Computationally Limited Agents in Auctions

Auctions provide eAEcient and distributed ways of allocating goods and tasks among agents. In this paper we study optimal strategies for computationally limited agents, where agents must use their limited resources in order to compute valuations for (bundles of) the items being auctioned. Agents are free to compute on any valuation problems including their opponents'. The deliberation actions are incorporated into the agents' strategies and di erent auction settings (both single{item and combinatorial) are analyzed in order to determine equilibrium strategies. We show that is some auction mechanisms, but not others, in equilibrium the bidders compute on others' problems as well. It is shown that under our model of bounded rationality, the generalized Vickrey auction (GVA) looses its dominant strategy property. The model of bounded rationality impacts the agents' equilibrium strategies and so must be considered when designing mechanisms for computationally limited agents.