Validation of Agent-Based Models of Financial Markets

Abstract Agent-based models take into account the limited rational behaviour of individuals acting on financial markets. Explicit simulation of this behaviour and the resulting interaction of individuals provide a description of aggregate financial market time series. Although the outcomes of such simulations often exhibit similarities with real financial market time series, methods for explicit validation are required. This paper proposes validation using simulation based on indirect estimation. It uses typical characteristic moments of financial market data to assess the similarity of simulation outcomes. Furthermore, the parameters of the agent-based models can be estimated by maximizing this similarity. The paper presents details of this estimation approach and initial results for the US$/DM exchange rate.