Theory and Behavior of Multiple Unit Discriminative Auctions

This paper reports the results of controlled experiments designed to test the HarrisRaviv generalization of the Vickrey theory of bidding in multiple unit discriminative auctions. The paper also discusses further development of the theory-in a way suggested by the experimental results-to include bidders with distinct risk preferences. IN A WELL-KNOWN PAPER, Vickrey [20] formulated a Nash equilibrium model of bidding by risk neutral economic agents in single unit auctions. This analysis was subsequently extended in numerous papers. Vickrey [21] generalized his original model to include multiple unit auctions in which each of N risk neutral bidders can bid on one out of a total of Q homogeneous items up for auction, where 1 c Q < N. In both of the Vickrey papers, individual values for the auctioned object(s) were assumed to be drawn from a uniform distribution. Holt [12] and Riley and Samuelson [19] for single unit auctions and Harris and Raviv [11] for multiple unit auctions have extended the Vickrey model to the case in which valuations are from a general distribution function and all agents have identical concave utility functions.' This paper reports the empirical properties of individual bidding behavior and seller revenue for a group of 28 laboratory experiments designed to test the Harris-Raviv generalization of the Vickrey (hereafter, VHR) model of Nash equilibrium behavior in multiple unit discriminative auctions. In Section I, we summarize briefly the theoretical results of the VHR model which form the basis for the hypotheses that we test. Section II describes the experimental design that we use. Section III describes the experimental results and various tests based on the bidding behavior of individual subjects and on aggregate market (seller revenue) data. The results from six of the ten (N, Q) parameter designs reported