Biased Bayesian Learning and the Risk-Free Rate Puzzle

According to the risk-free rate puzzle the return on safe assets is much lower than predicted by standard representative agent models of consumption based asset pricing. Based on non-additive probability measures arising in Choquet decision theory we develop a closed-form model of Bayesian learning in which the Choquet estimator of the mean consumption growth rate does not converge to its 'true' value. It rather expresses a bias that reflects the agent's ambiguity about his estimator. We calibrate the standard equilibrium conditions of the consumption based asset pricing model to demonstrate that our approach contributes to a resolution of the risk-free rate puzzle when the agent's learning process exhibits a moderate degree of ambiguity that is resolved in a pessimistic way