Learning, Structural Breaks, and Asset-Return Dynamics

This paper studies a representative-agent asset-pricing model of an endowment economy in which the agent has incomplete knowledge about exogenous stochastic endowment process and has incentive to learn about the process with adaptive learning rules. There is the well documented fact that when underlying economic environment is known and is common knowledge to investors, asset-pricing models under rational expectations with complete knowledge about model structure cannot account for such basic features of the data in the U.S. asset market as relative volatility between equity premium and risk-free rate, highly persistent excess return, and long-horizon predictability of asset returns. Although much effort to understand those aspects of asset-return dynamics have been done by relaxing underlying assumptions in Mehra and Prescott (1985) economy, for example more sophisticated forms of preference (Epstein and Zin, 1990, 1991; Weil, 1989; Kandel and Stambaugh, 1991; Abel, 1990; Constantinides, 1990; Heaton, 1995; Campbell and Cochrane, 1999), trading costs (Aiyagari and Gertler, 1991), and incomplete asset market (Weil, 1992), empirical performance has not been very successful.1 Recently, one growing line of research takes an alternative approach that allows for departures from the assumption of common knowledge and fully rational agents. For example, Cecchetti, Lam, and Mark (2000) examine a representative agent who has distorted beliefs about transition probabilities in a two-state