectful of expected utility – regardless of the<br />
market mechanism. These underlying preferences are discovered after bidders<br />
repeatedly take decisions, receive feed-back on the outcomes of their decisions,<br />
and are given incentives to discover which actions best satisfy their preferences.<br />
Discovered preference hypothesis suggests an equality of mean bids across<br />
rounds. Since anomalies to standard theoretical requirements are the results of<br />
bidders‟ irrationality, only later market trials reveal the true preferences.<br />
Experimentalists want individuals to reveal their preferences truthfully.<br />
Therefore they use incentive-compatibility constraints, where truthfully<br />
announcing private information is an optimal strategy for all individuals<br />
participating in the auction mechanism. Incentive-compatibility is dependent on<br />
23 The true itself is made.
the restrictive assumption that individuals have independent private values. In<br />
strategic interactions under incomplete information, different types of bidders –<br />
such as high- or low-value types – select from a menu of strategies. In principle,<br />
incentive-compatibility forbids the possibility that a given type of bidder mimics<br />
the behavior of other types and adjusts her bids to theirs.<br />
One of the critics against the incentive-compatibility is the argument of<br />
uncertainty (Horowitz 2006a). After an individual reports her bid, she faces<br />
uncertainty over her chances to win the auction and over the final cost she will<br />
incur. On the assumption that the absence of affiliation is verified, repeating<br />
auctions in experiments reduces the uncertainty faced by bidders, because<br />
repeated-rounds provide market feedback from which they learn their preferences<br />
and produce reliable value estimates.<br />
Knetsch et al. (2001) find that bids are influenced by the choice of auction<br />
mechanism. They show that willingness-to-pay (WTP) bids submitted in the later<br />
rounds of a second-price auction are significantly higher than those submitted in<br />
the later rounds of a ninth-price auction. Shogren et al. (2001a) report that mean<br />
WTP bids increase in repeated second-price and random nth-price auctions, but<br />
not in a repeated BDM mechanism (the Becker-DeGroot-Marschak mechanism,<br />
described later on). Lusk and Rousu (2006) find that the BDM mechanism is less<br />
accurate than NPA (random nth-price auction, described later on) in generating<br />
bids consistent with true values and recommend the use of endogenous clearing-<br />
price mechanisms when estimating nonmarket goods and services. Indeed, under<br />
BDM, the price is determined separately from the bids, preventing interactions<br />
between bidders plus providing poor market learning. As such, bidders have no<br />
opportunity to perform in a competition that normally imposes discipline on their<br />
bidding behavior (Bohm et al. 1997). Ergo market anomalies and violations of<br />
economic theory are fostered (Lusk and Rousu 2006, Lusk and Shogren 2007).<br />
Still, only a default of interaction makes the independence of bids certain, as the<br />
probability of winning does not depend on others‟ preferences. Shogren and Hays<br />
(1997) assert that “the repeated signals sent by the endogenous market price<br />
91
contaminate individual bids into unreliable and unreasonable beacons of true<br />
preferences”.<br />
Under BDM, the distribution of clearing prices is often known in advance.<br />
When the price distribution is fixed in reference to the common endowment, the<br />
ambiguity of the potential price disappears. On the contrary, under NPA, the<br />
distribution of prices depends on what her opponents are ready to pay for the<br />
good. The nth highest bid will be linked to the highest value. A bidder thus bids as<br />
if she held the highest private value conditional on her subjective estimation of the<br />
distribution of her opponents‟ private values; she assesses her opponents and their<br />
expected valuations for the good. As a result, a complementary issue on<br />
uncertainty appears: uncertainty over the bids of opponents. Of course, bidders<br />
should always bid sincerely because the randomness of the market-clearing price<br />
prevents them from fixing on a stable cost such as with BDM (Shogren et al.<br />
2001b), but they are counter-incited to chase other bidders‟ true valuations.<br />
Several previous experimental studies advocate that affiliation between<br />
private values is factual. List and Shogren (1999) unearth affiliation between<br />
naïve bidders for new goods and influence of posted prices. Similarly, Bernard<br />
(2005) finds affiliation, loss of information about bidders‟ initial values and<br />
recommends the use of single-round auctions. Indeed, if the object of the<br />
experiment is to elicit actual preferences and to test them for consistency, price<br />
information is a potential source of contamination (Cubitt et al. 2001). Cox and<br />
Grether (1996) discover that bids are positively correlated with previous market-<br />
clearing prices. Although it can simply prove interaction between the learning<br />
processes of different subjects, it can also be the result of imitation. Knetsch et al.<br />
(2001) and Cubitt et al. (2001) also report experimental results which imply that<br />
bids are influenced by observations of past prices and by expectations of future<br />
prices. They argue that the provision of price information in repeated auctions<br />
induces cross-subject contamination. This is all the more un<strong>sur</strong>prising, for posted<br />
prices are the norm, unlike bargaining (Hanemann 1994).<br />
In this chapter, we relax the assumption of private values‟ independence in<br />
the repeated-round auctions such as BDM and NPA, when the market-clearing<br />
92
prices are made public at the end of each round. Instead of using game-theory<br />
learning models, we introduce a behavioral model that shows that bidders bid<br />
according to the anchoring-and-adjustment heuristic which neither ignores the<br />
rationality and incentive-compatibility constraints, nor rejects the posted prices<br />
issued from others‟ bids. Bidders simply weight information at their disposal and<br />
adjust their discovered value using reference points encoded in the sequential<br />
price weighting function. The general hypothesis is that selection among strategies<br />
is adaptive, in that a decision maker will choose strategies that are relatively<br />
efficient in terms of effort and accuracy as task and context demands are varied.<br />
For unfamiliar choices, individuals make up a decision rule at the moment they<br />
need to use it (Bettman 1988). Of particular interest is the finding that under time<br />
constraints, some heuristics are more accurate than a normative procedure such as<br />
expected value maximization (Payne et al. 1988). In fact, real people are cognitive<br />
misers: they tend to choose in the simplest way possible (Hanemann 1994). Put to<br />
the test, our model shows that bidders and offerers are sincere boundedly rational<br />
utility maximizers. Still, they act rationally even if they operate inside a correlated<br />
equilibrium. Instead of handling affiliation of values after market prices are<br />
revealed 24 , we prefer to speak in terms of reference point adaptation and posted<br />
prices‟ weighting mechanisms.<br />
The chapter is organized as follows. Section 3.2 introduces the auction<br />
mechanisms. Section 3.3 deals with the interactions among bidders and the<br />
incentive compatibility constraints. Section 3.4 presents a method for adjusting<br />
reference points according to a sequential price weighting function. Section 3.5<br />
examines the empirical validity of such a model. Section 3.6 concludes.<br />
3.2. Auctions and incentive-compatibility<br />
The BDM mechanism (Becker et al. 1964, Shogren et al. 1994) and the<br />
random nth-price auction (Shogren et al. 2001) are two market based mechanisms<br />
24 Another restatement proposed by Morrison (2000) is the leading, which is the following of the<br />
randomly chosen exchange price.<br />
93
often used in experiments. Determination of the market clearing-price and the<br />
expected payoff, which ensues from the market price, is different in the two<br />
mechanisms. In theory, they perform the same. In practice, this assertion no<br />
longer holds true.<br />
Under BDM, an individual reports a bid for a good; a price is then<br />
exogenously and randomly drawn from a price list. If the individual bids above<br />
the price, she receives the good and pays the drawn price. If the individual bids<br />
below the price, she does not receive the good and pays nothing. The mechanism<br />
is regarded as a quasi-market mechanism, its market price being exogenously<br />
determined.<br />
Under NPA, the market price is endogenously determined. The mechanism<br />
works as follows (see Shogren et al. 2001, List 2003): each bidder submits a bid;<br />
all bids are rank-ordered; a number between 2 and n (n being the number of<br />
bidders) is randomly selected as a market-clearing price; a unit of the good is sold<br />
to each of the n � 1 highest bidders at the nth-price drawn from the bids. Because<br />
of the endogenous market price, NPA is considered to be a full-active market.<br />
Following the induced value payoff theory, whatever the auction<br />
mechanism, an individual faces the following payoff rule:<br />
�vi<br />
� p if p < bi<br />
�<br />
�0<br />
if p�bi where i v is bidder i‟s value, b i her bid, and p the market price. Whenever<br />
optimal bidding arises with bi � vi,<br />
an auction mechanism is said to be incentive-<br />
compatible. Put differently, an auction is truth-telling when the individual pays a<br />
price independent from what she bids. As Lusk and Shogren (2007) point out, the<br />
incentive to value truthfully can easily be proved.<br />
When the individual i bids, she is ignorant of the price she will pay. So she<br />
draws an estimate of the price from the probability density function fi�p � with<br />
support p , p � �<br />
�<br />
� � �<br />
and the cumulative distribution function Fi�p � where<br />
94
p , p �<br />
�� �<br />
�<br />
�<br />
corresponds to the bid. The rational individual submits a bid that<br />
i<br />
�<br />
maximizes her expected payoff which corresponds to her expected utility u i ,<br />
which is twice continuously differentiable and increasing 25 :<br />
�<br />
bip � � � � � � � � � �0� � �<br />
� �<br />
E u u v p dF p u dF p<br />
i<br />
p<br />
i i i<br />
b<br />
i i<br />
�<br />
� � � � �0� b p<br />
� �<br />
�<br />
i<br />
�<br />
� �<br />
u v p f p dp u dp<br />
p<br />
i i i<br />
b<br />
i<br />
i<br />
i<br />
�<br />
The first integral describes the expected payoff for random prices below<br />
her bid (where she expects a positive <strong>sur</strong>plus). The second integral describes the<br />
expected payoff for random prices between her bid and the maximum possible bid<br />
(where she expects a loss). The maximum over b i occurs when the derivative of<br />
� �<br />
i<br />
Eu with respect to b i is null:<br />
� �<br />
�Eu<br />
�b<br />
i<br />
i<br />
where � �<br />
� � � � 0<br />
� u v �b f b �<br />
i i i i i<br />
ui 0 � 0.<br />
When bi � vi,<br />
the probability distribution that the individual‟s<br />
bid equals the price is strictly positive or we assume positive support on p , p � �<br />
�<br />
� � � .<br />
The individual maximizes her expected utility when she bids her true value.<br />
In BDM, the market-clearing price is drawn from a uniform distribution<br />
with the probability density function f � p � and a cumulative distribution function<br />
F�p � . Bidders have different values but face the same price which is modeled as<br />
the mean of the price distribution in the support of b i . The probability of winning<br />
the auction given i‟s bid is F�b i � . Taking her bid as given, the price that i expects<br />
to pay conditional upon winning is:<br />
25 Assumptions that satisfy the von Neumann-Morgenstern utility function.<br />
95<br />
[1]<br />
[2]
� �<br />
� �<br />
bi<br />
f p<br />
f � p p < bi � � � p dp<br />
[3a]<br />
�� F b<br />
i<br />
Then her expected utility is her expected payoff:<br />
� �<br />
� �<br />
� bi<br />
f p �<br />
E�� i � ��vi�� p dp�F�bi� [3b]<br />
��<br />
�� F bi<br />
��<br />
In NPA with n bidders, one of the bidders‟ values, from the uniform<br />
distribution 26 with PDF g�v � and CDF G�v � , is independently drawn at random<br />
and set as the market price. Conditional on v i being the nth value, the chance that<br />
a bid from the opponents is drawn as the n-order statistic is � 1�<br />
96<br />
n� n.<br />
The<br />
probability of winning given i‟s bid is Gb � i � . Taking her bid as given, the price<br />
that i expects to pay conditional upon winning is:<br />
� �<br />
� �<br />
n �1<br />
� bi<br />
g v �<br />
g � p v < bi � � � v dv<br />
n � �<br />
[4a]<br />
��<br />
�� G bi<br />
��<br />
Her expected utility is her expected payoff:<br />
� �<br />
� �<br />
1 i<br />
� �<br />
� �<br />
b � n � g v �<br />
E �i<br />
��vi� v dv G bi<br />
n � �<br />
[4b]<br />
��<br />
�� G bi<br />
��<br />
The BDM and NPA are proved to be incentive-compatible (Kahneman et<br />
al. 1990, Shogren et al. 2001b). Lusk et al. (2007) analyze the cost of<br />
misbehaving or deviating from truthful bidding in terms of foregone expected<br />
earnings, and show that suboptimal bidding has equivalent effects for BDM and<br />
NPA. For a uniform distribution of values, the incentive to bid their value is<br />
