The microstructure of markets involves not only human traders’ learning and erring processes but also their heterogeneity. Much of this part has not been taken into account in the agent-based artificial markets, despite the fact that various computational intelligence tools have been applied to artificial-agent modeling. One possible reason for this little progress is due to the lack of good-quality data by which the learning and erring patterns of human traders can be easily archived and analyzed. In this chapter, we take a pioneering step in this direction by, first, conducting double auction market experiments and obtaining a dataset involving about 165 human traders. The controlled laboratory setting then enables us to anchor the observing trading behavior of human traders to a benchmark (a global optimum) and to develop a learning index by which the learning and erring patterns can be better studied, in particular, in light of traders’ personal attributes, such as their cognitive capacity and personality. The behavior of artificial traders driven by genetic programming (GP) is also studied in parallel to human traders; however, how to represent the observed heterogeneity using GP remains a challenging issue.
Financial Decision Making Using Computational Intelligence, Springer Series Optimization and Its Applications, 70, 2012, 35-69