Although there has been a good deal of research on auction theory, especially from the perspective of auction mechanisms [KlempererKlemperer1999], studies of autonomous bidding agents and their interactions are relatively few and recent. TAC is one example. FM97.6 is another auction test-bed, which is based on fishmarket auctions [Rodriguez-Aguilar, Martin, Noriega, Garcia, SierraRodriguez-Aguilar et al.2001]. Automatic bidding agents have also been created in this domain [Gimenez-Funes, Godo, Rodriguez-Aguiolar, Garcia-CalvesGimenez-Funes et al.1998]. There have been a number of studies of agents bidding for a single good in multiple auctions [Ito, Fukuta, Shintani, SycaraIto et al.2000,Anthony, Hall, Dang, JenningsAnthony et al.,Preist, Bartolini, PhillipsPreist et al.2001]. Outside of, but related to, the auction scenario, automatic shopping and pricing agents for internet commerce have been studied within a simplified model [Greenwald KephartGreenwald Kephart1999].
Twenty-two agents from 6 countries entered TAC, 12 of which qualified to compete in the semi-finals and finals in Boston. The designs of these agents were motivated by a wide variety of research interests including machine learning, artificial life, experimental economics, real-time systems, and choice theory [Greenwald StoneGreenwald Stone2001].
Our own approach was motivated by our research interests in multiagent learning [LittmanLittman1994,StoneStone2000,Singh, Kearns, MansourSingh et al.2000]. Based on the problem description, we expected to find several learning opportunities in the domain. As noted above, detailed opponent modeling was precluded by the system dynamics. Nonetheless, ATTac-2000's adaptivity is one of the keys to its success, particularly in avoiding skyrocketing hotels.
The 2nd and 3rd place agents both used a different strategy to prepare for the possibility of skyrocketing hotels. Rather than avoiding popular hotels entirely by tracking closing prices across game instances, they both discouraged their agents from bidding for too many of any particular hotel room, thus spreading their demand across the rooms [Greenwald StoneGreenwald Stone2001]. While such a strategy is safer in the limit (i.e., it continues to work even if everyone uses it), it has a greater potential to cost the agent in the event that hotel prices do not skyrocket, since the agent will still distribute its demand to the less desirable rooms. On the other hand, ATTac-2000 would notice that the prices are not skyrocketing and thus bid for the optimal travel packages given current prices.