An Agent-Based Model With Realistic Financial Time Series: A Method for Agent-Based Models Validation

Abstract

This paper proposes a methodology to empirically validate an agent-based model (ABM) that generates artificial financial time series data comparable with real-world financial data. The approach is based on comparing the results of the ABM against the stylised facts -- the statistical properties of the empirical time-series of financial data. The stylised facts appear to be universal and are observed across different markets, financial instruments and time periods, hence they can serve to validate models of financial markets. If a given model does not consistently replicate these stylised facts, then we can reject it as being empirically inadequate. We discuss each stylised fact, the empirical evidence for it, and introduce appropriate metrics for testing the presence of these in model generated data. Moreover we investigate the ability of our model to correctly reproduce these stylised facts. We validate our model against a comprehensive list of empirical phenomena that qualify as a stylised fact, of both low and high frequency financial data that can be addressed by means of a relatively simple ABM of financial markets. This procedure is able to show whether the model, as an abstraction of reality, has a meaningful empirical counterpart and the significance of this analysis for the purposes of ABM validation and their empirical reliability.