Crises and Recoveries in an Empirical Model of Consumption Disasters

We estimate an empirical model of consumption disasters using a new panel data set on consumption and output for over 20 countries and more than 100 years. The model allows for permanent and transitory effects of disasters, nesting both unit root and trend stationary consumption growth. We estimate the model using Bayesian methods. Our empirical estimates imply that the probability of entering a disaster state is 3.5%, and the probability of continuing in the disaster state is 47%. During the average disaster, consumption falls by 19%, but a large fraction of disasters involve consumption drops of over 30%. Mean reversion after a disaster is rapid; however, the long-run effect of disasters is estimated to be about 2/3 of their original size. We investigate the asset pricing implications of these rare disasters.