Learning Models And Asset Market Dynamics

Finance is witnessing an important paradigm shift, from a representative,rationalagent approach towards a heterogeneous, boundedly rational agents approach inwhich the market evolution is an important content,while the learning process is thecomposition of the market evolution.Agent-Based computational finance now play auseful role in observing and studying the market evolution and agent learning forits bottom-up modeling approach.This dissertation is composed of following research:Through agent-based computational experiment, I observed that differentlearning model imposed on market dynamics and researched what lead to volatilityclustering and asymmetric volatility, two important―stylized facts‖.By introducing―social learning bonus‖,I designed and implemented a new classifier system integratedof social learning and individual learning and apply it to an artificial stock market.Ifound volatility clustering and asymmetric volatility when the social learning bonuschanged.The strength of imitation lead to volatility clustering and asymmetricvolatility.After building an artificial stock market composed of lower rationality,limitedstrategy,and routine-based learning agents,I found that this market rapidly reached thestatus of random walk and proved it‘s an efficient market.While introducing highrationality agents who continually updated its strategy and created new strategy,themarket left random walk and volatility clustering emerged.It suggested that theproportion of heterogeneous,different rational agents impose on market dynamicsand the more sophisticated agents don‘t help to reach homogeneous rationalexpectation equilibrium.