What Is Mean-Variance Optimization? Portfolio Analysis Explained
By Markets & Investing DeskUpdated Nov 10, 2026
Definition
Mean-variance optimization is a method for choosing portfolio weights that give the highest expected return for a given level of risk, measured by variance, or the lowest risk for a target return.
Key takeaways
- Harry Markowitz introduced the framework in his 1952 paper "Portfolio Selection," the basis of modern portfolio theory.
- The model needs expected returns, variances and pairwise correlations or covariances for every asset considered.
- Portfolio variance sums every covariance term, so lower correlations shrink the cross term and cut total risk.
- Small errors in return estimates can swing optimal weights sharply, a flaw critics call "error maximization."
Mean-variance optimization (MVO) is a method for choosing portfolio weights that deliver the highest expected return for a given level of risk, or the lowest risk for a given expected return, where risk is measured by the variance (or standard deviation) of portfolio returns. Harry Markowitz introduced the framework in his 1952 paper “Portfolio Selection,” which became the foundation of modern portfolio theory (MPT) and contributed to his 1990 Nobel Memorial Prize in Economic Sciences.
The key insight is that a portfolio’s risk depends not only on how volatile each holding is but on how the holdings move relative to one another. Two risky assets that do not move in lockstep can form a portfolio less volatile than the weighted average of their individual volatilities.
The Inputs and the Formula
A mean-variance model needs three sets of estimates for every asset under consideration:
- Expected return (the mean): the return the investor anticipates over the holding period.
- Variance or standard deviation: how widely returns are expected to spread around that mean.
- Correlations or covariances: how each pair of assets tends to move together, on a scale from -1 to +1 for correlation.
From these, two portfolio figures follow. Expected return is a simple weighted sum:
E(Rp) = Σ wᵢ × E(Rᵢ)
Portfolio variance is not a weighted average. It sums every pairwise covariance term:
σp² = Σᵢ Σⱼ wᵢ wⱼ σᵢⱼ
For two assets, that expands to w_A²σ_A² + w_B²σ_B² + 2·w_A·w_B·σ_A·σ_B·ρ_AB. The cross term is where diversification lives: the lower the correlation ρ, the smaller the cross term and the lower the total risk.
The optimization itself solves one of two equivalent problems. Either minimize portfolio variance subject to the weights summing to 100% and hitting a target expected return, or maximize expected return minus a penalty for variance scaled by the investor’s risk aversion. The second form reflects mean-variance preferences: the assumption that investors care only about the mean and variance of outcomes, liking the first and disliking the second. Practical versions add constraints such as no short selling, position limits or sector caps.
The Efficient Frontier
Running the optimization for every possible target return traces a curve on a mean-variance plot, with standard deviation on the horizontal axis and expected return on the vertical. The upper edge of that curve is the efficient frontier (also called the mean-variance frontier). Every portfolio on it offers the best available return for its risk level.
Portfolios below the frontier are mean-variance inefficient: another mix delivers more return for the same risk, or the same return for less. The leftmost point of the curve is the global minimum variance portfolio, the lowest-risk combination available regardless of return.
When a risk-free asset is added, the best risky portfolio becomes the one where a straight line from the risk-free rate touches the frontier. This is the tangency portfolio, and it has the highest Sharpe ratio of any risky combination. Investors then adjust overall risk by blending it with the risk-free asset rather than moving along the curve.
Worked Example
Hypothetical figures for illustration only. An investor holds $100,000 in Investment A and $300,000 in Investment B, a $400,000 total.
| Investment A | Investment B | |
|---|---|---|
| Weight | 25% | 75% |
| Expected return | 5% | 10% |
| Standard deviation | 7% | 14% |
Assume a correlation of 0.65 between them.
Expected return: (0.25 × 5%) + (0.75 × 10%) = 8.75%
Variance: (0.25² × 0.07²) + (0.75² × 0.14²) + (2 × 0.25 × 0.75 × 0.07 × 0.14 × 0.65) = 0.000306 + 0.011025 + 0.002389 = 0.01372
Standard deviation: √0.01372 ≈ 11.71%
A naive weighted average of the two standard deviations would be 12.25%. The 0.54-point gap is the diversification benefit from a correlation below 1.
The optimization step asks whether a different split improves the trade-off. Here the correlation (0.65) exceeds the ratio of the two volatilities (7% ÷ 14% = 0.5), so adding B never lowers risk relative to holding A alone. Under a no-short-sale constraint, the minimum variance portfolio is 100% A at 7%. Without that constraint, the formula would put about 112.5% in A and short 12.5% of B, a position many investors cannot or would not hold.
How It Is Used
Mean-variance optimization underpins strategic asset allocation at pension funds, endowments, model-portfolio providers and robo-advisors, often applied to asset classes or ETFs rather than individual stocks. Advisors use it to map a client’s risk tolerance to a point on the frontier. It also serves as a benchmark: a proposed portfolio can be checked against the frontier to see how far it sits from efficient.
Limitations
- Estimation error: the output is only as good as the inputs, and expected returns are especially hard to forecast. Small changes in return estimates can swing the optimal weights sharply, a tendency critics call “error maximization” because the optimizer overweights assets whose returns happen to be overestimated.
- Concentrated or extreme portfolios: unconstrained solutions often pile into a few assets or take large short positions.
- Variance treats upside and downside alike: a surprise gain counts as risk just as a loss does, and fat tails or skewed returns are not captured.
- Unstable correlations: relationships estimated from history can break down, often in market stress when diversification is most needed.
- Single period: the classic model optimizes over one horizon and ignores taxes, transaction costs and rebalancing.
Refinements address these issues. The Black-Litterman model starts from market-implied returns and blends in investor views; resampled efficiency averages many optimizations; shrinkage estimators stabilize covariance inputs; and risk parity sidesteps return forecasts entirely by balancing risk contributions.
Commonly Confused Terms
Mean-variance analysis describes evaluating assets by expected return and variance; optimization is the step of solving for the best weights. Modern portfolio theory is the broader body of ideas that includes both. The Capital Asset Pricing Model builds on the mean-variance framework with a market portfolio, but it is a pricing model for expected returns, not an allocation procedure. Portfolio variance is one output of MVO, not the method itself.
The Bottom Line
Mean-variance optimization matters to pension funds, endowments, advisors and robo-advisors that set strategic asset allocations and check portfolios against the efficient frontier. Its results depend heavily on input estimates, especially expected returns, and unconstrained solutions can produce concentrated or short positions. Constraints and refinements such as Black-Litterman, resampled efficiency, shrinkage estimators or risk parity help make the output more stable and usable.
