Extended Sample Tests of Established Equity Premium Predictors

Do equity premium predictors published in the past still work after extending their respective discovery samples through 2020? In their September 2021 paper entitled “A Comprehensive Look at the Empirical Performance of Equity Premium Prediction II”, Amit Goyal, Ivo Welch and Athanasse Zafirov reexamine the power of 29 variables found to predict the equity premium (stock market return minus U.S. Treasury bill yield) in 26 prominent published papers, with data samples ending between 2000 and 2017, by extending these samples through 2020. They test not only their predictive powers, but also their performances when applied to four simple market timing strategies with Treasury bills as the alternative asset:

1. Untilted, Unscaled – go long the equity premium when the predictor is above its historical median and short otherwise.
2. Tilted, Unscaled – go long the equity premium unless the predictor is below its historical 25th percentile.
3. Untilted, Z-scaled – first calculate a Z-score by subtracting the historical median from the current value and dividing by the historical standard deviation, and then scale the equity premium allocation by the Z-score.
4. Tilted, Z-scaled – first calculate a Z-score by subtracting the historical 25th percentile value from the current value and dividing by the historical standard deviation, and then scale the equity premium allocation by the Z-score.

The benchmark for these strategies is buying and holding the equity premium. Extending the original discovery sample for each of the 29 predictors through December 2020 (typically about 10 years additional data), they find that: (more…)