Forecasting the equity premium with frequency-decomposed technical indicators

Abstract

Practitioners widely use technical trading rules 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 this frequency band. This pattern can be observed for common strategies based on volume, momentum, and moving average rules. In line with previous work, I find that the statistical predictability of technical indicators is centered in recessions and vanishes after sharp market rebounds. However, the out-of-sample R2 increases from less than 4% for the unadjusted indicators to more than 8% for the filtered counterparts in these periods. The medium-frequency components have a better market timing ability, and the economic gains are statistically significant. A mean–variance investor would be willing to pay a fee of 100 to 150 basis points annually to access the filtered indicators. I show that the best-performing frequencies can be identified in real time. The predictive power of medium-frequency components stems from their ability to anticipate discount rate news and changes in investor sentiment. Results are robust to different country indices and transaction costs.