Instantiating the Contingent Bids Model of Truthful Interdependent Value Auctions

Abstract

We consider the problem of auction design with agents that have interdependent values, i.e. values that depend on each others' private signals. We adopt the contingent bids model of Dasgupta and Maskin [3], and allow agents to submit bids of the form "if player 1 bids $x for good A then I will bid $y." Our main contribution is to identify a specific linear valuation model for which there exists an efficient auction for a single item, and then extend this to provide an approximately efficient combinatorial auction with single-minded bidders. In both auction, winners and payments are computed from the fixed point of the valuation mapping defined by contingent bids. We also adopt search in order to construct a variation on the single-item auction with improved revenue. In closing, we discuss the (many) challenges in moving to more general models of interdependent valuations.