Agent-based simulation on real stock market

This paper tries to present a more suitable computer simulated stock market model according to the statistics of real stock data. As the real stock market change violently due to the financial crisis, we divide the real market into the one before financial crisis and the one after it. Time series are analyzed from the standpoint of well-known empirical features in real markets as benchmark. Firstly, some modifications are done to make agents more "real" according to the statistics of real stock data before financial crisis, the modifications including agents with different learning speeds, strategy-sizes, utility functions. Secondly, when the best modified model before financial crisis failed to replicate the statistics feature of the real market after financial crisis, we make further research to find out the best simulate model after financial crisis. Finally, the best simulated model after financial crisis is able to replicate several of the real market phenomenon, including first-order autocorrelation, Kurtosis, standard deviation of yield series and the first-order autocorrelation of yield square.