Out-of-Sample Equity Premium Prediction: Combination Forecasts with Frequency-Decomposed Variables

Abstract

Technical trading rules are widely used by practitioners to forecast the U.S. equity premium. I decompose technical indicators into components with frequency-specific information, showing that the predictive power comes from medium-frequency variation in buy and sell signals, without much evidence of predictability outside of this frequency band. This pattern can be observed for commonly used strategies based on volume, momentum, and moving-average rules. A mean-variance investor who only forecasts with these medium-frequency components generates economically sizable gains compared to the historical average and the basic technical indicators. Combined forecasts from filtered indicators provide utility gains that are often twice as large as gains from the basic indicators. I show that the improvements mainly result from a better market timing around recessions. Overall, the choice of frequency matters more than the choice of technical indicator. I provide evidence that filtered buy and sell signals increase economic gains for several country indices, thereby ruling out data-snooping concerns.