Data mining the future: genetic discovery of good trading rules in agent-based financial market simulations

The field of agent-based simulation of financial markets has grown considerably in the last decade. However, the interpretation of simulation results has received far less attention. Typically, the output of a large number of simulations is reduced to one or two summary statistics, such as sample moments. While such summarization is useful, it overlooks a vast amount of additional information that might be revealed by examining patterns of behavior that emerge at lower levels. The authors propose an approach to interpreting simulation results that involves the use of so called data mining techniques to identify the rules of behavior that govern computer simulations of a complex system. They demonstrate the approach by analyzing data from runs of two different agent-based financial market simulations. In each case they derive rules that describe characteristics of the underlying structure of the simulated markets. In one case they are able to recover evidence of known dynamics of the simulation, and in the second case they find evidence of emergent behavior not explicitly a part of the simulation design