Empirical studies have found that during bad times return predictability is higher. Thus, variation in discount rate news should be relatively higher as economic conditions worsen. We propose a parsimonious model for expected returns that captures the countercyclical dynamics of stock return predictability. The implications of the model are studied within a flexible Bayesian predictive system. In our empirical tests, the estimates from the proposed model of expected return present lower prediction errors than the historical mean, and than the predictive regression using the dividend yield. Furthermore, the persistence parameter of expected returns presents long-term dynamics similar to the autocorrelation of realized returns.