There's a pretty good article in the October Dobb's about complexity and the use of gaming agents. One caveat, though; the article includes a figure that shows a model outperforming various other strategies in a financial market. The text mentions something that anyone interested in using computers to do financial modeling should know: "transaction costs in the actual market may prevent this profit from being achieved." Absolutely. It's trivial to create a strategy that can outperform financial markets, if you have perfect information, no transaction charges, and perfect order fulfillment.