The Modified Gerber Statistic: A Practical Nonparametric Covariance Estimator for Portfolio Construction

This paper introduces the Modified Gerber Statistic (MGS), a threshold-free co-movement statistic for covariance estimation and portfolio construction. The original Gerber statistic counts same-and oppositedirection joint moves only after both assets cross selected thresholds. MGS instead uses a nonparametric rank transformation: it ranks each observation relative to the asset's own history and uses every observation, giving the most influence to episodes in which both assets occupy similarly unusual states and less influence to one-sided events. This limits the effect of extreme raw values, makes unlike assets and signals comparable, and produces a covariance input that can enter an existing portfolio optimizer. We compare MGS with the original Gerber statistic, Ledoit-Wolf shrinkage, and historical covariance in two out-of-sample studies from January 1990 through December 2025. In the 10-asset study, MGS has the highest annualized geometric return at every displayed target-risk level from 3% through 15%. At the 9% target, MGS earns 11.26%, compared with 11.04% for the original Gerber statistic, 10.92% for shrinkage, and 11.05% for historical covariance. An equal-weight return-plus-four-week-divergence illustration reduces realized volatility, maximum drawdown, and turnover in every sleeve; at the 15% target, it raises geometric return from 15.17% to 15.38% and reduces maximum drawdown from-42.71% to-40.01%. In a monthly study of the point-in-time 30 largest S&P 500 constituents, return-only MGS has the highest geometric return among the four core estimators and a higher Sharpe ratio than the original Gerber statistic at every displayed target from 3% through 21%. At the 9% target, return-only MGS earns 9.15% with a 0.57 Sharpe ratio, compared with 8.42% and 0.55 for the original Gerber statistic. Combining return and trend-divergence evidence raises these figures to 9.76% and 0.60. The results support MGS as a practical co-movement estimator and show how managers can incorporate additional evidence about correlation without changing the portfolio construction workflow.