On the problem of calibrating an agent based model for financial markets

Agent based models are very widely used in different disciplines. In financial markets, they can be used to explain well known features called stylised facts and fit statistical properties of data. For this reason, they can model price movements better than standard models using gaussianity. Calibration and validation are essential issues in agent-based modeling. However, calibrating such models is not yet sufficiently considered in the literature. In this paper, a Nelder–Mead simplex algorithm coupled with threshold accepting algorithm (Gilli and Winker in Comput Stat Data Anal 42:299–312, 2003) and a genetic algorithm have been implemented to calibrate the model presented by Farmer and Joshi (J Econ Behav Org 49:149–171, 2002) and the outcomes have been compared and discussed. The data used are closing prices of S&P500 Composite index and a particular attention has been devoted to the choice of the objective function.