Equilibrium Stock Return Dynamics Under Alternative Rules of Learning About Hidden States

We examine the dynamic properties of equilibrium stock returns in an incomplete information economy in which the agents need to learn the hidden state of the endowment process. We consider both the case of optimal Bayesian learning and suboptimal learning, including near-rational learning, over- or under-confidence, optimism or pessimism, adaptive learning, and limited memory. We find that Bayesian learning can quantitatively explain short-run momentum, long-run mean-reversion, predictability, volatility clustering, and leverage effects in stock returns. Only over-confidence can marginally improve some aspects of the model (add short-run momentum) without substantially deteriorating other aspects. We conclude that the success of the incomplete information model is quite dependent on optimally learning agents.