Semi-Structural Modeling of Empirical Auction Models

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

Each year billions of dollars of goods and services are acquired through auctions. By analyzing data that are generated in these transactions, key insights can be obtained on critical issues such as auction design. The data also may be used to understand the competitive character of these markets and bring information to bear on possible collusive behavior and market performance as reflected in economic efficiency. "Structural" analysis of auctions traditionally is based on game theoretic models that assume perfect rationality. Yet, the findings from auction experiments are pervasive and consistent: The behavior of auction participant deviates significantly from theory prescriptions. The extent of deviation sharply contradicts the assumption of perfect rationality that is so fundamental to structural analysis. The objective of this proposal is to bridge the gap between experimental economics and empirical auction modeling. We will explicitly allow bidders to deviate from the theoretical solutions of an inherently oversimplified "benchmark" theory model. Such deviations could be non-strategic (mistakes), or they may be strategic in the sense that they reflect real-life factors that are not appropriately captured by the benchmark model. Traditional bid functions, which assign a single bid to a player's valuation, are replaced by a full probability distribution for bids, conditional on valuations. The proposal develops operational numerical procedures for the analysis of bids under this new class of "semi-structural" models and explains their statistical properties. The analytical tools developed in this framework will be applied to several real-life situations: (1) Auction data from timber sales in the Northwest Pacific will be analyzed, with a specific focus on the identification of possible collusive behavior. (2) Data related to school milk procurement in Georgia will be analyzed, where issues of market efficiency may be highlighted. (3) Internet auctions, where observed behavior sharply deviates from standard theory prescriptions, will be used to demonstrate the explanatory power of the techniques developed. These empirical studies are expected to provide critical new insights in the operations of real-life auctions and result in suggestions for corresponding adjustments in their design.