26 This time, the distribution comes from others‟ bids, not from a price list.
identical for both high- and low-type individuals, engaging all bidders to valuate<br />
truthfully.<br />
Let<br />
*<br />
� i be individual i‟s optimal payoff, which is achieved when an<br />
individual submits a bid equal to her value. Consider a bidder with a valuation<br />
slightly above or under v i . The deviation is profitable only if deviating is costless.<br />
The expected cost of deviating from bi � vi<br />
to bi�vi� �i<br />
, with � i >0,<br />
is given by<br />
ˆ � i :<br />
� � *<br />
E �� ˆ �i vi , bi , �i ���E� ��ivi, v � i ��E���ivi,<br />
vi<br />
��<br />
i ��<br />
Equation [5] represents the expected loss of an individual who does not bid<br />
her true value. It is a non-negative number that equals zero when �i � 0 . For both<br />
the BDM and NPA, the derivative of the expected cost of deviating with respect to<br />
� i at the point where vi bi<br />
� ˆ �<br />
�E<br />
�<br />
��<br />
i<br />
i v �b<br />
i i<br />
� 0<br />
� yields:<br />
Equation [6] states that only bidding sincerely is costless. If a bidder<br />
deviates and bids above her value, she may increase her chance of winning the<br />
auction, but her payoff will be negative even if she wins the auction. If a bidder<br />
deviates and bids under her value, she loses the auction and has zero payoff,<br />
which means that she loses the chance of winning the auction with some positive<br />
payoff. It is useful to think of the magnitude of deviation at the disposal of the<br />
bidder, which is the difference between her value and the highest bid. This would<br />
be the amount by which she could reduce her bid and still take part to the trades,<br />
or increase her bid to augment her chances of winning without supporting<br />
negative payoffs, once the distribution of high bids is known.<br />
97<br />
[5]<br />
[6]
In spite of the theoretical incentive-compatibility equivalence between<br />
elicitation mechanisms that employ endogenous and exogenous clearing prices,<br />
empirical evidence suggests that the two approaches generate divergent results. If<br />
the market price is based upon the preferences of other bidders, the risk of<br />
deviating from truthful bidding comes out. It is hard to distinguish between<br />
refining and copying, not only for experimentalists but for bidders too.<br />
3.3. Interactive incentive-compatibility<br />
Standard game models prescribe dominant strategies. Each individual has<br />
beliefs about the types of other individuals, how each individual values the good,<br />
and these beliefs are independent rational expectations, so the individuals‟ bidding<br />
strategies are constrained not to evolve. Indeed, incentive-compatibility requires<br />
that truth telling is best averaged over the types of other bidders in the auction.<br />
Incentive-compatibility constraints guarantee that it is optimal for the<br />
bidder to make a bid (send a signal to announce her type) truthfully. Let us<br />
consider two bidders i � 1,2 with unit demands, which are ex ante identical. Their<br />
valuations v 1 and 2<br />
v are independent, that is, each bidder‟s beliefs about the type<br />
of the other bidder are independent of the other bidder‟s belief distribution. Let b 1<br />
and 2 b denote the outcomes of the bidders‟ strategies � 1 and � 2 . The auction<br />
mechanism specifies the probability � , �<br />
*<br />
*<br />
p �b , b � . Let � �� � and � �<br />
i<br />
1 2<br />
1<br />
2<br />
f b b that the good is carried by i at price<br />
i<br />
1 2<br />
� � denote Bayesian Nash equilibrium strategies in<br />
the auction mechanism. For bidder 1, the rationality constraint is that, for each 1 v<br />
and for each 1<br />
v *<br />
2 �2<br />
v2<br />
*<br />
b belonging to the support of � � v<br />
� � 1 1 � 1 2 � 1 � 1 2 �<br />
98<br />
� :<br />
E E ��vfb, b � p b , b ��<br />
�0<br />
[7]<br />
1 1<br />
The rationality constraint en<strong>sur</strong>es that the bidder is willing to participate in<br />
the auction only in the case of nonnegative payoffs, since withdrawing from the
auctioning gives her null expected payoff. The probability distribution can be<br />
understood in different ways. Provided that bidder 1 controls 1<br />
99<br />
b but not 2<br />
b , we<br />
can think of a bidder as choosing a conditional probability distribution f1 �b1 b 2 � ,<br />
where b 2 has some exogenous probability distribution. Another interpretation is<br />
that � , �<br />
f b b is the result of a very complicated information mechanism by<br />
1 1 2<br />
which the bidder learns and updates her beliefs about b 2 . Finally, it can be<br />
understood as bidder i‟s actions over time.<br />
The incentive-compatibility constraint is such that, for each 1<br />
*<br />
the support of � � v<br />
ˆb :<br />
1 1 � and each deviation 1<br />
�<br />
� � 1 1 � 1, 2 � � 1 � 1, 2 ��� � � 1 1 � ˆ<br />
1, 2 � � 1 � ˆ<br />
1, v v<br />
2 �<br />
* *<br />
2 �222 �22<br />
v , each 1<br />
b in<br />
Ev E v f b b p b b Ev E �vfbbpbb� � � � � [8]<br />
The left-hand side of the constraint is the expected payoff if she reports her<br />
true bid b 1 , and the right-hand side of this constraint is the expected payoff if she<br />
deviates and reports 1<br />
ˆb . The idea here is that when bidder 1 bids 1<br />
ˆb instead of b 1 ,<br />
her payoff changes but the resulting probability distribution over b 2 does not<br />
change, since she cannot control b 2 , and hence she gets a different expected<br />
payoff. The incentive-compatibility constraint asserts that her expected payoff<br />
from honesty is not less than her expected payoff from deviating, i.e. by deviating<br />
she cannot gain more. The same applies to bidder 2. If the two bidders announce<br />
untruthful types � ˆ<br />
1, ˆ<br />
2�<br />
� � * *<br />
� �v �, � �v �<br />
b b , the probability of winning the auction is:<br />
fˆ bˆ , bˆ � E ��fb, b ��<br />
[9]<br />
� � � �<br />
1 1 2 2<br />
i 1 2 i 1 2<br />
The expected price is:
� ˆ ˆ � * *<br />
� �v �, � �v �<br />
pˆ b , b � E ��pb, b ��<br />
[10]<br />
� � � �<br />
1 1 2 2<br />
i 1 2 i 1 2<br />
The incentive-compatibility constraint en<strong>sur</strong>es that a Bayesian Nash<br />
equilibrium for both bidders is to announce the truth ( ˆv1 v1<br />
Regardless of how � , �<br />
i<br />
1 2<br />
100<br />
ˆv � v ).<br />
� and 2 2<br />
f b b occurs, if it violates the incentive-compatibility<br />
constraint, the bidder is not maximizing her expected payoff.<br />
Hausch (1986) asserts that an individual has an incentive to underbid in<br />
sequential auctions, i.e. to provide misleading information about her valuation of<br />
the good in the first round to deceive her opponents, in order to secure winning in<br />
the second round. Jeitschko (1998) demonstrates that bidders face a trade-off<br />
between increasing the probability of winning the early auction and increasing<br />
expected payoffs in the later auction. As a corollary, bidders place lower bids in<br />
the early auction, because they are aware of the learning effects.<br />
However, there is a strong information requirement. Each bidder must<br />
know the distribution of types of all the other bidders as well as the ability to<br />
determine the Nash strategies of every other bidder in the auction. In practice,<br />
equilibrium computation is usually infeasible. Moreover, the distribution over the<br />
possible types of n individuals in repeated-round auctions is complex and makes<br />
the space of types go of hand. One could calculate the equilibrium, but in the<br />
absence of common knowledge of type space and prior beliefs, it is unlikely to<br />
expect it (Saran and Serrano 2007). As a consequence, it is pragmatic to stress that<br />
individuals observe how others value the good, and some kind of equilibrium<br />
emerges (Boutilier et al. 2000) 27 .<br />
Theorists assume incentive-compatibility in the strict case of independent<br />
private values, which means that the individual‟s value is independently drawn<br />
from a commonly known distribution. In this case, the individual has only a prior<br />
on her signal. The setting of independent private values is reasonable for domains<br />
in which individuals‟ valuations are unrelated to each other, depending only on<br />
their signals. But when the bidder‟s valuation depends on both her signal and<br />
27 Recent literature shifts the analysis to the ex post equilibrium so any type space fits.
others‟ signals, those signals are likely to be affiliated: a phenomenon known as<br />
affiliated values pioneered by Milgrom and Weber (1982) 28 . For example, if a<br />
signal from an individual is a high value, this will increase the probability that<br />
other individuals will have high signals as well. As a consequence, a higher value<br />
for one bidder makes higher values for other bidders more likely (Kagel 1995).<br />
Values are drawn from an affiliated distribution if the posted price – which serves<br />
to signal the relative value of the good – shifts bids‟ distribution. Corrigan and<br />
Rousu (2006) make a distinction between bid affiliation and value affiliation, and<br />
prefer the bid affiliation as a broader concept. According to them, positive<br />
correlation between bids may not be caused by positive correlation between<br />
values: experimentalists observing bids, bid affiliation is a more relevant concept.<br />
When individuals actively interrelate, such as under NPA, they cannot<br />
circumvent estimating the probability distributions over maximal bids of other<br />
bidders and their chances of winning the auction given their true value. If the<br />
individual observes that others‟ bids are higher than her own, she learns she has<br />
little chance of winning the auction. In this case, the literature shows that<br />
individuals tend to submit higher bids afterwards (Fox et al. 1998, Cummings and<br />
Taylor 1999, List 2001). Likewise, Corrigan and Rousu (2006) experimentally<br />
find that posted prices have a statistically and economically significant impact on<br />
bids submitted in subsequent rounds. Furthermore, according to their study, the<br />
bidder‟s propensity to increase her bid is independent of her initial bid.<br />
Individuals combine their own signal with the signals received from<br />
others, which creates affiliation between values or bids (Klemperer 1999). For that<br />
reason, their value is given by:<br />
vi ��ti��� t<br />
i�j j<br />
[11]<br />
28<br />
Let x � x x �<br />
� be the vector of signals observed by the bidders. Let there be another vector<br />
1 ,..., n<br />
of signals containing information important to value the good. Bidders‟ values for the good are<br />
affiliated if v u �s, x�<br />
� . Otherwise, that is v � x , bidders‟ values of the good are privately and<br />
i i<br />
i i<br />
independently distributed.<br />
101
where t i is bidder i‟s signal, t j is j‟s signal, � is the weight assigned to i‟s signal<br />
and � the weight assigned to j‟s signal, with � � � . It is non-realistic to believe<br />
that individual i ignores others‟ signals. Her private value does not remain<br />
independent thus � � �0,1� . Finally, the individual does not bid her true value, and<br />
her over- or underbidding depends on the magnitude of � . Since the individual<br />
does not know other bidders‟ signals, she forms expectations on them.<br />
Learning preferences by repeating bidding is part of the methodological<br />
consensus. However, learning may also provoke unintended effects that challenge<br />
stricto sensu the constraints of incentive-compatibility. The reasoning is quite<br />
intuitive. An individual is given an initial endowment she uses as a reference to<br />
submit her bid. She reveals her value upon her preferences and this initial amount.<br />
Provided that a randomly selected round is chosen as the binding round in<br />
experimental repeated-round auctions, the individual bids in reference to the same<br />
endowment at the beginning of each round. In theory, this cannot compromise the<br />
property of demand-revealing. Nevertheless, she is told all the bids and the market<br />
price before submitting her bid in the next round, and revealing their distribution<br />
provokes an adaptive bidding behavior. Indeed, the individual extracts<br />
information on value perception from price formation in the auction, and price<br />
posting makes her update her values iteratively without fear of deviation.<br />
It is hard to believe that the process by which an individual maximizes her<br />
expected utility is one of assigning an independent value to the good after market<br />
information has been revealed. Assuming independent distributions implies that<br />
the individual is assumed to reason as if the bids for subsequent rounds were<br />
issued from independent beliefs. In other words, such a basic bidder is insensitive<br />
to strategic implications of varying � i in [5] and to the information content of t in<br />
[11]. Indeed, even if signals are mostly irrelevant to the payoffs, it is hard to<br />
exclude the possibility that they may find themselves into the equilibrium, which<br />
suggests existence of a correlated equilibrium (Aumann 1974). Moreover,<br />
Bayesian rational players play a correlated equilibrium as long the Harsanyi<br />
102
common prior assumption is verified (Aumann 1987) 29 . We think that the<br />
individual builds a bidding policy by which her bid is conditioned on the outcome<br />
of earlier rounds. Henceforth, the uncertainty is over opponents‟ bids. With an<br />
endogenous market-clearing price, the individual forms beliefs on the unknown<br />
distribution of the highest bid according to others‟ preferences. Her uncertainty<br />
over the parameters of this distribution is reflected by her prior distribution over<br />
the probability space of bid distributions. But, the use of equilibrium to describe<br />
the uncertainty relies on the existence of a type space as common knowledge,<br />
which is an important limitation.<br />
3.4. The behavioral model<br />
Consider dynamic settings where bidders interact repeatedly. We call a<br />
rule of behavior an adaptive heuristic. Invariably making the same choice is a sort<br />
of heuristic but not an adaptive one, since it is not responsive to a situation. At<br />
each stage, a bidder plays a strategy which is optimal against the distribution of<br />
the past actions of other bidders. Adaptive heuristics are boundedly rational<br />
strategies 30 . However, in the long run, such simple strategies yield highly<br />
sophisticated and rational behavior (Hart 2005).<br />
Now consider an individual who is aware of the strategic implications<br />
inbuilt in the auction, such as the effects of varying expectations on the adjacency<br />
of potential opponents‟ values to hers. We believe that instead of using a single<br />
bidding policy at every round, individuals use the distribution of bids they‟ve<br />
observed at earlier rounds to update their bidding policy and their estimate of the<br />
true distribution of high bids. Their bidding strategy in the next round is based on<br />
the updated distributions and all individuals play a Nash equilibrium in a Bayesian<br />
29 Common prior only requires the bidders‟ mutual beliefs on the fundamentals of the interaction<br />
be elicited, like expected payoffs entailed by the possible actions.<br />
30 Learning dynamics are levels of full rationality, whereas evolutionary dynamics are completely<br />
irrational actions. Adaptive heuristics are in-between.<br />
103
manner 31 . If the individual updates her bidding policy based on past observations,<br />
her true bids at early rounds are not reflective of the bids she submits at latter<br />
rounds, which means she is learning based on observations drawn from a<br />
nonstationary distribution. It has been shown that myopic learning models such as<br />
fictitious play – which is an adaptive heuristic – converge to a stationary<br />
distribution despite the initial nonstationarity (Fudenberg and Levine 1998). In a<br />
