Tyler Odean, Victor Naroditskiy, Amy Greenwald, John Donaldson
We present a fast and effective bidding strategy for the Trading Agent Competition in Supply Chain Management (TAC SCM). In TAC SCM, manufacturers compete to procure computer parts from suppliers, and then sell assembled computers to customers in reverse auctions. To address the bidding problem, an agent decides how many computers to sell and at what prices to sell them. We propose a greedy solution, Marginal Bidding, inspired by the Equimarginal Principle, which states that revenue is maximized among possible uses of a resource when the return on the last unit of the resource is the same across all areas of use. We show experimentally that Marginal Bidding performs as well as a computationally intensive integer linear programming approach on small problem instances. Moreover, unlike our ILP solution, Marginal Bidding can cope with large problem instances. Hence, it can incorporate Lookahead, that is, it can effectively reason about predicted future as well as current demand.
Subjects: 7.2 Software Agents; 7.1 Multi-Agent Systems
Submitted: May 16, 2007