We decompose earnings yield into a smoothing component and a stationary residual component to isolate the fluctuations due to variation in expected returns from those due to the change in the forecast of dividend dynamics. The residual component forms a powerful predictor of dividend growth motivated by linking the dividend partial adjustment model to the present-value framework. Empirically, the proposed predictor displays statistically significant in-sample and out-of-sample predictive power for aggregate dividend growth at monthly and annual frequencies, robust in dividend reinvestment strategies, in pre- and post-war data, in good and bad times, and at short and long horizons.