AI

Boosted Second-price Auctions for Heterogeneous Bidders

Abstract

Due to its simplicity and desirable incentive aspects, the second-price auction is the most prevalent auction format used by advertising exchanges. However, even with the optimized choice of the reserve prices, this auction is not optimal when the bidders are heterogeneous, i.e., when the bidder valuation distributions differ significantly. We propose a new auction format called the boosted second-price auction, which favors bidders with lower inverse hazard rates (IHRs), roughly speaking, bidders with more stable bidding behavior. Based on our analysis of auction data from Google’s advertising exchange, we found bidders to be heterogeneous and can be ordered based on their IHRs. In this paper, we theoretically analyze and describe how our proposed boosted second-price auctions increase revenue over that of the widely used second-price auctions by favoring bidders with lower IHRs. We also provide practical guidelines for determining boost values and validate these guidelines both theoretically and empirically. Our counterfactuals, based on actual transaction data, show that boosted second-price auctions that follow our guidelines perform almost optimally and obtain up to 3% more revenue than second-price auctions.