The Efficiency of an Artificial Stock Market with Heterogeneous Intelligent Agents

In this paper, we construct an artificial equity market using an Artificial Neural Network (ANN). Based on the heterogeneous beliefs and trading strategies prevailing in actual equity markets, we divide traders into three groups: value traders, momentum traders and noise traders. Characteristically, value traders are endowed with an ANN learning mechanism that allows them to forecast the dividend growth rate. A double-auction market setting is adopted as our market mechanism for simulating the market structure. Our artificial market is able to replicate several important features of real markets, including excess volatility, volatility persistence, and serial autocorrelation of stock returns. The profitability of three trading strategies are compared. On average, value traders exhibit the highest Sharpe ratio and significantly positive excess returns. The rational expectations equilibrium emerges for only one of the three experiments. I am grateful to my thesis supervisors Prof. Bryan Campbell and Prof. Lawrence Kryzanowski for their insightful advice and valuable supports. I would also like to thank Prof. Anastas Anastapoulos and Prof. Michael Sampson for their useful comments.