Learning from the past : the role of personal experiences in artificial stock markets

Recent survey evidence suggests that investors form beliefs about future stock returns by predominantly extrapolating their own experience: They overweight returns they have personally experienced while underweighting returns from earlier years and consequently expect high (low) stock market returns when they observe bullish (bearish) markets in their lifespan. Such events are difficult to reconcile with the existing models. This paper introduces a simple agent-based model for simulating artificial stock markets in which mean-variance optimizing investors have heterogeneous beliefs about future capital gains to form their expectations. Using this framework, I successfully reproduce various stylized facts from the empirical finance literature, such as underdiversification, the predictive power of the price-dividend ratio, and the autocorrelation of price changes. The experimental findings show that the most realistic market scenarios are produced when agents have a bias for recent returns. The study also established a link between underdiversification of investor portfolios and personal experiences.