Let’s take a smooth break: Stock return predictability revisited

Abstract

The hypothesis that dividend-price ratio predicts stock returns has a long tradition, but the empirical evidence has been mixed. Lettau and Van Nieuwerburgh (2008) find that by taking account of discrete structural changes in the steady state can improve the in-sample fit of a univariate predictive regression, whereas we argue such changes could be smooth rather than abrupt. Using a modification of flexible Fourier form to approximate the smooth shifts in the steady state of dividend-price ratio, our smooth break approach outperforms Lettau and Van Nieuwerburgh’s sharp break approach both in sample and out of sample. Monte Carlo experiments also support our empirical findings and suggest that our approach tends to be more appropriate if the underlying true structural change is indeed smooth. Overall, our novel approach provides a simple, yet effective, way to improve return predictability with economic motivation.