This paper introduces a new out-of-sample forecasting methodology for monthly market returns using the variance risk premium (VRP) that is both statistically and economically significant. This methodology is motivated by the ‘beta representation,’ which implies that the market risk premium is related to the price of variance risk by the variance risk exposure. Hence, when the slope of the contemporaneous regression of market returns on variance innovation is larger, future returns are more sharply related to the current VRP. Also, predictions are more accurate when market returns are highly correlated to variance shocks.

We consider a model in which the correlation between shocks to consumption and to expected future consumption growth is nonzero and varies over time. We validate this assumption empirically using the model’s implication that time variation in consumption growth persistence (CGP) drives the correlation between stock and bond returns. Our model implies that the stock–bond correlation is also related to the predictive relation between bond yields and future stock returns. Finally, we provide suggestive evidence that asset price fluctuations are the primary driver of changes in CGP.