fictitious play, the individual is enabled to learn if she can realistically win the<br />
auction given her true value. She learns by observing the history of past bids –<br />
prior to the beginning of the next round – and forms a belief about her opponents‟<br />
bids in the next period. She believes that her opponents are using a stationary<br />
strategy which is the empirical distribution of past bids, and thus updates her<br />
beliefs, her best reply and bid, computing a new bidding policy based on earlier<br />
outcomes. Although truth-telling is theoretically proved to be optimal, computing<br />
optimal bids as best replies defies the assumption of true valuation 32 .<br />
Instead of using these learning models, let us introduce a descriptive<br />
behavioral model based upon reference point adaptation. We introduce a parallel<br />
model to rank-dependent expected utility, because we consider agents to derive<br />
utility from changes in wealth relative to their reference point. If an agent<br />
perceives her payoff to be higher than the reference point, she perceives a gain;<br />
and perceives a loss, otherwise. We exploit the idea of linear and non-linear<br />
probability weighting and propose a sequential information weighting because we<br />
assume that strategic bidders convert objective linear weighting into subjective<br />
nonlinear decision weights.<br />
In this case, let us assume that bidders adjust their starting values.<br />
Anchoring-and-adjustment is a heuristic that influences the way individuals<br />
intuitively assess probabilities. According to this process, individuals start with a<br />
31 In the long run, irrational behavior can lead to Bayesian rationality (Aumann 1987).<br />
32 Shlomit et al. (1998) analyze a repeated first-price auction in which the types of the players are<br />
determined before the first round and do not vary in time. When each player uses a fictitious play<br />
learning scheme, the equilibrium vector of bids is the same as in a one-shot auction with the types<br />
of players being common knowledge. However, their players are too basic for they do not attempt<br />
to learn their opponents‟ types or to hide their own types.<br />
104
eference point (the anchor) and make adjustments to it to reach their estimate 33<br />
(Tversky and Kahneman 1974). In the case of repeated one-shot auctions, their<br />
true value is a reference point that bidders discovered in time, i.e. people use<br />
practice rounds to refine their values with regard to their vague or naïve start.<br />
Deviation from true value is then an adjustment from the self-generated anchor in<br />
order to win the auction in the late rounds. When bidders long to increase either<br />
their payoffs or their probability of winning the auction, given that a rational agent<br />
is programmed to maximize her payoff, deviating can be considered rational.<br />
The reference point is formed after observing the last posted price. The<br />
bidder thus makes her bid in i � 1 according to i<br />
or i i 1<br />
105<br />
r . Depending on whether pi> ri� 1<br />
p < r� , she scales her bid up and down, respectively. The adaptation of the<br />
reference point corresponds to the following phase diagram:<br />
r0 p0<br />
� 1 r 2 p 2 r<br />
p1<br />
p1 r0<br />
� p2 � r1<br />
Arkes et al. (2008) term the adaptation of the reference point the rule<br />
where bidders shift their reference point in the direction of a realized outcome. If<br />
the reference point is 0 r and the price is 1 p , the difference between 1 p and 0 r<br />
should be equal to the difference between 2 p and 1 r , or p1 r0 p2 r1<br />
� � � . This is<br />
standard rationality. It is due to the linear shape of the utility function where<br />
bidders are indifferent to rank-dependence. We term this the uniform or linear<br />
adaptation of the reference point. If the utility function v is linear, the reference<br />
point is a weighted average of posted prices. With p0 � r0<br />
as the anchor in a<br />
fictitious period i � 0 , the next bid is formulated along with:<br />
33 Einhorn and Hogarth (1985) have also considered the anchoring-and-adjustment process to<br />
describe how people make judgements under ambiguity; their adjustment is made according to<br />
some probability p which could come from any distribution.<br />
� 0 � 0
1 n<br />
ri � � w<br />
i 1 i p<br />
� i<br />
[12]<br />
i<br />
ˆ ˆ<br />
The expected gain of deviating or adapting the reference point from b1 � v1<br />
ˆ � :<br />
to b1 � v1,<br />
conditional on 2 b is given by 1<br />
� �<br />
E �ˆ �1 b1, bˆ ˆ<br />
1, v �<br />
2 �E��1b1, b �<br />
2 �E���1b1, b2<br />
�<br />
� � � �<br />
�<br />
106<br />
[13]<br />
There are several competing notions of rationality, and one among them is<br />
the correlated equilibrium, which has the advantage of being reasonable, simple<br />
and is guaranteed always to exist. The rationality constraint says that a bidder has<br />
no reason to bid in case of null payoff. Since losing in auctioning means absence<br />
of payoff, increasing the probability of winning the auction and consequently the<br />
chance of earning some positive payoff by deviating is rational. In parallel, a<br />
rational bidder seeks to maximize her payoff which is the difference between her<br />
value and the cost of the item. If by deviating, a bidder increases her expected<br />
payoff with some extra gain, she is acting rationally.<br />
In terms of interactions between two players, the deviation of player 1 is<br />
such that, for all b 1 and 1<br />
ˆb in � � v<br />
1 1 � and all 2 2<br />
ˆ b , b in � � v � :<br />
2 2<br />
� � � ˆ<br />
1 1,2 1, ˆ<br />
2 � � � ˆ<br />
1 1, ˆ<br />
2 � � � � 1 1,2 � 1, ˆ<br />
2 � � 1 � ˆ<br />
1, v v<br />
2 �<br />
Ev E �v f b b p b b � E<br />
2 2 2 v E �v f b b p b b �<br />
� � � 2 �22��<br />
[14]<br />
If joint distribution 1,2<br />
f with � ˆ ˆ<br />
v �<br />
1 v f<br />
2 1,2 b1 b2<br />
� � , � 1 is a correlated strategy,<br />
equilibrium is achieved when no player ignores the public signal, which is to<br />
make an expected gain from deviating with some positive probability, given that<br />
others follow this rule as well. This implies that deviating is worthwhile only if a<br />
public signal such as a posted price recommends doing so and all submit to it<br />
because the suggested strategy is the best in expectation. The right hand-side<br />
expression is when player 1 is the only one not to follow the recommendation
issued from the public signal and chooses some bid b 1 instead of ˆb 1 , provided the<br />
endogeneity of the market-clearing price.<br />
Proposition 3.1.: When bidders follow the public recommendation leading them<br />
to rationally deviate from their anchor, there exists a correlated equilibrium.<br />
Proof: In the appendix.<br />
The incentive-compatibility constraint en<strong>sur</strong>es that truthful bidding<br />
maximizes utility. Let us now consider this point. Following the work on rank<br />
dependent expected utility (Bleichrodt and Pinto 2000) and reference point<br />
adaptation (Arkes et al. 2008, Baucells et al. 2008), we introduce a model of<br />
sequential decision analysis. First, let us recall the existing decision theoretic<br />
background.<br />
According to cumulative prospect theory (Tversky and Kahneman 1992),<br />
people weight outcomes when they choose between lotteries. Let<br />
�� , p ; � , p ;...; � , p ; � , p �<br />
1 1 2 2 n�1n� 1 n n be a lottery that yields outcome i<br />
probability i<br />
107<br />
p with<br />
� . A lottery can be defined as a set of n outcomes � p , p ,..., p , p �<br />
with respective probability �� , � ,..., � , � �<br />
1 2 n� 1 n<br />
1 2 n� 1 n<br />
. The rank-dependent expected utility<br />
of this lottery is a junction between the value or utility function v�� � and the<br />
weighting function w :<br />
n<br />
�� , ; � , ;...; � , ; � , � � �<br />
v p p p p � � v p w<br />
[15]<br />
where<br />
1 1 2 2 n�1n�1nn i�1<br />
i i<br />
i i�1<br />
� � 1 � � � 1 �1<br />
�<br />
� � [16]<br />
w �w�w i i i
in particular w � w���. The weighting function w is increasing with � �<br />
1 1<br />
and w�1�� 1.<br />
It is a function of the cumulative distribution at i<br />
108<br />
� and i 1<br />
w 0 � 0<br />
� � . If w is<br />
an identity transformation and corresponds to a positive linear transformation of<br />
v , the rank-dependent expected utility theory is equivalent to the expected utility<br />
theory. In this case, bidders are considered rational: they have linear or uniform<br />
preferences for money, separately from the rank position. Tversky and Kahneman<br />
(1992) rather take w�� � as nonlinear, that is, a monotonic s-shaped function,<br />
which implies deviations from linearity and irrationality because of insensitivity<br />
to or misperceiving of mean probabilities. That takes the form as follows:<br />
w<br />
�� �<br />
�<br />
�<br />
�<br />
� � ��<br />
1<br />
� � �<br />
� � 1��<br />
[17]<br />
for 0< � � i for i close to 1 or n.<br />
By analogy, we assume that bidders weight all the sequential information<br />
at their disposal to build their bidding strategy, in particular their anchor and the<br />
posted market-clearing prices. Bidders start with an outcome p0 � r0<br />
which is<br />
their original reference point and which corresponds to their subjective and asocial<br />
valuation of a good. Put differently, their first reference point is their value after<br />
the practice rounds: a true value issued from discovered preference hypothesis. In<br />
repeated-round auctions, bidders are told the market-clearing price – which can be<br />
endogenous to the bids – before submitting their next bid, so all posted prices<br />
correspond to subsequent outcomes of the outcome set.<br />
Instead of ordering outcomes from worst �i � 1�<br />
to best �i n�<br />
� as in<br />
cumulative prospect theory, we assume that bidders sort the outcomes backwards,<br />
from the latest to the anchor, according to pi� pn�� i 1,<br />
with i being the rank of the<br />
round. Posted prices arrive following the sequence of rounds. Therefore, the price<br />
vector is sequentially sorted. By analogy to the probability weighting function
(Einhorn and Hogarth 1985, Tversky and Wakker 1995), we assume a sequential<br />
price weighting function, such that bidders give a weight of 1/n to each price, with<br />
n the length of the price sequence. We define the sequential rank-dependent<br />
function.<br />
Definition 3.1.: W � A� � W �B� whenever A� B.<br />
If W is additive, i.e.<br />
W � A� B� �W � A� � W � B�<br />
for all disjoint outcomes A and B , then it is a<br />
weighting mea<strong>sur</strong>e. A sequential price weighting mea<strong>sur</strong>e is a strictly increasing<br />
function w: �0,1� � �0,1� with w�0�� 0 and � �<br />
109<br />
w 1 � 1.<br />
A weighting mea<strong>sur</strong>e W<br />
on P , with P the outcome space, is a function whose components are included in<br />
�0,1 � such that W ���� 0 and W�P�� 1.<br />
Bidders rank prices following the mirror reflection. Henceforth, the<br />
sequential sorting is: � p , p ,..., p , p � � p , p ,..., p , p �<br />
accumulated such that:<br />
��n �n�1 �2 �1<br />
�<br />
� . Ranks are then<br />
1 2 n�1nnn�121 �121� , ,..., , : �� , ,...,1 � ,1�<br />
�nnn� [18]<br />
where 1/n corresponds to the weight of the latest posted price, and the last<br />
increment corresponds to the weight of the anchor. A sequential weighting<br />
function is introduced to transform the ranks into cumulative decision weights:<br />
� 1 � � 2 � � 1 �<br />
�w�� n�, w�� n�1�,...,<br />
w�� 2 �, w��1 ��<br />
: � w� �, w� �,..., w�1 � �,<br />
w�1�<br />
� �<br />
� �<br />
� � n � � n � � n � � [19]<br />
Following [16], the weighting factor is an increment between two rounds 34 :<br />
34 Uniform or linear weighting results as a special case, and we have<br />
� � � 1 � �1 �<br />
w i n � w i � n � w n for all i.
�� � �� �1<br />
�<br />
� i � � i�1�<br />
wi � w i � w i : � w� � � w�<br />
� , i� 1,..., n<br />
[20]<br />
� n��n �<br />
The cumulative prospect theory suggests an s-shaped weighting function<br />
that overweighs extreme outcomes which occur with small probabilities and<br />
underweighs average outcomes which occur with high probabilities. In lieu of this<br />
point, we assume that bidders overweight the beginning and the end of time<br />
series 35 . Indeed, one reference point in the context of stock investment is the<br />
starting point which enjoys a privileged role (Spranca et al. 1991). As well,<br />
investors partially update their reference point after a stimulus is presented to a<br />
price between the purchase price and the current price, but they do it incompletely<br />
(Chen and Rao 2002) 36 .<br />
The sequential weighting function presented in Fig. 3.1. is s-shaped: it is<br />
steep near 0 and 1 and mild in-between. Thus, a low interval �0,1 n � and a high<br />
��11 ,1��<br />
interval ��<br />
n�<br />
have more impact than a middle interval 1 n,1�1n� �<br />
110<br />
�� ��<br />
.<br />
To compute her next bid, the bidder takes into account a reference point,<br />
and adjusts her estimates upon the weighted sequential price vector. If the posted<br />
price is higher than her latest reference point, she revises her value and her bid<br />
upwards to increase her chance of winning the auction, given that she learns that<br />
she earns a null payoff with her previous bid: where she does not maximize any<br />
utility. This could simply mean that she has a higher reservation price for a good<br />
than the bid she posted in the first round. If the posted price is lower than her<br />
latest reference point, she will revise her value and her bid downwards in order to<br />
35 We are drawn to the s-shaped decision-weighting function partly because of convenience to<br />
represent some non-linear weighting.<br />
36 Another reference point used by individuals is the historical peak (Gneezy 2005) and<br />
expectations about future outcomes (Koszegi and Rabin 2006).
augment her payoff, as she learns that she can deviate and still take part to the<br />
trades: where she maximizes her expected gain and accordingly her utility 37 .<br />
Fig. 3.1. The sequential price weighting function<br />
As we can see, introducing the sequential price weighting function<br />
modulates the linear or uniform adaptation of the reference point. In point of fact,<br />
w is s-shaped, so the latest posted price and the anchor will most impact the<br />
valuation of the reference point. Their respective weights amount w�1n � and<br />
� � ��<br />
1�w 1� 1 n . The rest is distributed among the in-between, that is,<br />
�1 �1 ��<br />
�1 �<br />
w � n � w n .<br />
� �<br />
�1� � 1��1� w w n<br />
�1��1��� �1 �<br />
w n w n<br />
�1 � � �0� w n w<br />
This model lies between evolutionary dynamics and adaptive heuristics. In<br />
the evolutionary literature, inertia means that the bidder will invariably repeat a<br />
bid in i � 1 she used in i. If her bid is sincere, it implies that she is always bidding<br />
truthfully. In our case, she will adjust her bid in the direction of the last posted<br />
price, and an adaptive rule based on the posted prices has an important component<br />
of heuristics. Since we are dealing with posted prices issued from others‟ bids,<br />
37 Aumann has argued that rationality should be examined in the context of rules rather than acts,<br />
i.e. rules of behavior that are better to other rules.<br />
0 1�1 n 1<br />
1 n � �<br />
111
linear or uniform weighting supposed to reveal rationality (Van de Kuilen 2009)<br />
no longer holds. Bidders with well-defined preferences exploit the market<br />
mechanism to discover their true preferences. If their preferences satisfy standard<br />
theoretical requirements, the discovered preference hypothesis implies that<br />
irrationality is the results of individuals‟ errors, and these can be reduced by<br />
market experience. However, only later market trials can reveal their true<br />
