Building Financial Markets With Artificial Agents: Desired goals, and present techniques

Financial markets operate as a large interacting group of agents each in a constant struggle to better understand and interpret current prices and information. The complex interconnections between prices and information is probably more dramatic in financial markets than any other economic situation. Economic theory has been capable of describing many different financial equilibria, but it remains quiet on the types of dynamics that can occur while learning is still active and equilibrium is never quite obtained. Models of learning agents allow a direct attack on this problem. However, even though this approach may seem appealing at first it does come with many costs. The first of these is the modeling of the agents themselves. Boundedly rational agents can come in many forms, and an important question for theorizing is where to“set the dial” of rationality. This paper will describe some of the methods that have worked, and which directions look promising. Some methods for endogenously setting the level of rationality will be discussed. Finally, comparisons to the single agent, homogeneous belief world will be made, stressing why this is still a useful benchmark. A second issue involves the actual trading mechanisms, and this will be briefly discussed in relation to how outcomes can be affected. In closing, some of the policy questions centered on market stability and structure will be compared with certain computational issues.