Testing Exponential Weighting for Sharpe Ratio Estimates

Summary

This study asks whether exponentially weighted estimates of strategy Sharpe ratios improve portfolio optimization compared with using the full available history. It tests several exponential spans alongside an all-history estimate, across different in-sample and out-of-sample periods. For each subsample, it estimates Sharpe ratios using the selected window, estimates correlations from all in-sample data, applies fixed shrinkage, optimizes, and compares median out-of-sample Sharpe ratios. Some tests also include inverse versions of trading rules to reduce the effect of selecting only apparently successful rules.

The reported results vary across sample lengths and forecast horizons. In several cases longer estimation histories perform better, while other comparisons show little or no statistically clear advantage for a particular span. The overall conclusion is that exponential weighting does not consistently outperform the simpler full-history estimate, though a very slow decay may still help discount stale data. These results are conditional on the instruments, rules, shrinkage settings, and test design used; the article does not establish a universally optimal estimation window.

Key ideas

• The study compares exponentially weighted Sharpe ratio estimates with estimates using all available history.
• It evaluates the alternatives through repeated in-sample optimization and out-of-sample Sharpe ratios.
• Correlations use the full in-sample period, while Sharpe ratios use the candidate estimation windows.
• Results vary by sample length and forecast horizon, with no consistent advantage for exponential weighting.
• The findings depend on the study's selected rules, fixed shrinkage levels, and available histories.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.