Should stock returns predictability be 'hooked on' long-horizon regressions?

This paper re-examines stock returns predictability over the business cycle using price-dividend and price-earnings valuation ratios as predictors. Unlike prior studies that habitually implement long-horizon/predictive regressions, we conduct a testing framework in the frequency domain. Predictive regressions support no predictability; in contrast, our results in the frequency domain verify significant predictability at medium and long horizons. To robustify predictability patterns, the analysis is executed repetitively for fixed-length rolling samples of various sizes. Overall, the stock returns are predictable for wavelengths higher than 5 years. This finding is robust and independent of time, window size and predictor.