Predicting Stock Returns — The Information Content Of Predictors Across Horizons

We evaluate and compare the information contents of dividend-price ratio and consumption-wealth ratio (cay) for predicting stock returns at different horizons. To do this, we conduct a canonical correlation analysis of wavelet-decomposed stock returns and a selected group of predictors. We show that predictive information is often wasted due to a weak signal problem: The highly predictive component is met with very low variation. Nevertheless, we find that cay contains valuable information about the long run and that, after allowing for structural breaks, dividend-price ratio becomes very informative about short-to-medium-horizon returns and outperforms cay in terms of in-sample R2.