Standard predictability regressions use return as dependent variable, which is the sum of change in dividend-price (dp) ratio and dividend growth, even though dividend growth is known to be noisy. We construct a new predictor based solely on the dp dynamics and show it outperforms the historical aver-age. Our predictor does not even require estimation for long-horizon return, yet it outperforms the best known prediction approach based on dp. We re-late our approach to structural estimation, which reveals that dp dynamics is informative for return prediction in real time. Our evidence cautions against using reduced-form return regressions to evaluate channels of predictability.
展开▼