preferences. According to this rationale, when the bidder has discovered her value,<br />
moving from it becomes irrational. In fact, because the bidder has discovered her<br />
preferences, adjusting her bid upon posted prices cannot be considered rational,<br />
for truth-telling is rational and affiliating private values on public signaling is not.<br />
Although we accept the model of discovered preferences, because we consider it<br />
to reveal the anchor, we believe that bidders can partially adapt their reference<br />
point according to posted prices and still be sincere.<br />
We thus model the concept of inertia as high weighting of the anchor,<br />
which stands for truthful bidding and high regard to freshly discovered<br />
preferences. Adjustment means adaptive rule based on adaptation of the reference<br />
point in the direction of the posted price. It helps a bidder to maximize her<br />
expected payoff, which is after all the only purpose that matters to rationality.<br />
From the above, the two components simply suggest that sincere bidders are<br />
boundedly rational. Once a bidder has discovered her preferences, she is<br />
considered insincere only if she scales her references point upon the posted prices<br />
issued from others‟ bids with uniform sequential weighting, i.e. null inertia, where<br />
her anchor – a result of discovered preferences hypothesis – would be drowned by<br />
the sequence of posted prices. The following proposition comes into existence.<br />
Proposition 3.2.: A bidder is truth-telling inasmuch as she behaves as a<br />
(boundedly rational) utility maximizer 38 , i.e. so long as she bids pursuant to the<br />
sequential s-shaped weighting function.<br />
Proof: In the appendix.<br />
38<br />
This can be connected to the equation [11] where � � � .<br />
112
The correlation between bids comes from the commonly observed history<br />
of play and each bidder‟s actions are determined by the history. Uniform<br />
weighting means that at round i each bidder knows the history of the repeated<br />
one-shot auction; that is, each bidder uniformly considers all prices that were<br />
posted in all previous rounds. We consider bidders to be sincere if they have<br />
limited memory and confine their reference point adaptation to their anchor and<br />
the latest posted price. S-shaped weighting mechanism reflects such a bidding<br />
strategy 39 .<br />
Our model predicts that different-type bidders will pursue a similar rule as<br />
they get into interactions via endogenous market-clearing prices, no matter what<br />
their anchors are. Of course, preferences are no longer invariable in time due to<br />
the local weighting function, but this guarantees the high weight given to freshly<br />
discovered preferences. Besides, bidders still seek to maximize their expected<br />
payoff. Although bidders would orthodoxically be regarded as irrational, this<br />
model shows that sincere bidders are just boundedly rational.<br />
3.5. The empirical study<br />
Let us now test the empirical relevance of the sequential weighting<br />
function. We reprocess the home-grown data from the BDM and NPA<br />
experimental auctions on the carbon offset (regarded as an unfamiliar good)<br />
realized by <strong>Dragicevic</strong> and Ettinger (2009). We analyze the five – out of ten – last<br />
rounds because we consider bidders and offerers to have discovered their<br />
preferences after a sufficient number of practice rounds. If bidders or offerers are<br />
to deceive and compute their bids or offers insincerely, they reasonably do it from<br />
this point of time.<br />
Under BDM, the market-clearing price is exogenously and randomly<br />
chosen from a price list, so the value of the good is worth any market-clearing<br />
39 Contrary to Dasgupta and Maskin (2000) we do not use all available information. In addition,<br />
we work with time series and accumulated ranks reflect such a sequential optimization.<br />
113
price. If every posted price is uniformly weighted, subjects are naïve. Under NPA,<br />
the market-clearing price is endogenously and randomly chosen, so the value of<br />
the good is worth anybody‟s value participating in the auction. If every posted<br />
price is uniformly weighted, subjects are insincere because they are copying<br />
others‟ values.<br />
As shown in equations 21 and 22, we correspondingly compute the<br />
following theoretical bids:<br />
� 1 i 1 �<br />
1 n<br />
bi m b pi<br />
i � � �� [21]<br />
�<br />
n�1<br />
� i�<br />
�<br />
� n�1� n�21<br />
bi � �1��h�b1��m� p<br />
1 i ��<br />
h pn<br />
[22]<br />
� n � n n<br />
We estimate bids and offers of the subsequent round according to the<br />
uniform and s-shaped reference point adaptations previously explained. We use<br />
one-parameter specification factors � m � 0.61 for moderate weighing and<br />
� � 0.69 for high weighting from Tversky and Kahneman (1992) 40 .<br />
h<br />
Table 3.1. Unitary sequential weight coefficients<br />
Round estimate Uniform weighting S-shaped weighting<br />
Anchor In-between <strong>La</strong>st Price Anchor In-between <strong>La</strong>st Price<br />
10th (5–8) 41 0.102 0.102 0.102 0.425 0.102 0.115<br />
Normalization 0.167 0.167 0.167 0.449 0.107 0.121<br />
9th (5–7) 0.122 0.122 0.122 0.448 0.122 0.138<br />
Normalization 0.200 0.200 0.200 0.471 0.128 0.145<br />
8th (5–6) 0.153 0.153 0.153 0.483 0.153 0.173<br />
Normalization 0.250 0.250 0.250 0.503 0.159 0.180<br />
7th (5–5) 0.203 0.203 0.203 0.540 0.203 0.230<br />
Normalization 0.333 0.333 0.333 0.555 0.209 0.236<br />
�<br />
40<br />
We rather use linear � �1 n�<br />
instead of power �1 n� underweighted otherwise.<br />
41 (. – .): in-between rounds.<br />
114<br />
factoring, because the anchor gets
Table 3.2. Summary statistics of the uniform and s-shaped theoretical estimates<br />
115<br />
WTP bids WTA offers<br />
Auction mechanism nth round 7 8 9 10 7 8 9 10<br />
BDM First bid or offer (5th round) 8.29 8.29 8.29 8.29 8.92 8.92 8.92 8.92<br />
<strong>La</strong>st posted price (n – 1) 1.50 5.00 6.50 13.50 1.50 5.00 6.50 13.50<br />
Average real bid or offer 8.39 8.71 8.82 8.61 9.53 9.19 8.67 8.03<br />
Average bond between two rounds 0.32 0.11 –0.21 –0.35 –0.56 –1.05<br />
Uniform bid or offer average estimate 7.92 7.20 7.06 8.15 8.13 7.35 7.18 8.25<br />
Average bond between two rounds –0.72 –0.14 1.09 0.78 –0.17 1.07<br />
t-test* of bonds between two rounds<br />
7.24 1.34 –3.87 1.19 –0.49 –5.02<br />
Average SSE 42 (uniform residual) 7.10 10.39 12.19 12.77 10.87 12.81 12.96 13.83<br />
S-shaped bid or offer average estimate 7.88 7.53 7.47 8.24 8.23 7.85 7.77 8.53<br />
Average bond between two rounds –0.35 –0.06 0.77 –0.38 –0.08 0.76<br />
t-test* of bonds between two rounds<br />
5.04 0.98 –2.98 0.08 –0.69 –4.11<br />
Average SSE (s-shaped residual) 3.85 5.49 6.62 7.33 6.56 7.17 5.90 7.04<br />
NPA First bid or offer 4.77 4.77 4.77 4.77 9.86 9.86 9.86 9.86<br />
<strong>La</strong>st posted price (n – 1) 1.50 8.51 7.84 7.03 10.00 5.00 5.88 7.96<br />
Average real bid or offer 6.18 6.12 6.85 6.72 9.17 9.14 9.23 9.37<br />
Average bond between two rounds –0.06 0.73 –0.12 –0.03 0.09 0.14<br />
Uniform bid or offer average estimate 4.14 5.33 5.83 6.04 9.11 8.09 7.65 7.71<br />
Average bond between two rounds 1.19 0.50 0.21 –1.02 –0.44 0.07<br />
t-test* of bonds between two rounds –2.41 0.67 –0.60 3.19 2.06 0.09<br />
Average SSE (uniform residual) 7.78 5.06 6.64 7.92 10.80 9.90 12.36 19.62<br />
S-shaped bid or offer average estimate 4.37 5.21 5.50 5.60 9.39 8.64 8.37 8.42<br />
Average bond between two rounds 0.84 0.29 0.10 –0.76 –0.26 0.05<br />
t-test* of bonds between two rounds<br />
–1.77 0.50 –0.40 2.66 1.42 0.10<br />
Average SSE (s-shaped residual) 6.41 4.22 5.84 6.14 8.66 6.56 7.27 14.81<br />
* H0: The difference between experimental and theoretical average bonds is zero at 5% significance.<br />
42 SSE: the sum of the squares of the residuals.
We normalize the sequential weights to one (Table 3.1.) in order to compute<br />
the reference point from which the bid or offer is figured out and to compare it to the<br />
real bid or offer (Table 3.2.). We study both the (insincere) uniform weighting and the<br />
(sincere) s-shaped weighting.<br />
Table 3.2. presents the summary statistics of the uniform and s-shaped<br />
theoretical estimates and their comparison to the experimental results of trials 7–10.<br />
The WTP market-side is analyzed as follows. If the real bid is greater than or equal to<br />
the theoretical bid, the bidder overbids regarding her reference point. If the real bid is<br />
lower than the theoretical bid, the bidder underbids regarding her reference point.<br />
When the bidder overbids, she values the good more than what her reference point<br />
suggests. She increases her chances of winning the auction but decreases her expected<br />
payoff regarding her true value. If the uniform residual is higher than the s-shaped<br />
residual, the bidder is considered insincere. The WTA market-side is analyzed as<br />
follows. If the real offer is greater than the theoretical offer, the offerer overoffers<br />
regarding her reference point. Otherwise, she underoffers. When the offerer<br />
underoffers, she values the good less than what her reference point suggests. She<br />
increases her chances of winning the auction but decreases her expected payoff<br />
regarding her true value. If the uniform residual is higher than the s-shaped residual,<br />
the offerer is considered insincere.<br />
Our first investigation reveals that within BDM, only 26% of offerers and<br />
22% of bidders stick to their discovered value. Within NPA these figures even<br />
collapse to 13% for both offerers and bidders, making the auctioning tactical until the<br />
last round. Let us now see whether agents‟ strategies are based upon market-clearing<br />
prices by looking at the average bonds in bids and offers between two rounds. Given<br />
that we believe that agents incorporate public signals into their bidding and offering<br />
strategies, we analyze the impact of posted prices on their bids and offers, i.e. their<br />
freshly discovered preferences. We thus look at Student‟s t distribution between<br />
experimental and theoretical data and regard whether they fit. With NPA and under<br />
both adaptation weightings, the theoretical bonds in bids and offers are not<br />
significantly different from the real bonds in bids and offers. The t-test fails to reject<br />
the null hypothesis that the theoretical bonds in offers and the real bonds in offers<br />
116
come from the same distribution at the p
3.3. All estimates are significant, i.e. all p-values amount less than 0.001, and all R-<br />
squares are higher than 0.9. Despite the fact that they are comparable, we find that<br />
each one of the market-sides has its own � . We do not identify � m in s-shaped<br />
weighting because the factor oscillates around zero and is not significant; therefore,<br />
bidders and offerers simply weight the anchor and the last posted-price, which proves<br />
their sincerity as well as the relevance of our descriptive model. As one notices, the<br />
regression factors we used to compute theoretical estimates are higher than those<br />
usually elicitated in the gain and loss perception. However, they are in accordance<br />
with experimental data <strong>sur</strong>passing our predictions.<br />
Table 3.3. � -factors statistics<br />
� estimate 43 Uniform weighting S-shaped weighting<br />
BDM NPA BDM NPA<br />
Bidders 1.18 (0.02) 1.24 (0.03) 1.24 (0.03) 1.16 (0.06)<br />
Offerers 1.19 (0.02) 1.17 (0.03) 1.15 (0.03) 1.21 (0.07)<br />
Let us now discuss about the implications of the differences between the<br />
uniform and s-shaped estimates and the real bids or offers. Because market-clearing<br />
prices are exogenously determined under BDM, even though the risk of a uniform<br />
reference point adaptation exists, it does not compromise the incentive-compatibility.<br />
In experiments, bidders and offerers are foreseen as sincere. At worst, they are naïve,<br />
for it is irrational to run after luck. On the WTA market-side, the average s-shaped<br />
SSE is lower than the average uniform SSE, indicating that offerers are sincere: they<br />
refine their values in time. We observe the predominance of the s-shaped weighting<br />
on the WTP market-side as well. Finally, we denote that the average SSE is higher on<br />
the offerers‟ side than on the bidders‟ side. This is due to loss aversion of three out of<br />
eighteen offerers, who systematically proposed a ceiling WTA. When we ignore<br />
them, the average SSE is similar between both market-sides.<br />
43 Standard errors in parentheses.<br />
118
Because market-clearing prices are endogenously determined under NPA, not<br />
only does the risk of a uniform reference point adaptation exist, but it compromises<br />
the incentive-compatibility of the market mechanism. Bidders and offerers are<br />
foreseen as potentially insincere. On the WTA market-side, the average s-shaped SSE<br />
is lower than the average uniform SSE, suggesting that the offerers are sincere.<br />
Furthermore, the average SSE is higher on the offerers‟ side than on the bidders‟ side.<br />
This can be explained by loss aversion of two out of sixteen offerers, who proposed a<br />
ceiling WTA in each round, and by one offerer who had no stable strategy. When we<br />
ignore them, the average SSE remains above the SSE of the buyers‟ market-side, but<br />
is similar to the average SSE of the BDM sellers‟ market-side. We also observe lower<br />
average s-shaped SSE on the WTP market-side, which illustrates sincere bidding. The<br />
NPA average buyers‟ SSE is the lowest, all auction mechanisms and market-sides<br />
taken into account.<br />
At last, we notice that average WTP estimates, under both auction<br />
mechanisms, are beneath the average real bids: the real bids and offers are always<br />
higher than what the model suggests. Given that the bidders and offerers were<br />
confirmed to be sincere, we believe this is due to the reference point adaptation<br />
overstated by the combination of regret and competitive pres<strong>sur</strong>e. Indeed, theory of<br />
disappointment aversion (Horowitz 2006b) says that a bidder reports a higher value<br />
than the true one, simply because she is more disappointed from not receiving the<br />
good than from receiving it overpriced 44 . Let us recall that the WTP and WTA value<br />
mea<strong>sur</strong>es approach equality more by virtue of the steady increase of the buyers‟ bids<br />
than the weak decrease of the sellers‟ offers. When we proceed to the computation of<br />
the NPA s-shaped estimates on the WTP market-side, but consider the posted prices<br />
issued from the offers instead of the bids, not only do we obtain an average SSE of<br />
4.17 but also average WTP estimates almost equal to the real bids 45 . This unexpected<br />
result can stand for a high influence of the sellers‟ clearing prices on the bidders‟<br />
44 An alternative formulation of joy-of-winning was tested by Goeree et al. (2002) but they find that it<br />
does not add anything to the explanation of overbidding.<br />
45 This cannot hold with BDM, because the market clearing price is the same for both market-sides and<br />
because it is exogenously and randomly drawn from a known list.<br />
119
eference point adaptation. It can also mean that the WTA posted prices – unlike the<br />
WTP posted prices – incorporate a behavioral effect of loss aversion combined with<br />
competitive pres<strong>sur</strong>e, which works as a catalyst, from the very beginning. The<br />
magnitude of disappointment aversion and loss aversion would then be similar, and a<br />
way to verify it is to call to mind the likeness of � -factors between bidders and<br />
offerers. The excess in these values could be the quantitative mea<strong>sur</strong>e of the<br />
competition‟s pres<strong>sur</strong>e.<br />
To end this section, let us verify if it is worthwhile deviating such as<br />
suggested by rational deviation. We compute the percentage of expected bidders and<br />
