Financial crises appear to have long-lasting effects, even after the
crisis itself has past. This paper offers a simple explanation
through Bayesian learning from rare events. Agents face a latent and time-varying
probability of economic disaster. When a disaster occurs, learning
results in greater effects on asset prices because agents update their
probability of future disasters. Moreover, agents’ belief that the
disaster risk is high can rationally persist for years, even when it
is in fact low. We generalize the model to allow for a noisy signal of
the disaster probability. This generalized model explains excess
stock market volatility together with negative skewness, effects that
previous models in the literature struggle to explain.