J. Doyne Farmer of the Sante Fe Institute and his colleagues designed a computer model in which traders placed orders to buy and sell at random. In the simulation, the so-called "minimally intelligent agents" were subject only to the rules of the market. When the researchers tested their approach using 21 months of data from the London Stock Exchange, they found that it successfully predicted two basic market properties with surprising acuity. The model accounted for 96 percent of the variance in the spread, which is the difference between the best buying and selling prices and is the main determinant of transaction costs. It also explained three quarters of the fluctuations in the diffusion rate, which is a standard measure of financial risk.
The authors note that their results do not imply that stock traders are unintelligent. Instead they say that the report, published online this week by the Proceedings of the National Academy of Sciences, "suggests there are circumstances where the strategic behavior of agents may be dominated by other considerations." The new model could be employed to help regulators lower transaction costs and detect irregular market behavior.