offerers who win some positive payoff by sticking to their anchor, and the percentage<br />
of expected bidders and offerers who win some positive payoff by deviating from<br />
their anchor according to the last market-clearing price. We then do the same<br />
computation on real payoffs obtained by not deviating and deviating from the anchor.<br />
The results presented in Table 3.4. show that deviating pays, since both expected and<br />
real deviating gainers outnumber.<br />
Table 3.4. Comparison between extra expected and real winners from deviation<br />
per cent BDM WTP BDM WTA NPA WTP NPA WTA<br />
Extra expected deviant<br />
gainers 2.63 4.17 10.00 10.94<br />
Extra real deviant<br />
gainers 0.00 2.78 3.33 3.13<br />
Second, we mea<strong>sur</strong>e up the average expected payoffs with and without<br />
deviation with real payoffs with and without deviation. The results are presented in<br />
Table 3.5. We observe that deviating is in general gainful, for only BDM offerers are<br />
penalized for having moved from their anchor (they get a negative payoff on average)<br />
which is un<strong>sur</strong>prising in view of the fact that the exogenous market-clearing price<br />
makes it inevitably a naïve strategy. Within NPA, adjusting the discovered value<br />
upon posted prices paid in both expected and real scenarios.<br />
120
Table 3.5. Comparison between extra expected and real gains from deviation<br />
on average BDM WTP BDM WTA NPA WTP NPA WTA<br />
Extra expected gain<br />
from deviation 0.13 –0.36 0.72 0.09<br />
Extra real gain<br />
from deviation 0.13 –0.22 0.26 0.08<br />
3.6. Concluding remarks<br />
The validity of incentives for truthful value revelation is questioned whenever<br />
someone‟s probability of winning depends on the moves of others, such as with<br />
endogenous market-clearing price auctions. Still, should this imply that results<br />
obtained from experiments in the random nth-price auction have no meaning because<br />
of the risk of uniform reference point adaptation? It amounts to saying that<br />
experimentalists have to choose between the absence of market learning under BDM<br />
and the risk of dependence of private values that exists under NPA.<br />
In repeated-round experimental auctions, the private-value-independence<br />
assumption behind the incentive-compatibility may be unrealistic and malapropos.<br />
When bids get correlated, the observed bid for a good after a round impacts the<br />
estimated price of the good at the next round. Individuals then revise their beliefs to<br />
reflect this information. With endogenous market-clearing prices, we believe that<br />
bidders start their valuation with a naïve anchor – their first reference point – and then<br />
adjust their value using market reference points encoded in the sequential price<br />
weighting function. They sort prices from present to past and weight these prices<br />
using an s-shape function.<br />
Although quite simple and sometimes looked upon with a critical eye, our<br />
behavioral model underlines the validity of incentive-compatibility of both the BDM<br />
and NPA auction mechanisms. Contrary to conventional models, it shows that<br />
accounting for posted prices without rejecting the incentive-compatibility enables to<br />
differentiate sincere from insincere bidding or offering. Until some proper method<br />
121
enables to distinguish learning from affiliating, we believe incentive-compatibility<br />
need not be excluded in presence of reference point adaptation, as long as one verifies<br />
the heavy weighting of the anchor. We thus suggest a different form of rationality<br />
within incentive-compatibility constraints where the correlated equilibrium plays a<br />
key role.<br />
Nevertheless, the nature of the market-clearing price plays a significant role.<br />
When it is endogenous, i.e. issued from the bids of offers instead of drawn from a<br />
uniform price list, subjects tend to fix it and refine their reference point according to<br />
it, even if it is randomly chosen. In detail, our results suggest that bidders tend to<br />
overstate their bids as if posted prices were of the WTA level, because these<br />
incorporate lifting behavioral effects. Accordingly, market discipline and competition<br />
seem not only to reveal preferences and to moderate early loss aversion, but also to<br />
unveil belated disappointment aversion and competitive pres<strong>sur</strong>e which can arise<br />
when buyers interactively value an unfamiliar nonmarket good. This avenue of<br />
research requires more attention.<br />
Instead of condemning behaviors that tie in and considering the NPA<br />
mechanism as a lesser evil, we believe that the good approach is to investigate<br />
conditions under which incentive-compatibility constraints can be remade. In this<br />
case, the notion of truth, which is undeniably contingent on human perception,<br />
convention, and social experience, should be reformulated. Our model is one attempt.<br />
3.7. References<br />
Alfnes F. and Rickertsen K. (2003), “European Consumers‟ Willingness to Pay for<br />
U.S. Beef in Experimental Auction Markets”, American Journal of Agricultural<br />
Economics, 85: 396–405.<br />
Arkes, H., Hirshleifer, D., Jiang, D. and Lim, S. (2008), “Reference Point Adaptation:<br />
Tests in the Domain of Security Trading”, Organizational Behavior and Human<br />
Decision Processes, 105: 67–81.<br />
Aumann, R. (1974): “Subjectivity and Correlation in Randomized Strategies,”<br />
Journal of Mathematical Economics, 1: 67–96.<br />
Aumann, R. (1987): “Correlated Equilibrium as an Expression of Bayesian<br />
Rationality,” Econometrica, 55: 1–18.<br />
122
Baucells, M., Weber, M. and Welfens, F. (2008), “Reference Point Formation and<br />
Updating: Theory and Experiments”, Working paper.<br />
Becker, G., DeGroot, M., and Marschak, J. (1964), “Mea<strong>sur</strong>ing Utility by a Single<br />
Response Sequential Method”, Behavioral Science, 9: 226–232.<br />
Bernard, J. (2005), “Evidence of Affiliation of Values in a Repeated Trial Auction<br />
Experiment”, Applied Economics Letters, 12: 687–691.<br />
Bettman, J. (1988), “Processes of Adaptivity in Decision Making”, Advances in<br />
Consumer Research, 15: 1–4.<br />
Bleichrodt, H. and Pinto, J. (2000), “A Parameter-Free Elicitation of the Probability<br />
Weighting Function in Medical Decision Analysis”, Management Science, 46:<br />
1485–1496.<br />
Bohm, P., Linden, J. and Sonnegard, J. (1997), “Eliciting Reservation Prices: Becker-<br />
DeGroot-Marschak Mechanisms vs. Markets”, Economic Journal, 107: 1079–<br />
1089.<br />
Boutilier, C., Goldszmidt, M., Monteleoni, C. and Sabata, B. (2000), “Resource<br />
Allocation Using Sequential Auctions”, LNCS: Agent Mediated Electronic<br />
Commerce II, 1788: 131–152.<br />
Chen, H. and Rao, A. (2002), “Close Encounters of Two Kinds: False Alarms and<br />
Dashed Hopes”, Marketing Science, 21: 178–196.<br />
Cherry, T., Crocker, T. and Shogren, J. (2003), “Rationality Spillovers”, Journal of<br />
Environmental Economics and Management, 45: 63–84.<br />
Corrigan, J. and Rousu, M. (2006), “Posted Prices and Bid Affiliation: Evidence from<br />
Experimental Auctions”, American Journal of Agricultural Economics, 88: 1078-<br />
1090.<br />
Cox, J. and Grether, D. (1996), “The Preference Reversal Phenomenon: Response<br />
Mode, Markets and Incentives”, Economic Theory, 7: 381–405.<br />
Cubitt, R., Starmer, C. and Sugden, R. (2001), “Discovered Preferences and the<br />
Experimental Evidence of Violations of Expected Utility Theory”, Journal of<br />
Economic Methodology, 8: 385–414.<br />
Cummings, R. and Taylor, L. (1999), “Unbiased Value Estimates for Environmental<br />
Goods: A Cheap Talk Design for the Contingent Valuation Method”, American<br />
Economic Review, 89: 649–65.<br />
Dasgupta, P. and Maskin, E. (2000), “Efficient Auctions”, Quarterly Journal of<br />
Economics, 115: 341–388.<br />
123
<strong>Dragicevic</strong>, A. and Ettinger, D. (2009), “Private Valuation of a Public Good in Three<br />
Auction Mechanisms”, Working paper.<br />
Einhorn, H. and Hogarth, R. (1985), “Ambiguity and Uncertainty in Probabilistic<br />
Inference”, Psychological Review, 92: 433–461.<br />
Fox, J., Shogren, J., Hayes, D. and Kliebenstein, J. (1998), “CVM-X: Calibrating<br />
Contingent Values with Experimental Auction Markets”, American Journal of<br />
Agricultural Economics, 80: 455–65.<br />
Fudenberg, D. and Levine, D. (1998), “The Theory of Learning in Games”, MIT<br />
Press.<br />
Gneezy, U. (2005), “Updating the Reference Level: Experimental Evidence”, in R.<br />
Zwick and A. Rapoport (eds), Experimental Business Research Volume III,<br />
Springer:263–284.<br />
Goeree, J., Holt, C. and Palfrey, T. (2002), “Quantal Response Equilibrium and<br />
Overbidding in Private-Value Auctions”, Journal of Economic Theory, 104: 247–<br />
272.<br />
Hanemann, M. (1994), “Valuing the Environment Through Contingent<br />
Valuation”, Journal of Economic Perspectives, 8: 19–43.<br />
Hart, S. (2005), “Adaptive Heuristics”, Econometrica, 73: 1401–1430.<br />
Hausch, D. (1986), “Multi-object auctions: Sequential vs simultaneous<br />
sales”, Management Science, 32: 1599–1610<br />
Hayes, D., Shogren, J., Shin, S. and Kliebenstein, J. (1995), “Valuing Food Safety in<br />
Experimental Auction Markets”, American Journal of Agricultural Economics,<br />
77: 40–53.<br />
Horowitz, J. (2006a), “What do we know about the performance of the Becker-<br />
DeGroot-Marschak mechanism?”, available at<br />
http://faculty.arec.umd.edu/jhorowitz/BDM-Empirics-1.doc.<br />
Horowitz, J. (2006b), “The Becker-DeGroot-Marschak mechanism is not necessarily<br />
incentive compatible, even for non-random goods”, Economics Letters, 93: 6–11.<br />
Jeitschko, T. (1998), “Learning in Sequential Auctions”, Southern Economic Journal,<br />
65: 98–112.<br />
Kagel, J. (1995), “Auctions: A Survey of Experimental Research.”, in J. Kagel and A.<br />
Roth (eds), Handbook of Experimental Economics, Princeton University Press:<br />
501–585.<br />
124
Kahneman, D., Knetsch, J. and Thaler, R. (1990), “Experimental Test of the<br />
Endowment Effect and the Coase Theorem”, Journal of Political Economy, 98:<br />
1325–1348.<br />
Klemperer, P. (1999), “Auction Theory: A Guide to the Literature”, Journal of<br />
Economic Surveys, 13: 227–286.<br />
Knetsch, J., Tang, F. and Thaler, R. (2001), “The Endowment Effect and Repeated<br />
Market Rounds: Is the Vickrey Auction Demand Revealing?”, Experimental<br />
Economics, 4: 257–69.<br />
Koszegi, B. and Rabin, M. (2006), “A Model of Reference-Dependent Preferences”,<br />
Quarterly Journal of Economics, 121: 1133–1165.<br />
List, J. and Shogren, J. (1999), “Price Information and Bidding Behavior in Repeated<br />
Second-Price Auctions?”, American Journal of Agricultural Economics, 81: 942–<br />
949.<br />
List, J. (2001), “Do Explicit Warnings Eliminate the Hypothetical Bias in Elicitation<br />
Procedures? Evidence from Field Auctions for Sportscards”, American Economic<br />
Review, 91: 1498–1507.<br />
List, J. (2003), “Using random nth-Price Auctions to value Non-Market Goods and<br />
Services”, Journal of Regulatory Economics, 23: 193–205.<br />
Lusk, J., Daniel, M., Mark, D. and Lusk, C. (2001), “Alternative Calibration and<br />
Auction Institutions for Predicting Consumer Willingness to Pay of<br />
Nongenetically Modified Corn Chips”, Journal of Agricultural and Resource<br />
Economics, 26: 40–57.<br />
Lusk, J. and Rousu, M. (2006), “Market Price Endogeneity and Accuracy of Value<br />
Elicitation Mechanisms”, in List, J. (ed.), Using Experimental Methods in<br />
Environmental and Resource Economics, Northhampton, MA: Edward Elgar<br />
Publishing, 21–40.<br />
Lusk, J. and Shogren, J. (2007), “Incentive compatible auctions: theory and<br />
evidence”, in Experimental Auctions, Methods and applications in Economic and<br />
Marketing Research, Cambridge University Press, 19–27.<br />
Lusk, J., Alexander, C. and Rousu, M. (2007) "Designing Experimental Auctions for<br />
Marketing Research: The Effect of Values, Distributions, and Mechanisms on<br />
Incentives for Truthful Bidding”, Review of Marketing Science, 5: 1–30.<br />
Milgrom, P. and Weber, R. (1982), “A Theory of Auctions and Competitive<br />
Bidding”, Econometrica, 50: 1089–1122.<br />
Morrison, G. (2000), “WTP and WTA in Repeated Trial Experiments: Learning or<br />
Leading?”, Journal of Economic Psychology, 21: 57–72.<br />
125
Payne, J., Bettman J. and Johnson, E. (1988), “Adaptive Strategy Selection in<br />
Decision Making”, Journal of Experimental Psychology: Learning, Memory, and<br />
Cognition, 14: 534–552.<br />
Plott, C. (1996), “Rational individual behavior in markets and social choice<br />
processes: the Discovered Preference Hypothesis”, in Arrow, K., Colombatto, E.,<br />
Perleman, M. and Schmidt, C., (eds.), Rational Foundations of Economic<br />
Behavior, Macmillan and St. Martins London, 225–250.<br />
Saran, R. and Serrano, R. (2007), “The Evolution of Bidding Behavior in Private-<br />
Values Auctions and Double Auctions”, IMDEA Working Paper Series in<br />
Economics and Social Sciences, 2007/10.<br />
Shlomit, H., Monderer, D. and Sela, A. (1998), “A Learning Approach to<br />
Auctions”, Journal of Economic Theory, 82: 65–88.<br />
Shogren, J., Shin, S., Hayes, D. and Kliebenstein, J. (1994), “Resolving Differences<br />
in Willingness to Pay and Willingness to Accept”, American Economic Review,<br />
84: 255–270.<br />
Shogren, J. and Hayes, D. (1997), "Resolving Differences in Willingness to Pay and<br />
Willingness to Accept: Reply," American Economic Review, 87: 241–244.<br />
Shogren, J., Cho, S., Koo, C., List, J., Park, C., Polo, P. and Wilhelmi, R. (2001a),<br />
“Auction Mechanisms and the Mea<strong>sur</strong>ement of WTP and WTA”, Resource and<br />
Energy Economics, 23: 97–109.<br />
Shogren, J., Margolis, M., Koo, C. and List, J. (2001b), “A Random nth-price<br />
Auction”, Journal of Economic Behavior and Organization, 46: 409–421.<br />
Spranca, M., Minsk, E. and Baron, J. (1991), “Omission and Commission in<br />
Judgement and Choice”, Journal of Experimental Social Psychology, 27: 76–105.<br />
Tversky, A. and Kahneman, D. (1974), “Judgment Under Uncertianty: Heuristics and<br />
Biases”, Science, 185: 1124–1130.<br />
Tversky, A. and Kahneman, D. (1992), “Advances in Prospect Theory: Cumulative<br />
Representation of Uncrtainty”, Journal of Risk and Uncertainty, 5: 292–323.<br />
Tversky, A. and Wakker, P. (1995), “Risk Attitudes and Decision Weights”,<br />
Econometrica, 63: 1255–1280.<br />
Van de Kuilen, G. (2009), “Subjective Probability Weighting and The Discovered<br />
Preference Hypothesis”, Theory and Decision, 67: 1–22.<br />
126
3.8. Appendix<br />
Proof of Proposition 3.1.<br />
Let us give evidence through a numerical example that deviation in repeated-<br />
round auctions, in terms of adaptation of the reference point, is preferable.<br />
Suppose an initial reference point value r . The last posted price amounts r � 1.<br />
The bidder faces an equiprobable trend of the value, i.e. bullish r � 1 or bearish<br />
r � 1.<br />
She can update her reference point according to the last posted price.<br />
Updating a reference point equals to the average differential such as:<br />
v �<br />
�r � r �1� � 0.5�r �1� r� � 0.5�r �1� r�<br />
3<br />
Let us now compare cases with and without adaptation of the reference point. In<br />
spite of the last posted price, the reference point is not updated and remains at r .<br />
In this case, the expected value is:<br />
� 1 � 1<br />
v � ( r � r �1) � 0.5( r �1 � r) � 0.5( r �1 � r)<br />
� �1��1���0.5 � 2 � 3<br />
The reference point is updated to r � 1 due to the last posted price. In this case,<br />
the expected value is:<br />
127<br />
� � 1<br />
v � ( r �1� r �1) � 0.5( r �1� r �1) � 0.5( r �1� r �1) � 2 � 0 �1 � � 1<br />
3<br />
One can directly see that v< v, which implies a larger expected payoff by<br />
reference point updating, or rationally deviating. Therefore, adaptation of the<br />
reference point is preferred to the current state of affairs.
