Designing the Market Game for a Trading Agent Competition

Trading in electronic markets is a topic of increasing interest within the artificial intelligence (AI) and electronic commerce research communities. As Internet marketplaces proliferate, programs that monitor and bid in these markets automatically—what we call “trading agents”—will play a significant role. Various AI research communities have held competitions to compare different approaches to common problems, and showcase the state of the art.1-3 Inspired by their successes, we organized a trading agent competition (TAC), held at the most recent International Conference on Multiagent Systems. The competition pitted agents from 20 teams around the world against each other in a market game. The agents ran from participants’ home institution using TCP/IP socket connections to exchange messages with the auction server at the University of Michigan. In this article, we discuss general criteria for the design of a market game using trading agents and present our competition as a model problem description for this domain; we also describe the competition’s genesis, its technical infrastructure, and its organization. The article, “Autonomous Bidding Agents in the Trading Agent Competition”4 (page 52), describes the competition from a participant’s perspective and describes the strategies of some of the top-placing agents. A visualization of the competition and a description of the preliminary and final rounds of the TAC are available in IC Online (http://computer.org/internet/ tac.htm). Further information is available on the TAC Web site (http://tac.eecs.umich. edu/) and in other articles on TAC agents.5