control and full abstraction

We present a novel method for automatically acquiring strategies for the double auction by combining evolutionary optimization together with a principled game-theoretic analysis. Previous studies in this domain have used standard co-evolutionary algorithms, often with the goal of searching for the “best ” trading strategy. However, we argue that such algorithms are often ineffective for this type of game because they fail to embody an appropriate game-theoretic solution-concept, and it is unclear, what, if anything, they are optimizing. In this paper, we adopt a more appropriate criterion for success from evolutionary game-theory based on the likely adoption-rate of a given strategy in a large population of traders, and accordingly we are able to demonstrate that our evolved strategy performs well. Keywords auctions and electronic markets, multi-agent evolution, adaptation and learning, game theoretic foundations of agent systems 1