Consider the following payoff matrix with a mixed strategy. Now suppose a third<br />
trusted party posts the market-clearing price which reveals some public signal.<br />
Each player has an incentive to rationally deviate instead of sticking to her anchor<br />
with some positive probability, if the public signal instructs to do so, by learning<br />
she has no chance of winning any positive payoff. While each player wants to<br />
deviate and to increase her expected payoff, she hopes that the other player does<br />
not act alike, as her payoff can depend on the bid of the other player: illustrating<br />
the endogeneity of the market price. Therefore, we have:<br />
� Each player discovers alone her preferences and thus her expected payoff � .<br />
� A player deviates to increase her expected payoff to � � 1 if she finds out that<br />
her expected payoff is close to zero, but she hopes that the other player does<br />
not move from her own anchor; conversely, she expects a payoff of � � 2 if<br />
she stays while the other player deviates.<br />
� The market price comes from the bids. When a player follows the market price<br />
or v� p,<br />
she risks a null payoff because v� p� � , explaining the absence of<br />
payoff in the last cell where all bids converged.<br />
stay deviate<br />
stay � , � � � 2 , � � 1<br />
deviate � � 1,<br />
� � 2 0 , 0<br />
The third party only tells each player what she is supposed to do. There is a<br />
correlated equilibrium if no player refuses to follow the instruction. So if the row<br />
player receives the signal „deviate‟ given she has no chance to win some positive<br />
payoff, she has no incentive not to follow, because she can make a positive payoff<br />
by deviating from her anchor, which is better in expectation. The row player<br />
assigns a positive conditional probability of 0.5 to each of the two pairs of signals<br />
(stay, deviate) and (deviate, deviate). If the column player follows the same rule,<br />
the (uncorrelated) expected payoff of the mixed strategy equilibrium by:<br />
128
1 1 1<br />
� � � � 2 � 2� � 2 � � � 1<br />
2 2 2<br />
- staying is: � � � � � �<br />
1 1 1<br />
� �1 � 0 � � �1 � 0.5� � 0.5<br />
2 2 2<br />
- deviating is : � � � � � �<br />
The expected payoff must be positive, that is, the row player‟s expected gain of<br />
staying must verify � >1,<br />
whereas her expected gain of deviating must verify<br />
� > � 1.<br />
Therefore, she is better-off by deviating, meaning that she is better-off by<br />
seeking the higher expected payoff. The game being symmetrical, the column<br />
player has no incentive not to follow her instruction either. We know that a player<br />
will never refuse to follow the recommendation resulting from the public signal in<br />
case of higher expected payoff.<br />
If we now look at the correlated equilibrium, as there is necessarily one, it yields<br />
probabilities of ⅓ to each combination that yields some positive outcome. In this<br />
case, the expected payoff verifies � greater than ⅓, which in return weakly<br />
dominates the strategy of staying (sticking to the anchor). Provided that expected<br />
payoffs are increased, each player takes the public price into account, making the<br />
decisions correlated and bids follow the same trend, which induces „affiliation of<br />
values‟ in view of the standard rationality. �<br />
Proof of Proposition 3.2.<br />
Then, let us show that a bidder who updates her reference point is sincere if she<br />
assigns a high weight on the anchor or she is subject to high inertia. Suppose a<br />
low-type bidder value and posted market-clearing prices issued from high-type<br />
bidders, such that market prices are greater than the anchor. For the purpose, let us<br />
once again take a numerical example. Assume a weight of 0.69 for the anchor and<br />
the last posted price and a weight of 0.61 for the in-between, after the losses and<br />
129
gains factors in Tversky and Kahenman (1992). Assume there are five rounds at<br />
stake.<br />
With an unnormalized uniform weighting, we obtain the following cumulative<br />
weighting:<br />
� 1��1��1��1��1��1� 0.61� � �0.61� � �0.61� � �0.61� � �0.61� � �0.61�<br />
�<br />
� 6��6��6��6��6��6� last price in-between posted prices anchor<br />
� 6� 0.102 � 0.610<br />
As one can see, each round receives an equal weight, which means that the anchor<br />
is drowned in time by the sequence of posted prices.<br />
With an unnormalized s-shaped weighting, we obtain the following cumulative<br />
weighting:<br />
� � 1 � � � � 1 � � 1 �� � � 1 ��<br />
�0.69� � � 0� � �0.61�1� � � 0.61� �� � �1� 0.69�1 � ��<br />
� � 6 � � � � 6 � � 6 �� � � 6 ��<br />
last price in-between posted prices anchor<br />
� �<br />
� 0.115 � 4� 0.102 � 0.425 � 0.947<br />
This graphical representation corresponds to the example of cumulated ranks on<br />
the x-axis and accumulated weight on the y-axis:<br />
130
1� 0.69�5 6�<br />
0.61�4 6�<br />
0.69�1 6� � 0<br />
�16� �56� As we can see, the anchor, i.e. the right-hand side of the graph, receives the<br />
highest weight. If the asocial valuation gets a high weight on the topic of the value<br />
refinement in time, the bidder‟s valuation is not fully captured by the sequential<br />
market-clearing prices. Regarding our low-type bidder, the risk of deviating from<br />
her anchor, while she updates her reference point, is lower with the s-shaped<br />
weighting than with the uniform weighting. �<br />
131
132
133<br />
Chapter 4<br />
Competitive Private Supply of Public Goods<br />
Abstract<br />
This chapter compares guilt alleviation and competition for social status in the private<br />
provision of a public good. When agents are intrinsically impulsed, that is, they<br />
mostly provide the public good in order to alleviate their guilt, they tend to free-ride.<br />
In contrast, when agents are extrinsically impulsed and compete for social status,<br />
their provisions become strategic complements. In the latter case, the aggregate level<br />
of the public good increases as the disparity between agents‟ incomes shrinks.<br />
Injecting competition for social status into utility functions increases provisions to a<br />
public good, and hence its aggregate level. Market competition thus creates incentives<br />
to overcome the free-riding issue.<br />
Keywords: public good private supply, guilt relieving, social status, competition,<br />
income transfer<br />
JEL Classification: A13, C7, H41
4.1. Introduction<br />
134<br />
"Guilt is the price we pay willingly<br />
for doing what we are going to do<br />
anyway." Isabelle Holland<br />
The voluntary offset market enables agents to pay for their negative<br />
externalities issued from carbon emissions by investing in projects that reduce<br />
emissions or sequester carbon, such as tree planting or renewable energy. The<br />
reduction of carbon emissions is a public good because, once provided, agents can<br />
enjoy the benefits devoid of rivalry, without excluding anyone from its consumption.<br />
Some people believe that the voluntary offset market is inefficient. One of the<br />
arguments put forward is that offsetting validates polluting behavior. Likewise,<br />
offsetting is said to operate like charities: voluntary supplies never provide enough<br />
public good because of the free-rider incentive. And when private arrangements<br />
finance a public good, free-riding on other people‟s provisions is rational.<br />
However, free-riding is limited to some extent because agents who purchase<br />
offsets may also derive private benefits. Olson (1965) advances the hypothesis that<br />
free-riding can be overcome through social incentives. According to him, agents do<br />
not privately supply a public good for its direct material benefit, but to achieve social<br />
objectives like prestige or respect; this would explain why individuals do less free-<br />
riding than what the economic theory suggests. Following this rationale, Hawkes et<br />
al. (1993) show that in ancient times hunters and gatherers tended to share their<br />
resources because the cost of exclusion from the group – where every agent prefers a<br />
supplier to a consumer as a neighbor – was too high to risk, thus making resources a<br />
public good.<br />
This impure approach of pro-social behavior has been modeled by Andreoni<br />
(1990) who justifies private provisions in terms of warm-glow or joy-of-giving. Our<br />
approach differs from Andreoni‟s and rejoins Olson‟s, for we consider social status<br />
gained by agents who privately supply a public good from its relative perspective. As<br />
a matter of fact, supplying to the public good can generate benefits of guilt relief –<br />
which we find more convincing than warm-glow – and/or social status. In the first
case, agents want to feel better about themselves, because they want to recover self-<br />
esteem after producing a public bad. If an agent feels guilty, because she believes she<br />
bears responsibility for carbon excesses, then guilt alleviation through carbon<br />
offsetting is a private benefit derived from the supply of the public good 46 . Despite<br />
the private benefit, the motivation for it is internal. It is thus an intrinsic incentive.<br />
Since guilt arousal is positively related to donation intention (Hibbert et al. 2007),<br />
guilt alleviation has positive impacts on environmental awareness. Then, agents<br />
compete to be formally acknowledged as being the most concerned about the public<br />
good. This prosocial behavior can be due to social pres<strong>sur</strong>e and norms and<br />
corresponds to an extrinsic incentive. An agent who offsets receives a proof<br />
acknowledging her provision to the public good. She thus sends a signal to make<br />
other agents aware of her polluting abatement. Following this rationale, producers<br />
will also promote their offsets as part of their corporate social responsibility policy<br />
(Kotchen 2009).<br />
People have a preference for showing altruism in situations that facilitate<br />
broadcast opportunities, and the provision of a public good is certainly one such<br />
situation (Smith and Bliege Bird 2000). De facto, what type of incentives should be<br />
introduced to increase private provisions? Are competitive settings such as auctions a<br />
good solution to the inefficient provision of a public good? Do agents become more<br />
generous by guilt or by craving for social status?<br />
If high status brings with it high earnings, then status seeking behavior can be<br />
explained as a part of economic behavior (Ball and Eckel 1998). According to<br />
competitive altruism, despite the dearness of being publicly generous, agents can<br />
promote their generosity as potential exchange partners, reaping the benefits later on<br />
(Roberts 1998). Agents also refuse transactions that are in their best economic interest<br />
when they feel they are an insult to their dignity (Bénabou and Tirole 2006).<br />
Experimental literature has confirmed the role of individual status as an incentive<br />
affecting market outcomes (Ball et al. 2001) and donors (Duffy and Kornienko 2005).<br />
Because of the rivalry and excludability in social hierarchy, agents have to compete<br />
46 Gilbert (1997) speaks about membership guilt over group wrongs. This collective guilt will be<br />
shared by members of the collective in question in their capacity as group members.<br />
135
efore attaining some desired social status: if an agent desires to be the first or among<br />
the first in some venture, she might have to make the most efforts to reach her goal.<br />
Making the most efforts means that she has knowledge of her challengers and of the<br />
efforts she has to invest. In this case, how does competition influence an agent‟s<br />
voluntary supply of a public good? Competitive mechanisms, such as contests, have<br />
shown to increase the voluntary provision of a public good (Kolmar and Wagener<br />
2008).<br />
This chapter investigates how competition influences private provisions of the<br />
public good when agents are stirred by an intrinsic impulse, meaning that they mainly<br />
maximize utility from guilt relief, as opposed to when they are stirred by an extrinsic<br />
impulse, suggesting that they mainly maximize utility from social status. Our public<br />
good game unveils several results: first, we find that when status seeking dominates<br />
guilt relief, private provisions become strategic complements: an attribute which<br />
increases the aggregate level of the public good. Then, we prove sufficient conditions<br />
for existence and uniqueness of a Nash equilibrium. At last, when agents behave<br />
according to their best-response functions, we find that the aggregate level of the<br />
public good depends on the disparity between agents‟ incomes, which – depending on<br />
the nature of the provisions – induces a particular income transfer policy.<br />
We give a basic account of the social status function and present the public<br />
good game in Section 4.2. We provide a model of logarithmic best-response functions<br />
and describe explicit properties of a Nash equilibrium in Section 4.3. Concluding<br />
comments are given in Section 4.4.<br />
4.2. The public good game<br />
Let us first introduce the social status function. Consider n agents who<br />
produce the public good by devoting some of their endowment w into the public good<br />
g . Following Frank (1985), let us suppose that each agent cares about her social<br />
status with respect to the other n � 1 agents.<br />
136
Definition 4.1.: The social status function is a continuously differentiable function<br />
� , �<br />
si � s gi g�i where i g is the provision of agent i, g� i is the provision of other<br />
agents. The level of the provision to the public good determines social status. If<br />
f �g � is the density function for g values which determines the social status of the<br />
agents and g 0 is the smallest provision to the public good g among the n agents, then<br />
an agent with 0 < g g n will have a social status function such that:<br />
g<br />
� � � �<br />
s g � � f g dg<br />
[1]<br />
n<br />
g<br />
0<br />
where f �g � increases as g moves towards the maximum value of its domain.<br />
Let us consider two agents i and j, with j � i.<br />
Let w i be agent i‟s endowment,<br />
let x i denote her consumption of the private good, let G be the aggregate level of<br />
public good and let g i account for her provision to the public good. The aggregate<br />
level of public good is the sum of the two agents‟ provisions G � gi � g j.<br />
Agent i‟s<br />
social status is determined by her relative contribution si � gi �gj<br />
preferences represented by the following utility function:<br />
� , , �<br />
i i i i<br />
137<br />
47 . Agents have<br />
u � u x G s<br />
[2]<br />
Considering agent j‟s provision g j as exogenous, agent i maximizes her<br />
utility by solving the following program:<br />
x , g<br />
� �<br />
max u x , G, s subject to xi �gi� wi<br />
and g � 0<br />
[2']<br />
i i<br />
i i i<br />
47 According to Auriol and Renault (2008), social status is a scarce resource: increasing an agent‟s<br />
status requires that another agent‟s status decreases.<br />
i
Let us now determine the Nash equilibrium of the public good game. Each<br />
agent‟s best-response function fully specifies her equilibrium strategy. This strategy<br />
involves choosing a level of private supply to the public good, the private supply of<br />
the other agent being exogenous. We first analyze the best response functions of each<br />
agent. We thus study the two motives for contributing to the public good: to relieve<br />
guilt and to acquire social status.<br />
Assume the marginal utility from the provision to the public good to be:<br />
� �<br />
H x G s<br />
�u �u �u<br />
� � � [3]<br />
i i i<br />
i i, , i � � �<br />
G si xi<br />
The first term denotes the marginal utility from the public good. The second term<br />
represents the marginal efficacy of social status. The last term is the marginal fall in<br />
the consumption of private goods. We then make three assumptions on H .<br />
A1:<br />
�H � u � u � u<br />
� � � � 0<br />
�x �x �G �x �s �x<br />
2 2 2<br />
i i i i<br />
2<br />
i i i i i<br />
A1 says that an increase of income increases the marginal utility of the supply of the<br />
public good. The assumption is referred to as the normality assumption because it is<br />
satisfied if we assume that both private and public goods are normal with respect to<br />
income. It simply says that agent i‟s demand for the public good increases with<br />
income and her demand for private goods does not decrease with income.<br />
A2:<br />
�H � u � u � u<br />
� � � � 0<br />
�G �G �G�s �G�x 2 2 2<br />
i i<br />
2<br />
i i<br />
i i i<br />
A2 states that the marginal utility of the public good decreases with G. As a matter of<br />
fact, if the level of the public good increases independently of agent i‟s supply, there<br />
138<br />
[4]<br />
[5]
is no incentive to contribute to the public good. This is a formal foundation for the<br />
free-riding issue. Considering negative externalities, it simply means that any agent<br />
can compensate for the damage caused, and all agents can profit from its reparation 48 .<br />
A3:<br />
�H � u � u � u<br />
� � � � 0<br />
�s �s �G �s �s �x<br />
2 2 2<br />
i i i<br />
2<br />
i<br />
i i i i i<br />
A3 implies that an increase in social status creates negative incentives: the agent<br />
tends to reduce her supply to the public good, because she no longer has to compete<br />
for social status.<br />
According to the previous assumptions and following the work on warm-glow<br />
by Andreoni (1990), we now consider that individuals obtain guilt relief and social<br />
status from their private supply of the public good. Following the first order<br />
condition, agent i‟s best response, that is, � , �<br />
� �<br />
ri wi g i , is to have i<br />
139<br />
g such as:<br />
r � H w � g , g � g , g � g � 0<br />
[7]<br />
i i i i j i j<br />
A Nash equilibrium of the public good game is a couple of strategies<br />
each strategy is the best response to the other agent‟s strategy:<br />
� , �<br />
* *<br />
i i i j<br />
* *<br />
i j<br />
[6]<br />
g , g such that<br />
g � r w g with j � i<br />
[8]<br />
Let us now look at the second order condition to see whether contributing to<br />
the public good does in fact maximize an agent‟s function. The second order<br />
condition is satisfied for:<br />
48 According to Gilbert (1997) since feeling guilt is unpleasant, it is liable to move one who feels it to<br />
act. And this will not necessarily be the personal undertaking of reparative action.
