Forecasting the equity risk premium with frequency-decomposed predictors

Abstract

We show that the out-of-sample forecast of the equity risk premium can b e signi cantly improved by taking into account the frequency-domain relationship b etween the equity risk premium and several p otential predictors. We consider fteen predictors from the existing literature, for the out-of-sample forecasting p erio d from January 1990 to Decemb er 2014. The b est result achieved for individual predictors is a monthly out-of-sample R 2 of 2.98 % and utility gains of 549 basis p oints p er year for a mean-variance investor. This p erformance is improved even further when the individual forecasts from the frequency- decomp osed predictors are combined. These results are robust for di erent subsamples, including the Great Mo deration p erio d, the Great Financial Crisis p erio d and, more generically, p erio ds of bad, normal and go o d economic growth. The strong and robust p erformance of this metho d comes from its ability to disentangle the information aggregated in the original time series of each variable, which allows to isolate the frequencies of the predictors with the highest predictive p ower from the noisy parts.