Correlation & Covariance

Covariance measures how two assets' returns move together: Cov(A,B) = E[(Rₐ - μₐ)(Rᵦ - μᵦ)], where μ is the mean return. Positive covariance means the assets tend to move in the same direction; negative covariance means they move opposite. Covariance is hard to interpret in isolation because its magnitude depends on the units of each variable. Correlation standardizes this: ρ(A,B) = Cov(A,B) / (σₐ × σᵦ), producing a dimensionless measure between -1.0 and +1.0. A correlation of +1.0 means perfect positive co-movement; -1.0 is perfect negative (perfect hedge); 0 means no linear relationship.

In portfolio construction: Portfolio Variance = w₁²σ₁² + w₂²σ₂² + 2w₁w₂σ₁σ₂ρ₁₂. The correlation term ρ₁₂ is the key driver of diversification benefit. When ρ = 0, the cross-term disappears and portfolio variance is simply the weighted average of individual variances. When ρ < 0, the cross-term subtracts from portfolio variance — providing genuine risk reduction. This formula shows mathematically why combining even moderately correlated assets (ρ = 0.5) still reduces portfolio variance below the simple weighted average of component variances.

Historical long-run correlations provide the starting point for diversification analysis. US equities and US investment-grade bonds have historically had a correlation of roughly -0.2 to +0.1 — close to zero, with periods of negative correlation during equity downturns (the 'flight to quality' effect). US and international developed equities: 0.7-0.8. US equities and emerging markets: 0.6-0.7. US equities and gold: 0.0 to -0.1. US equities and commodities: 0.1-0.3. These figures suggest that bonds and gold provide the most genuine diversification for equity-heavy portfolios, while international equities provide modest diversification.

Crisis correlation is the critical limitation of historical averages. During severe market downturns — 2008, 2020 — correlations across almost all equity markets converge toward 1.0. Emerging markets, international developed markets, small-cap stocks, and US large-cap stocks all fell simultaneously and dramatically in March 2020. The diversification benefit of geographic or sector diversification disappears precisely when most needed. Only truly non-correlated assets — US Treasuries (flight to safety), gold (fear asset), and long volatility strategies — maintained or increased their diversification benefit during the crisis.

Correlation matrices are used to identify hidden concentration in a seemingly diversified portfolio. A portfolio with 20 positions across different sectors might show high pairwise correlations within specific factor groups — for example, all 5 technology positions highly correlated with each other and with a few 'different sector' stocks that have similar growth/momentum factor exposures. This hidden correlation makes the portfolio's effective diversification much less than the position count implies.

Correlation is dynamic and changes over time. Rolling 60-day correlations can differ substantially from rolling 2-year correlations, especially during regime changes. Monitoring correlation stability as part of ongoing portfolio risk management reveals when assets that previously provided diversification begin co-moving — signaling that the portfolio's risk structure has changed. Some sophisticated risk models use copulas (statistical functions that model dependency structures) rather than linear correlation to better capture the non-linear dependencies that emerge during tail events.