Randomized strategic demand reduction: getting more by asking for less

Auctions are an area of great academic and commercial interest; however, many important auctions are extremely complicated, involving simultaneous auction of hundreds of non-interchangeable goods with value dependencies. This makes it difficult or impossible for humans to grasp all of the nuances of their strategic decisions. Computer aid could ease the burden of efficiently competing in these auctions; indeed, very simple autonomous agents have started to appear as bidders in some auctions (e.g. on eBay). Although agents that function in more complex experimental scenarios have been deployed [5], studies of autonomous bidding agents and their interactions are relatively few and recent. Our work explores agent bidding strategies in a large-scale and realistic auction scenario, namely the FCC Spectrum Auction Simulator, orFAucS [2]. That paper introducedknapsackagents that optimized the set of goods they bid on given a budget constraint, but without taking into account the needs and strategies of other agents. We extend this work by examining the possibilities of agent cooperation and uncertain knowledge. In this paper, we present punishing randomized strategic demand reduction (PRSDR), a strategy by which cooperative agents can significantly outperform the Knapsack agents despite having highly uncertain knowledge regarding each others’ goals and without any explicit inter-agent communication. The strategy is self-enforcing in that agents cannot benefit by defecting back to the Knapsack strategy, and serves as an example of a general strategy for efficient bidding in many realistic simultaneous multiple-round (SMR) auctions.