Empirical Game-Theoretic Analysis of the TAC Market Games

ABSTRACT We illustrate developing techniques for empirical game-theoretic analysis by application to two challenging market games employed in an annual Trading Agent Competition. These games exemplify relevant environments beyond analytic tractability, yet which can be investigated experimentally through simulation and careful measurement. Our analysis of agents from the 2005 TAC Supply Chain Management game reveals interesting interactions not seen in the tournament, and demonstrates the value of a post-competition agent repository. Our ongoing analysis of the TAC Travel game illustrates methods for scaling up the strategy and profile space, and demonstrates the value of empirical game-theoretic analysis for strategy selection. Interesting open issues remain, particularly in regard to controlling the experimental search process.