dHi �Hi �Hi �Hi<br />
� � � � 0<br />
dg �G �s �x<br />
i i i i<br />
The sign of the differential implies a diminishing marginal utility of the public good<br />
as the agent supplies the public good. Negativity depends on three terms. The first<br />
term mea<strong>sur</strong>es the outcome of any provision to the public good on the marginal utility<br />
of the public good. This is our indicator of free-riding. The second term values the<br />
outcome of a shift in the social status on the marginal utility of the public good. It<br />
allows us to study the interactions between the aggregate level of the public good and<br />
social status in the utility function. The third term assesses the impact of a decrease in<br />
private goods‟ consumption on the marginal utility of the public good.<br />
Let us now consider the effect of agent j‟s supply on the marginal utility of<br />
agent i‟s supply:<br />
dH �H�H � �<br />
dg �G�s i i i<br />
j i i<br />
This effect is ambiguous, for the first term is negative while the second one is<br />
positive. The first term denotes a typical free-riding issue: an increase of agent j‟s<br />
provision reduces agent i‟s incentive to contribute; except that the second term<br />
denotes status seeking, thus an opposite effect, as social status decreases with agent<br />
j‟s supply. Indeed, agent i suffers from the reduction in the level of public good due<br />
to carbon emissions, thus any private provision that increases the public good also<br />
increases agent i‟s utility. Provided that any supply removes her feelings of guilt, she<br />
can free-ride on others‟ provisions and allocate all her endowment to the private<br />
goods instead. This is a counter-incentive to supply the public good. In parallel, agent<br />
i suffers from status loss in social hierarchy every time others supply the public good.<br />
Therefore g j is also an incentive to contribute in order to maintain the level of social<br />
status.<br />
The sign of the best-response function slope of agent i is:<br />
140<br />
[9]<br />
[10]
�ri �<br />
dH dg<br />
i j<br />
�g�dHdg j i i<br />
The sign depends on which effect prevails: guilt relieving or status seeking.<br />
According to the terms of Bulow et al. (1985), if free-riding dominates social<br />
hierarchy or �r � g >0,<br />
we are in the presence of strategic substitutes, and strategic<br />
i j<br />
complements vice versa. Despite the fact that in standard public good games (even in<br />
the presence of an impure public good) the only effect at stake is free-riding and<br />
public good provisions are always strategic substitutes: injecting competition for<br />
social status converts the provisions into strategic complements in some cases.<br />
141<br />
[11]<br />
A Nash equilibrium is a set of provisions that satisfies the aggregation of<br />
supplies. Let us prove the existence and uniqueness of a Nash equilibrium. For a<br />
Nash equilibrium between agents to exist, one must verify:<br />
dHi dg j dH j dgi<br />
, 1,1<br />
dH dg dH dg ��<br />
� �<br />
i i j j<br />
� �<br />
The slopes of the best-response functions are bounds within the interval �� 1,1�<br />
. The<br />
binding conditions are sufficient for the existence of a unique Nash equilibrium.<br />
Proposition 4.1.: If [12] is satisfied, there exists a unique Nash equilibrium.<br />
Proof: In the appendix.<br />
[12]<br />
Let us now see what happens when the policy of income transfer is instituted.<br />
Consider the ratio which confronts the two motives involved in the public good‟s<br />
supply. The expression returns to an intrinsic impulse coefficient such as:<br />
�<br />
�H �x<br />
i i<br />
i �<br />
� i � i � � i � i<br />
� H x � �2 H s �<br />
[13]
The numerator mea<strong>sur</strong>es the marginal utility of the public good and stands for<br />
the intrinsic (contrite) impulse of guilt relief to supply the public good. It depends on<br />
agent i‟s income and thus on her opportunity loss when she doesn‟t purchase the<br />
private goods. Here, agent i is indifferent between consuming her own supply or<br />
benefiting from agent j‟s supply of the public good. In Andreoni‟s terminology, this<br />
phenomenon means pure altruism or selflessness of agent i. Here, we consider the<br />
numerator as a mea<strong>sur</strong>e of free-riding on others‟ provisions.<br />
The denominator represents the influence of social status on the marginal<br />
utility of the public good and stands for the extrinsic (social) impulse of status<br />
seeking to supply the public good. Just as with the numerator, it depends on agent i‟s<br />
income, but it depends on social status above all, that is, marginal utility of the public<br />
good derived from her own provision (analogue to Andreoni‟s impure altruism).<br />
Given that status is acquired by relative provisions, the effect of social status counts<br />
twice. First, consuming more of the x‟s decreases agent i‟s provision to the public<br />
good and thus her social status; second, more of g j implies lower social status for<br />
agent i, all else being equal. For those reasons, the intrinsic impulse coefficient is<br />
inversely proportional to status seeking.<br />
Proposition 4.2.: An income transfer from agent j to agent i , such that<br />
dwi � �dwj � 0 increases G if and only if � > � .<br />
Proof: In the appendix.<br />
i j<br />
Agents are unwilling to perfectly substitute their provisions to offset a<br />
transfer. If �i> � j then agent i can be considered to be less status seeking than agent<br />
j. Hence, the policy of income transfer will increase (decrease or not change) the<br />
aggregate level of the public good if and only if the income gainer is less status<br />
seeking than (more status seeking than or equally status seeking than) the income<br />
loser. This proposition is comparable to that of Andreoni, but our interpretation is<br />
different. In fact, since competition for social status encourages agents to supply the<br />
142
public good, only an increase in income will motivate the lower income agent to<br />
supply more 49 , for it enables her to compete for social status. Without transfer, her<br />
position discourages her to race for social status and she can only relieve her guilt.<br />
The direct consequence is free-riding on other agents‟ provisions. Another way of<br />
understanding the proposition is: since the higher income agent proves – with a<br />
higher level of supply which reflects higher income – to be more extrinsically<br />
impulsed, she does not have to contribute more to the public good. She is in no doubt<br />
to hold the social status ex ante.<br />
Our model is a way-out to Andreoni‟s impure altruism and warm-glow giving.<br />
What he calls pure altruism, we identify as guilt relief and free-riding, while his<br />
impure altruism corresponds to our willingness to compete for social status, which is<br />
observable via any non-anonymous donation. The model is thus an alternative and a<br />
more realistic way to explain prosocial behavior.<br />
4.3. The explicit logarithmic model<br />
4.3.1. The program<br />
Following the model by Kumru and Vesterlund (2008), agents have<br />
preferences represented by the following separable nonlinear utility function:<br />
� , , � ln � � ln �� � �<br />
u x G s � x � G � s<br />
[14]<br />
i i i i i i i<br />
where G � gi � g j and si � gi � g j.<br />
Private goods are included in the first term, while<br />
provisions are included in the second term which is nondecreasing in g i . The latter<br />
mea<strong>sur</strong>es utility derived from guilt relief based on the aggregate level of the public<br />
good G and social hierarchy s i which are separable.<br />
49 For example, OECD (2007) suggests monetary transfers in benefit of low income households when<br />
imposing environmental taxes.<br />
143
We assume that individuals originate guilt relief from their private supply of<br />
the public good. Agent i‟s preferences when she provides the public good by g i are<br />
defined by:<br />
�g g �<br />
� � for j � i<br />
[15]<br />
i i j<br />
The expression denotes the utility that agent i gets from supplying to the<br />
public good and the aggregate level of the public good scaled by a specific index<br />
� � 0 . The aggregation of provisions corresponds to the public good dimension of<br />
i<br />
the utility function. We assume that some willingness to relieve guilt is stated by<br />
either agent 50 . For example, either agent could relieve guilt with a single symbolic<br />
coin when participating in charity auctions.<br />
Agent i gets utility from social status when she provides the public good by<br />
gi 51 . Her status is given by the distance between her provision and that of agent j‟s<br />
such as:<br />
�g g �<br />
� � for j � i<br />
[16]<br />
i i j<br />
Agent i enhances her status in the social hierarchy if her provision<br />
outdistances agent j‟s; otherwise, her social status deteriorates. The status is scaled by<br />
a specific index � i , with �i � 0 , which mea<strong>sur</strong>es agent i‟s willingness to acquire<br />
social status. When agents provide identical provisions, the term vanishes. In the<br />
equilibrium, agent i knows whether she acquires social status through her private<br />
supply of the public good ( i g > g j ). The explicit maximization program is then:<br />
50 Social comparison theory suggests that individuals have a need to compare themselves to individuals<br />
whom they deem are similar to them (Goethals 1986).<br />
51 A status-based model of market competition has already been introduced by Podolny (1993).<br />
144
x , g<br />
� i � i j � � � � i � i � � �i�i�j���i�i�j� max u x , g , g , s ln w g ln � g g g g �<br />
� �<br />
i i<br />
subject to x � g � w, g � 0<br />
i i i<br />
145<br />
[17]<br />
The first term represents the utility derived from the consumption of private<br />
goods x i . The second term corresponds to the utility that agent i obtains from her<br />
supply of the public good. Agent j‟s provision is both a strategic substitute and a<br />
strategic complement of agent i‟s utility. As a strategic substitute, two obvious<br />
interpretations come out. First, agent i suffers from the public good diminishment due<br />
to carbon emissions, thus any private provision that increases the public good also<br />
increases agent i‟s utility indexed by � i . Second, since any provision removes her<br />
feelings of guilt, she can free-ride on others‟ provisions and allocate all her<br />
endowment to the consumption of the private goods. To consider agent j‟s provision<br />
as a strategic complement is to consider that agent i suffers from status loss in social<br />
hierarchy every time agent j provides the public good. Therefore, agent j‟s private<br />
provision decreases agent i‟s utility.<br />
4.3.2. Reaction functions<br />
Now suppose both agents decide to submit their provisions to the public good.<br />
Given g j , differentiating � � u � with respect to i g gives r i , agent i‟s best-response<br />
function:<br />
1 A<br />
� i<br />
wi w � i<br />
ri �wi , g j � � wi � g j if g j �max � , � �<br />
2 2<br />
�Ai Ai<br />
�<br />
A<br />
� � �<br />
� .<br />
� �<br />
i i<br />
where i<br />
i � i<br />
[18]<br />
Whether r i is constrained depends on the level of j g . For small values of g j ,<br />
agent i allocates a part of her income to the supply of public good. For sufficiently
high values of g j , agent i can supply either nothing or her full income. Whichever<br />
occurs depends on the sign of �i � �i.<br />
Corollary 4.1.: The difference between � i and � i determines whether provisions are<br />
strategic substitutes or strategic complements.<br />
Proof: In the appendix.<br />
When ><br />
i i<br />
� � or Ai � 0 , r i is the best-response only when g j � wi Ai<br />
, which<br />
is a nonnegative number. If agent j <strong>sur</strong>passes this threshold, agent i has fairly no<br />
incentive to make positive provisions. In point of fact, even a quasi-null level of i g<br />
(nonnegative by definition) enables agent i to maximize her utility by allocating her<br />
income to more private goods while alleviating her guilt through agent j‟s provisions.<br />
We could think of an individual who pays tribute to the collective high efforts in<br />
providing the public good while ending up self-pleased by giving a single coin.<br />
When �i � �i<br />
or A � 0 , agent i is equally concerned by guilt alleviation and<br />
i<br />
social hierarchy. This time, r i is equal to � � ½ w i for g j ��0, � � . Her provision is<br />
always the half of her income, but she has no incentive to contribute more than that.<br />
This is the behavior of an autonomous agent who disregards the provisions of the<br />
opponent. We could think of an individual who invariably contributes to the public<br />
good in order to alleviate her guilt – because some moral obligation incites her to do<br />
so –, but who does not discredit the positive spillover on her social rank, even if she is<br />
not centered upon the social ranking matter. This agent is either blind or denies the<br />
possibility of acting as a free-rider.<br />
At last, when <<br />
i i<br />
� � or Ai � 0 , r i holds if g j �� wi Ai<br />
, otherwise ri � wi<br />
and<br />
agent i allocates her full income to the supply of the public good. Provisions are then<br />
strategic complements: every time agent j increases her supply, agent i has an<br />
incentive to increase her supply to stay in the race for the social status up to the point<br />
where her full income is spent.<br />
146
According to the foregoing results, Fig. 4.1. illustrates the best-response<br />
functions which meet at the bisection line, observed from symmetric cases<br />
�i ��j � � and �i ��j � � . Each best-response function – initiated from the<br />
reference point which is the opponent‟s null provision – is v-shaped, i.e. separated<br />
into two segments following opposite slopes.<br />
The black straight lines depict agent i‟s best-response functions. The grey<br />
straight lines depict agent j‟s best-response functions. We have three cases: (i) when<br />
the intrinsic impulse dominates A >0,<br />
their best-response functions decrease in their<br />
opponents‟ provisions and the public good is weakly provided, for respective<br />
provisions are less than wi 2, w j 2 ; (ii) when the extrinsic impulse dominates A 0<br />
A �0<br />
wi<br />
A
�r<br />
�<br />
i<br />
igj ��<br />
�� � � �<br />
� �<br />
i i i<br />
2 0<br />
[20]<br />
The derivative �ri � �i<br />
is strictly positive and agent i‟s best-response function<br />
increases in � i , for all j g . As � i goes up, agent i is emulous and considers agent j‟s<br />
provision a threat to her status in the social hierarchy, which makes her increase her<br />
provision. In consequence, the higher � i , the higher the agent i‟s provision. The same<br />
reasoning applies to agent j.<br />
Corollary 4.3.: Agent i‟s best-response function increases in � i .<br />
4.3.3. The equilibrium
* *<br />
At a Nash equilibrium � i, j�<br />
g g each agent‟s provision is her best response to<br />
the other‟s. We first consider an interior equilibrium where both agents‟ provisions<br />
are strictly positive but inferior to their incomes:<br />
equilibrium, provisions amount:<br />
� 2w<br />
* i � Ajw j<br />
�gi<br />
�<br />
� 4 � AA i j<br />
�<br />
� 2w<br />
* j � Ai wi<br />
g j �<br />
�<br />
� 4 � AA j i<br />
� � �<br />
A � and A<br />
� �<br />
i i<br />
where i<br />
i � i<br />
j<br />
� j � � j<br />
� .<br />
� ��<br />
j j<br />
149<br />
*<br />
0 gi wi<br />
� � ,<br />
*<br />
0 gj wj<br />
� � . At such<br />
In this case, the aggregate level of the public good in equilibrium, that is,<br />
G � g � g , amounts:<br />
* * *<br />
i j<br />
1<br />
�2 � �2 �<br />
� �<br />
� �<br />
*<br />
G � � Aj wi � � Ai wj<br />
4 � AA i j<br />
[21]<br />
[22]<br />
As one can detect, when agents apply their best-response functions, the<br />
aggregate level of the public good depends on the relative distance between the social<br />
status and guilt relief indices.<br />
Corollary 4.4.: When A i increases, (i) the equilibrium provision of agent i<br />
decreases; (ii) the aggregate equilibrium quantity of public good decreases; (iii) and<br />
the equilibrium provision of agent j increases (decreases) if Aj ��� �0 .<br />
Proof: In the appendix.
