Click here for full text:
Pooling Information from Auctions with Varying Number of Bidders
Marquez, Carolina; Guler, Kemal; Tang, Hsiu-Khuern; Zhang, Bin
Keyword(s): auctions; bidding strategies; nonparametric estimation; econometrics of auctions
Abstract: First price auctions are among the most commonly used auction mechanisms, especially in procurement . Recent research activity in economics of auctions focuses on development of econometric methods to analyze the bid data from such auctions to estimate the underlying structural variables. Optimization of almost all auction-related decisions (reserve price, auction format, information rules) requires reliable estimates of these structural elements. Current nonparametric methods to estimate the distribution of bidders' private valuations in first price auctions make use of functionals of bid distributions. Currently known estimation methods require separate estimation of these functionals for each configuration of observable variates. In many situations the sizes of some subsamples are too small to obtain reliable estimates of the functionals of interest. In such cases, the practice is to discard such subsamples and work with the remaining observations to estimate the latent valuation distributions. This waste of information may have substantial impact on the reliability of the final estimates and the decisions based on such estimates. Especially in situations where the total size of available samples are small to start with, the need for methods and algorithms that make use of all available data is obvious. We propose a new fully nonparametric approach that makes full use of all available auction data. This paper describes the methods based on the new approach and presents Monte Carlo experimental evaluation of alternative methods of combining information from auctions with varying numbers of bidders.
Back to Index