Predictive Regressions under Arbitrary Persistence and Stock Return Predictability

Abstract

We present a novel approach to analyzing stock return predictability that accommodates (i) arbitrary predictor persistence, (ii) panels with common factors, (iii) multiple predictors, (iv) short- and long-horizon analysis, and relies on standard inference from least-squares estimation of a suitably adjusted predictive regression. We analyze US and international equity premia and find that dividend- and earnings-related price ratios have negligible predictive power over long horizons, whereas the dividend yield has considerable predictive power over short horizons, with positive coefficients, consistent with present value theory. Long-term government bond yields exhibit predictive power over all horizons from one month through five years.