Testing Averaged Portfolio Weights Against Shrinkage Methods

Summary

This study asks whether averaging portfolio weights from different mean-variance shrinkage methods can improve out-of-sample results. It compares individual grid settings, ranging from no shrinkage to full shrinkage of estimated Sharpe ratios and correlations, with combinations that average selected portfolios. The motivation is that averaging weights might be more robust than averaging model inputs, particularly when the best shrinkage choice is uncertain.

Results are reported across several in-sample and out-of-sample window combinations using median Sharpe ratio, a low-tail Sharpe measure, and t-statistics. No averaged combination consistently dominates the underlying methods. Outcomes shift with the evaluation horizon: some combinations are competitive in certain cases, while others perform poorly, and some reported statistics are missing or zero. The tables support a cautious conclusion that weight averaging is not a dependable fix for estimation uncertainty in these experiments. The document gives no asset universe, detailed portfolio constraints, or complete account of the experiment, so the findings should not be generalized beyond the reported setup.

Key ideas

• The experiment compares averages of portfolio weights with individual shrinkage settings in mean-variance optimization.
• Performance is evaluated across multiple combinations of in-sample and out-of-sample horizons.
• The reported tables do not show a single averaging scheme that consistently leads.
• Results vary by evaluation window, and some reported test statistics are missing or zero.
• The limited methodological detail makes it difficult to generalize the findings beyond this experiment.

Tags

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