In Joshi's model, agents must choose between two assets: a risky stock and a risk-free bond. Agents can follow one of two strategies: they can be technical traders and look for price patterns that they think will recur, or they can be fundamental traders and buy or sell stocks depending on their underlying financial values.
To determine how technical trading affected the overall market, Joshi looked at how a stock's price history was affected by the agents' use of "decision rules." These rules were expressed in the form of "if . . . then" statements, along the line of "if a stock's price has increased in each of the last 10 periods, then it will increase in the next period."
There were 25 agents in her model, and each was randomly assigned a set of 100 decision rules, divided among the technical and fundamental trading approaches. The model was designed so that agents would use the most accurate rules, in terms of recent history, to buy, sell, or hold a stock. After a trade would clear the model stock market, the agents would determine their profits and losses, and the model would update the accuracy of the rules they used. It would periodically replace the weakest rules in each agent's pool with new rules formed by slightly modifying the previously best-performing rules through the use of "genetic algorithms" (a computer tool used to model evolution in systems of artificial life). Thus, the model allowed agents' trading behavior to spontaneously evolve and adapt over time, reflecting not only their own trading results but also those of other agents.
In order to achieve this evolution of trading behavior, Joshi had to simulate stock market trading thousands and thousands of times, with each simulation involving two dozen agents making many hundreds of thousands of individual investment decisions. To do so, she harnessed the energy of Reed's five most powerful computers and ran the machines every night, night after night, from November through May.