Gerber Statistics for Robust Asset Correlation and Portfolio Optimization

Summary

Gerber statistics offer a threshold-based way to estimate dependence between asset returns. Each return is classified by whether it rises or falls beyond an asset-specific threshold; paired observations count as concordant when both move substantially in the same direction, discordant when they move in opposite directions, and otherwise contribute zero. The resulting measure uses significant moves rather than every return, and the article notes that setting the threshold to zero yields Kendall's tau. It also explains how the measure forms a correlation matrix and can be scaled into a covariance matrix.

The article identifies a weakness in one construction: its resulting covariance matrix can fail to be positive semidefinite. It describes a revised calculation based on dividing paired returns into nine regions, which the authors report produces positive semidefinite matrices in practice. An empirical comparison across equity, bond, real estate, gold, and commodity indexes reports better annualized returns for the Gerber approach than historical covariance and a shrinkage method across tested threshold and volatility-target settings. Those findings depend on the study's sample and long-only, volatility-constrained optimization setup; the article does not provide enough formula detail in the supplied text to reproduce the analysis fully.

Key ideas

• The method counts significant same-direction and opposite-direction return moves while ignoring smaller moves.
• A threshold controls which returns count as significant, and a zero threshold corresponds to Kendall's tau.
• The article reports that one covariance construction can be non-positive-semidefinite and presents a revised construction to address this.
• The empirical comparison reports higher annualized returns for the Gerber approach under the tested settings.
• The evidence is limited to the stated asset sample and long-only, volatility-constrained portfolio optimization setup.

Tags

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