Séminaire GREDEG - Paul Pezanis-Christou (University of Adelaide)

"Structural Analysis of First-Price Auction Data: Insights from the Laboratory"
Quand ? Le 16-06-2016,
de 13:00 à 15:30
Où ? Salle Picasso
S'adresser à
Ajouter un événement au calendrier vCal

Paul Pezanis-Christou

(The University of Adelaide, Australia)


Jeudi 16 juin 2016 


Abstract: We analyze first-price sealed bid auction data with structural econometric methods pertaining to field or laboratory settings to assess the robustness of structural inferences (i.e., estimates, expected revenues and optimal reserve price recommendations) to the information available. Our approach consists in manipulating the quantity and the quality of information available to highlight possible misbehaviors and check how they affect revenue predictions and policy recommendations. Assuming first a field setting in which the researcher has limited information, we find weak support for the Symmetric Bayes-Nash Equilibrium (SBNE) requirements: (i) the quantity of information available usually matters, (ii) bidders' estimated risk parameters are not invariant to the number of bidders, (iii) there is a behavioral spillover from sequential (multi-unit) to single-unit auctions, and (iv) parametric and nonparametric estimation procedures yield different risk parameter estimates. However, bidding behavior is found to be homogenous and assimilable to a SBNE one of bidders with either Constant Relative Risk Averse (CRRA) preferences (possibly risk neutral) or a power form of probability misperception. These interpretations explain bid data equally well and yield different revenue predictions, albeit not significantly so, as well as markedly different optimal reserve price recommendations. Improving the quality of information in such a setting dampens the effect of information quantity on inferences, confirms (ii)-(iv), and yields significantly larger estimates which reject homogeneity in auctions with 3, 4 and 6 bidders. Yet, despite this evidence of out-of-equilibrium behavior, the CRRA hypothesis or a power Probability Weighting Function (PWF), yields significantly better revenue predictions than the one of risk neutrality, especially in auctions with 3 bidders. Finally, when assuming a laboratory setting where the researcher knows the bidders' private values and their distribution, the SBNE model is typically rejected because of a nonlinear behavior, as reported in previous experimental studies and rationalized in terms of heterogeneous CRRA preferences. We show that such behavior may also obtain under risk neutrality if bidders are assumed to misread the uniform distribution as a Beta one. This interpretation captures the observed nonlinearity but does not outperform the homogenous CRRA hypothesis or probability misperception in terms of goodness-of-fit, neither does it lead to better revenue predictions or optimal reserve prices.

(with Andres Romeu)

En savoir plus sur l'invité : Site personnel