We examine the predictive ability of the aggregate earnings yield for market returns and earnings growth by estimating a variance decomposition. Based on weighted long-horizon regressions for the 1946-2015 period, we find that most of the variation in the earnings yield is due to return predictability, with earnings growth predictability assuming a secondary role. However, by estimating a variance decomposition based on a first-order vector autoregression (VAR), we find that the major driving force is earnings growth predictability. We show that the share of earnings predictability implied from the VAR is due to a misspecification driven by the unusual high volatility in aggregate U.S. earnings during the recent financial crisis (2007-2009). Our results suggest that the practice of analyzing long-run return and cash-flow predictability from a VAR can be misleading, and reinforce previous evidence that what drives variation in aggregate price ratios is return (instead of cash-flow) predictability.