Why Covariance Is Everything
Suppose you hold two assets, each with 20% annual volatility. If they move in perfect lockstep (correlation of +1.0), the portfolio has 20% volatility — no benefit from combining them. But if they are uncorrelated (correlation of 0), the portfolio volatility falls to around 14%. If they are negatively correlated (correlation of −1.0), you can theoretically construct a riskless portfolio from two risky assets.
This is the mathematical heart of diversification. The practical implication is profound: what matters when adding an asset to a portfolio is not how risky that asset is in isolation, but how it correlates with what you already hold. A volatile asset with low correlation to the rest of your portfolio can actually reduce overall portfolio risk.
Markowitz formalised this in matrix terms: portfolio variance equals the sum of all pairwise covariances between assets, weighted by their portfolio weights. The formula is deceptively compact but computationally demanding — for a portfolio of 50 assets, it requires estimating 1,225 distinct covariances.