Larry Epstein, Martin Schneider
This paper considers learning when the distinction between risk and ambiguity matters.
Working within the framework of recursive multiple-priors, the paper formulates a counterpart of the Bayesian model of learning about an uncertain parameter from conditionally i.i.d. signals.
The framework permits a distinction between noisy and indistinguishable signals and also between indistinguishable and identical experiments.
Other noteworthy features include: The set of conditional probabilities agents use for forecasting expands or shrinks in response to new data.
Ambiguous signals may increase the volatility of conditional actions and may prevent ambiguity from vanishing in the limit.
Properties of the model are illustrated with two applications. A dynamic portfolio choice model suggests that agents should exit (enter) the stock market after a string of bad (good) returns.
A representative agent asset pricing model shows how large unanticipated shocks are amplified by the prospect of ambiguous news.