An Evolutionary Approach to Finding Bidding Strategies in a Combinatorial Auction

Abstract

Auctions involve trading of variety of different items. Auctions that allow agents to bid for combinations of items are called Combinatorial Auctions (CAs). The Ascending k-Bundle Auction(AkBA) is a combinatorial auction founded on a notion of bundle price equilibrium. The purpose of this research is to explore the strategy space and help agents evolve strategies for a Proxy version of A1BA(P-AkBA). We use a Genetic algorithm to search the space of strategies. Several experiments were performed for different categories of problems and the resultsshow that the approach yields good solutions. We compare the outcomes of the evolved solutions with the outcomes that result from truthful bidding, and compare prices against those generated in the sealed-bid version of k-bundle auction and the standard GVA payments. We also make several observations about the effect of genetic parameters on the performance of search.