This paper addresses the issue of developing agents capable of participating in several potentially simultaneous auctions of different kinds (English, First-Price, Vickrey), with the goal of finding the best price for an item on behalf of their users. Specifically, a multi-agent architecture is proposed, in which a manager agent cooperates with several expert agents, each specialised in a specific kind of auction. The expert agents communicate their knowledge to the manager agent in the form of probability functions, capturing the likelihood that a bid of a given price may win an auction. Given a set of such functions, the manager agent builds a bidding plan that it executes in concert with the expert agents.
Marlon Dumas, Guido Governatori, Arthur H. M. ter