Predicting Stock Returns: Historical Mean vs. Dividend Yield

Abstract

This paper considers whether the log dividend yield provides forecast power for stock returns. While this is an oft-researched topic there is no consensus answer and yet it remains crucial in our understanding of asset pricing. Using a five-year rolling window we compare forecasts from the dividend yield model to those from the historical mean model across forecast magnitude, sign and investment metrics. Results show that in each case the dividend yield model provides superior forecasts. While the difference in, for example, RMSE and the success ratio is small, results support improved market timing and a higher Sharpe ratio using the dividend yield model. In explaining these results, we note that recursive forecasts do not perform as well and thus it is the nature of time-variation within the forecast parameter that is important. We also argue that such time-variation is linked to economic performance. Overall, these results support stock returns forecasting but stresses the importance of time-variation in the forecast model to ensure forecast power.

Keywords: Stock Returns, Dividend Yield, Forecasting, Rolling Regressions

JEL Classification: C22, G12