Portfolio Optimization with Shrinkage on Real Trading Returns

Summary

This study compares portfolio optimization methods using real trading-rule returns. Each trial samples nine rules from one instrument, varies the available in-sample history, and evaluates performance out of sample. Methods include mean-variance portfolios with different degrees of shrinkage applied to estimated Sharpe ratios and correlations, equal weighting, maximum diversification, empirical portfolio optimization, Monte Carlo, and bootstrap approaches.

Across the reported experiments, the author finds that shrinkage generally helps, particularly for correlations, whose estimates appear less stable than Sharpe ratios. The preferred settings vary with the length of available history; a heuristic increases shrinkage as data becomes shorter, reaching equal weights with less than a year of data. Monte Carlo and bootstrap methods perform poorly in many cases and are slower in this study. The results are noisy, and statistical tests often fail to distinguish shrinkage combinations. The conclusions are specific to the author's rules and holding periods, so the suggested settings should be treated as context-dependent guidance rather than universal parameters.

Key ideas

• The study evaluates portfolio optimizers using repeated samples of real trading-rule returns and out-of-sample periods.
• Shrinkage of estimated correlations is generally more useful than the author initially expected.
• The suggested balance of Sharpe-ratio and correlation shrinkage changes with the length of available data.
• Monte Carlo and bootstrap optimizers are slower and often underperform in the reported real-data experiments.
• The optimization results are noisy, and the proposed shrinkage schedule is domain-specific.

Tags

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