This paper studies whether the Rietz-Barro "disaster" model, extended for a time-varying probability of disaster, can match the empirical evidence on predictability of stock returns. It is shown that when utility is CRRA, the model cannot replicate this evidence, regardless of parameter values. This motivates extending the disaster model to allow for Epstein-Zin utility. Analytical results show that when the probability of disaster is i.i.d., the model with Epstein-Zin utility can match the evidence on predictability qualitatively if the intertemporal elasticity of substitution is greater than unity. The case of a persistent probability of disaster is studied numerically, with partial success.
