Abstract:
This paper explores the out-of-sample forecasting performance of 25 equity premium predictors over a sample period from 1973 to 2023. While conventional time-series methods reveal that only one predictor demonstrates significant out-of-sample predictive power, frequency-domain analysis uncovers additional predictive information hidden in the time series. Nearly half of the predictors exhibit statistically and economically meaningful predictive performance when decomposed into frequency components. The findings suggest that frequency-domain techniques can extract valuable insights that are often missed by traditional methods, enhancing the accuracy of equity premium forecasts.