A policy of income transfer from agent j to agent i such that dw � �dw � 1<br />
impacts the aggregate quantity of public good:<br />
dG dg dg<br />
� �g �g<br />
� ��g �g<br />
�<br />
� � � �<br />
�� �� �� ��<br />
i i<br />
j j<br />
� i � j � � � �<br />
�wi �wj �wi �wj<br />
That is:<br />
1<br />
dG � �Ai �A�<br />
j<br />
4 � AA � �<br />
where<br />
i j<br />
A �A �2<br />
i j<br />
� � ��<br />
�<br />
i j j i<br />
��i ��i���j��j� 150<br />
i j<br />
[23]<br />
[23']<br />
Corollary 4.5.: At interior equilibrium, an income transfer from agent j to agent i<br />
increases (decreases) the aggregate level of the public good if and only if<br />
� �<br />
� � � � � > < 0.<br />
i j j i<br />
In equilibrium, when � �� � 0 and each agent is indifferent between her<br />
i j<br />
supply and the other‟s supply, we have dG � 0 which is the standard result of<br />
neutrality obtained by Warr (1983).<br />
Let us now consider corner solutions with either null or full-income<br />
provisions. In corner equilibria, in case of strategic substitutes, one of the agents<br />
provides a null supply. In the case of strategic complements, one of the agents<br />
allocates her full income to the public good supply. If we analyze income transfers at<br />
the corner equilibria in the symmetric case, Figs. 4.2. and 4.3. depict provisions with<br />
respect to the income inequality. The x-axes denote agents‟ shares of total income:<br />
� � �,<br />
wj �wi wj<br />
�<br />
w w w<br />
i i j<br />
� . The y-axis represents the aggregate level of the public<br />
good. The total income is fixed. The broken black curve represents the provision of<br />
agent i while the broken grey curve represents the provision of agent j. The broken
grey curve decreases while the broken black curve increases as the transfer between<br />
agents j and i occurs. The equality arises at 0.5. The solid black curve illustrates the<br />
sum of provisions, i.e. the aggregate level of the public good.<br />
G<br />
G<br />
0 0.5<br />
1<br />
*<br />
g j<br />
0 0.5<br />
1<br />
Fig. 4.2. Income transfer with strategic substitutes<br />
0 0.5<br />
*<br />
g j<br />
0 0.5<br />
Fig. 4.3. Income transfer with strategic complements<br />
In the case of strategic substitutes (the standard scenario in public good<br />
games), where guilt relief prevails, the aggregate level of the public good decreases as<br />
the incomes‟ disparity shrinks. Indeed, at a corner solution, the lower income agent<br />
151<br />
*<br />
gi<br />
*<br />
gi<br />
1<br />
1
invariably free-rides on the supply of the higher income agent. If the income is<br />
transferred from the lower income agent to the higher income agent, the latter should<br />
allocate the extra income into the public good supply 52 and the aggregate quantity<br />
should increase. This is a similar result to Theorem 5 from Bergstrom et al. (1986)<br />
who show that equalizing income by transferring income from contributors to non-<br />
contributors will decrease the equilibrium supply of the public good, in the case of a<br />
pure public good ( �i � 0 in our case).<br />
In the case of strategic complements (the novel scenario in public good<br />
games), where status seeking prevails, the aggregate level of the public good<br />
decreases as the agents‟ income disparity grows. This time, the lower income agent<br />
allocates her full income to the supply of public good in order to gain social status,<br />
thus saturating her supply capacity, whereas the higher income agent contributes less<br />
than her full income. An income transfer from the higher income agent to the lower<br />
income agent should increase the quantity of public good, because the lower income<br />
agent should allocate the money transfer to the provision of the public good.<br />
G<br />
–1<br />
Aj<br />
Fig. 4.4. The aggregate level of provisions<br />
52 For example, this suggests that cutting taxes on the higher income agent and raising taxes on the<br />
lower income agent may increase private supply.<br />
0<br />
1<br />
152<br />
0<br />
Ai<br />
–1
i<br />
At last, Fig. 4.4. shows the aggregate provisions to the public good in view of<br />
A and A j , in both interior and corner equilibria. The kinks in the slope correspond to<br />
corner equilibria. When A i
Therefore, status (market) competition in the form of auctions can be an<br />
answer to free-riding 53 . Our results could explain the institution of charity auctions,<br />
honor rolls of donors and the construction of socially responsible finance indices.<br />
More generally, it could relate to why institutions make use of agents‟ willingness to<br />
demonstrate their generosity if not their apparent selflessness. To some extent, our<br />
model could be an illustration of the theory of crowding out of intrinsic motivations<br />
by extrinsic incentives. Further work consists in verifying the relevancy of these<br />
findings with field data.<br />
4.5. References<br />
Andreoni, J. (1990), “Impure Altruism and Donations to Public Goods: A Theory of<br />
Warm-Glow Giving?”, Economic Journal, Royal Economic Society, 100: 464–<br />
477.<br />
Auriol, E. and Renault, R. (2008), “Status and Incentives”, RAND Journal of<br />
Economics, 39: 305–326.<br />
Ball, S. and Eckel, C. (1998), “The Economic Value of Status”, Journal of Socio-<br />
Economics, 27: 495–514.<br />
Ball, S., Eckel, C., Grossman, P. and Zame, W. (2001), “Status in Markets”,<br />
Quarterly Journal of Economics, 116: 161–188.<br />
Bénabou, R. and Tirole, J. (2006), “Incentives and Prosocial Behavior”, American<br />
Economic Review, 96: 1652–1678.<br />
Bergstrom, T., Blume, L. and Varian, H. (1986), “On the Private Provision of Public<br />
Goods”, Journal of Public Economics, 29: 25–49.<br />
Bulow, J., Geanakoplos, J. and Klemperer, P. (1985), “Multimarket Oligopoly:<br />
Strategic Substitutes and Complements”, Journal of Political Economy, 93: 488–<br />
511.<br />
Frank, R. (1985), “The Demand for Unobservable and Other Nonpositional Goods”,<br />
American Economic Review, 75: 101–116.<br />
53 This is an opposite result to Holländer (1990) who finds that opening a market for the collective<br />
good lowers its provision.<br />
154
Gilbert, M. (1997), “Group Wrongs and Guilt Feelings”, Journal of Ethics, 1: 65–84.<br />
Goethals, G. (1986), “Social Comparison Theory: Psychology from the Lost and<br />
Found”, Personality and Social Psychology Bulletin, 12: 261–278.<br />
Hawkes, K., Altman, J., Beckerman, S., Grinker, R., Harpending, H., Jeske, R.,<br />
Peterson, N., Smith, E., Wenzel, G. and Yellen, J. (1993), “Why Hunter-<br />
Gatherers Work: An Ancient Version of the Problem of Public Goods [and<br />
Comments and Reply]”, Current Anthropology, 34: 341– 361.<br />
Holländer, H. (1990), “A Social Exchange Approach to Voluntary Cooperation”,<br />
American Economic Review, 80: 1157– 1167.<br />
Kolmar, M. and Wagener, A., (2008), “Contests and the Private Provision of Public<br />
Goods”, University of St. Gallen <strong>La</strong>w and Economics Working Paper No. 2008-<br />
27.<br />
Kotchen, M. (2009), “Offsetting Green Guilt”, Stanford Social Innovation Review,<br />
Spring 2009.<br />
Kumru, C. and Vesterlund, L. (2008), “The Effect of Status on Voluntary Provision”,<br />
Australian School of Business Research Paper No.2008 ECON 02.<br />
OCDE (2007), “L‟économie politique des taxes liées à l‟environnement”, Les<br />
Synthèses de l’OCDE, available at: www.oecd.org/env/taxes<br />
Olson, M. (1965), “The Logic of Collective Action: Public Goods and the Theory of<br />
Groups”, Harvard University Press, First edition.<br />
Podolny, J. (1993), “A Status-based Model of Market Competition”, American<br />
Journal of Sociology, 98: 829–872.<br />
Roberts, G. (1998), “Competitive altruism: From reciprocity to the handicap<br />
principle”, Proceedings of the Royal Society of London B: Biological Sciences,<br />
265: 427–431.<br />
Smith, E. and Bleige Bird, R, (2000), “Turtle Hunting and Tombstone Opening:<br />
Generosity as Costly Signaling”, Evolution and Human Behavior, 21: 245–261.<br />
Warr, P. (1983), “The Private Provision of a Public Good is Independent of the<br />
Distribution of Income”, Economics Letters, 13: 207–211.<br />
155
4.6. Appendix<br />
Proof of Proposition 4.1.<br />
* *<br />
First, � i, j�<br />
g g is a Nash equilibrium if and only if<br />
� i � j � , j �,<br />
i � and � �<br />
x r r x w w<br />
�r � �<br />
�g<br />
j<br />
r g � g .<br />
* *<br />
j i j<br />
�r � �<br />
�g<br />
i<br />
j<br />
Second, if � 1,1�<br />
and � 1,1�<br />
point and<br />
dHi dH j<br />
�ri �g j<br />
dg j<br />
�<br />
�dH dg<br />
�rj<br />
dgi<br />
, �<br />
�g<br />
�dH<br />
i<br />
dg<br />
i j<br />
i j<br />
Proof of Proposition 4.2.<br />
i<br />
156<br />
*<br />
g i is a fixed point of the function:<br />
� , , �<br />
x � r r x w w has a unique fixed<br />
, � �<br />
i j j i<br />
Consider the ratio which mea<strong>sur</strong>es the relative incentives to contribute to the public<br />
good:<br />
dHi �Hi<br />
dxi � i �<br />
dHi dHi �<br />
dg j dgi �xi<br />
�<br />
��H i �H � � i �Hi �Hi �H<br />
� i<br />
� �<br />
G s<br />
� � � � �<br />
i G si x<br />
�<br />
� � � � � � � � i �<br />
First of all, it is worth writing � i according to the partial derivative of the reaction<br />
function r i :<br />
�
�Hi �Hi �ri<br />
�xi �xi �wi<br />
� i � � �<br />
�Hi 2�Hi<br />
�Hi �Hi �ri<br />
� � 1�<br />
�x �s �g �g �g<br />
i i j i j<br />
The income transfer corresponds to dw � �dw � 0 . At the unique equilibrium<br />
i j<br />
*<br />
G �wi, w j�,<br />
agent i ‟s provision satisfies � , �<br />
relation gives<br />
�r �r<br />
dg � dg � dw .<br />
* i * i<br />
i<br />
�gj j<br />
�wi<br />
i<br />
Since<br />
dg dG dg<br />
* * *<br />
j i<br />
r g w � g and differentiation of this<br />
* *<br />
i j i i<br />
�r �r<br />
dg � dG � dg � dw . That is:<br />
* * *<br />
� � , we have � �<br />
i i<br />
i<br />
�gj i<br />
�wi<br />
i<br />
�ri �ri<br />
�ri<br />
�g * j<br />
dgi �<br />
�ri 1� �g * �wi<br />
dG �<br />
�ri 1� �g �gj<br />
dwi �<br />
�ri<br />
1�<br />
�g<br />
*<br />
dG � � idwi ,<br />
j j j<br />
A similar expression holds for<br />
� �ri �rj<br />
�<br />
� �g �<br />
j �<br />
�gi<br />
1�<br />
� � � � � �<br />
� �ri �rj<br />
�<br />
�<br />
1� 1�<br />
g j g<br />
�<br />
� � � i �<br />
*<br />
dg j and summing both expressions gives:<br />
� �<br />
*<br />
dG � idwi � jdw j � i � j dwi<br />
Because of [12], the first factor of the left hand side is positive thus:<br />
*<br />
dG 0 �i � j<br />
� � � �<br />
157<br />
,
Proof of Corollary 4.1.<br />
Given g j , differentiating � � u � with respect to g i gives best-response<br />
interior solution the first order condition is satisfied:<br />
1<br />
�i � �i<br />
� � � 0.<br />
w � r � r � g � � r � g<br />
� � � �<br />
i i i i j i i j<br />
Therefore, ��i �i �� wi ri � �i �ri g j � �i<br />
�ri g j �<br />
� � � � � � , and<br />
��i ��i�<br />
�� ��<br />
�<br />
1 1 1 1<br />
r �g , w � � w � g � w � A g<br />
2 2 2 2<br />
i j i i j i i j<br />
i i<br />
158<br />
*<br />
g i . At an<br />
This equation holds if the right hand side is between 0 and w i which is the case if<br />
g j � wi Ai<br />
when A � 0 (i.e. �i � �i)<br />
and g j �� wi Ai<br />
when A � 0 (i.e. �i � �i).<br />
When �� � �=0<br />
� , �½� i i<br />
i<br />
r � w for any g � 0.<br />
The same reasoning applies to agent<br />
i i<br />
j. �<br />
Proof of Corollary 4.4.<br />
At an interior equilibrium, the two following equations are satisfied:<br />
* *<br />
�� 2gi<br />
� Ai g j �wi<br />
�<br />
�� A g �2g�w * *<br />
j i j j<br />
� � �<br />
� � �<br />
i i<br />
j j<br />
where A � , A � ���1,1� i j<br />
�i ��i�j��j The aggregation of provisions amounts:<br />
.<br />
j<br />
i
� gi � 1 � 2 �Ai��wi�1 �2wi�A1wj� �<br />
g<br />
� � � �<br />
4 A A �A2 � �<br />
w<br />
� �<br />
4 A A 2w�Aw<br />
�<br />
� �<br />
� j � i j � j � � j � i j � j 2 i �<br />
And the total provision is:<br />
1<br />
�2 � �2 �<br />
� �<br />
� �<br />
*<br />
G � � Aj wi � � Ai wj<br />
4 � AA i j<br />
When � �� � 0 and agents are exclusively intrinsically impulsed<br />
i j<br />
2w2 * i � wi � wj � wj<br />
1<br />
G � � w � w<br />
4 �1<br />
3<br />
� i j�<br />
159<br />
�
160