Equity Return Prediction: Are Coefficients Time-Varying?

Abstract

Most papers in equity return prediction rely on the assumption that coefficients in linear predictive models are constant over time. We question and relax this assumption and find strong empirical support for the validity of models allowing for time varying regression coefficients. Analyzing model uncertainty, we document that uncertainty about the level of time-variation in coefficients and uncertainty about the choice of predictive variables are equally important sources of predictive variance. Furthermore, we document out-of-sample predictability of the average and individual predictive models. Most importantly, only individual predictive models with time-varying coefficients show out-of-sample predictability across all periods of time.