Agent-based simulation of dynamic online auctions

Abstract

The need to understand dynamic behavior in auctions is increasing with the popularization of online auctions. Applications include designing auction mechanisms, bidding strategies, and server systems. We describe simulations of a typical online auction, where the duration is fixed, and the second-highest price is continuously posted and determines the winner's payment. We modeled agents of exactly two types, idealizations and simplifications of those observed in practice: early bidders, who can bid any time during the auction period, and snipers, who wait till the last moments to bid. This allows us to study the interactions of the two types of bidders during the course of auctions, and the effects of the two strategies on the probability of winning, the final price, and the formation of price consensus in iterated auctions. Results show that 1) early bidders can win with a lower price on average than snipers, but much less often; 2) the late bidding strategy of snipers is effective; and 3) in iterated auctions, adjustment feedback of motivational parameters can lead to effective price consensus with small fluctuations.