The mathematics of diversification
For a two-asset portfolio with weights w and (1−w), the portfolio variance is: σ²ₚ = w²σ²₁ + (1−w)²σ²₂ + 2w(1−w)ρ₁₂σ₁σ₂, where ρ₁₂ is the correlation between the two assets. When ρ₁₂ = 1 (perfect positive correlation), no diversification benefit exists — portfolio risk is simply the weighted average of individual risks. As correlation falls below 1, portfolio variance falls below the weighted average. When ρ₁₂ = −1 (perfect negative correlation), it is theoretically possible to construct a zero-variance portfolio. In practice, correlations between most risky assets are positive but below 1 (0.2–0.7 for equities across regions and sectors), providing meaningful but imperfect diversification. For N assets, the portfolio variance involves N variance terms and N(N−1)/2 covariance terms — as N grows, covariances dominate and the marginal benefit of adding assets diminishes.
Practical limits of Markowitz optimisation
The theory is elegant; its practical implementation is beset by problems. Input sensitivity: the optimiser is extraordinarily sensitive to small changes in expected return estimates — a tiny change in an asset’s expected return produces dramatically different portfolio weights. Since expected returns are the most uncertain inputs (estimated from historical data with very low signal), optimised portfolios are often extreme and concentrated, not diversified. Estimation error: with N assets, there are N expected returns, N variances, and N(N−1)/2 covariances to estimate — all subject to error. Errors in covariances are particularly insidious. Correlation instability: correlations between assets are not constant. In crises, correlations between risky assets spike toward 1 — exactly when diversification is most needed, it provides least benefit. Non-normality: the framework assumes returns are normally distributed; real returns have fat tails and skew, making variance a poor risk measure.
“Diversification is the only free lunch in investing — but Markowitz’s contribution was showing exactly how much of that lunch you’re leaving on the table with a naive portfolio.” — Andrew Lo, paraphrased
What this means for you
Modern portfolio theory is the conceptual foundation for multi-asset investing: the efficient frontier, diversification, and the risk-return tradeoff are taught in every finance course for good reason. In practice, however, direct Markowitz optimisation should be approached with caution. Robust alternatives include risk parity (equal risk contribution from each asset), maximum diversification (maximise the portfolio’s diversification ratio), and Black-Litterman (combine market equilibrium returns with investor views). For most investors, a well-diversified low-cost index fund spanning global equities and bonds captures the most important benefits of Markowitz diversification without the input sensitivity of explicit optimisation. The insight that matters: holding a diversified mix of uncorrelated assets improves the return per unit of risk — the specific weights matter less than the principle.