Minimum Variance Optimization: Weighting for Lowest Portfolio Volatility

Minimum variance optimization is a portfolio construction technique that calculates the combination of asset weights producing the lowest expected portfolio volatility. Rather than treating all selected assets equally or weighting by momentum scores, the optimizer uses the historical covariance matrix — capturing each asset's volatility and its correlation with every other asset — to find the mathematically optimal blend that minimizes total portfolio risk.

How It Works

The optimizer takes two inputs: the set of eligible assets and a covariance matrix estimated from recent daily returns (typically the trailing 126 trading days). It then solves for the long-only weight vector that minimizes the portfolio variance: w = argmin(w'Σw), subject to the constraint that weights sum to one and are non-negative.

The solution naturally produces several desirable properties. High-volatility assets receive lower weights. Highly correlated assets receive lower combined weight (because their co-movement contributes disproportionate portfolio variance). Low-correlation assets receive elevated weight (because they reduce portfolio variance through diversification). The optimizer effectively identifies the "sweet spot" where portfolio risk is minimized given the available building blocks.

Why Optimize for Minimum Variance

The research motivation comes from a counterintuitive finding in portfolio theory: minimum variance portfolios have historically delivered comparable returns to cap-weighted or equal-weighted portfolios while experiencing substantially lower volatility. This suggests that the market does not efficiently compensate investors for bearing volatility risk — a finding documented across markets and time periods by researchers including Clarke, de Silva, and Thorley (2006).

In the context of tactical allocation, minimum variance optimization adds a risk management layer that pure momentum selection lacks. A momentum screen identifies what is trending well. The optimizer determines how to combine those trending assets for the least possible risk. This dual filter — momentum for selection, optimization for weighting — produces portfolios with both strong trend exposure and controlled risk characteristics.

Strategies That Use Minimum Variance

• Adaptive Asset Allocation (AAA) — selects top 5 by momentum, then min-var optimizes weights
• Defensive Adaptive Allocation (KDA) — combines canary protection with min-var during risk-on periods

Practical Considerations

Minimum variance optimization is sensitive to the estimation window for the covariance matrix. Shorter windows capture current market dynamics but are noisier. Longer windows are more stable but may reflect outdated correlation structures. Most implementations use a trailing six-month daily window as a compromise. The optimizer can also produce extreme weights when the covariance matrix is poorly estimated — a problem addressed by imposing position size constraints or using shrinkage estimators to regularize the input matrix.