Level III Core · Principles of Asset Allocation

Mean-Variance Optimization and Its Limitations

Updated 8 October 2026 · Fact-checked

Mean-variance optimization (MVO) uses expected returns, standard deviations and correlations to find portfolios with the highest return for each level of risk. Together they form the efficient frontier. MVO is very sensitive to its inputs and often gives concentrated portfolios. Fixes include constraints, resampling and Black-Litterman.

Understand Mean-Variance Optimization and Its Limitations

Mean-variance optimization (MVO) is a method to pick asset weights. You give it expected returns, standard deviations and correlations for each asset class. It then finds the weights that give the highest expected return for each level of risk. Plot those portfolios and you get the efficient frontier.

The efficient frontier is a curve in expected return and standard deviation space. Every portfolio on it is efficient. No other portfolio gives more return for the same risk, or less risk for the same return. The investor picks the point that fits the client's risk tolerance, objectives and constraints. Risk here means standard deviation, so MVO treats upside and downside moves the same.

MVO looks precise, but it is only as good as its inputs. Small changes in expected returns can swing the optimal weights a lot. This is input sensitivity. Expected returns are the hardest input to estimate and matter most. The optimizer tends to load up on assets with high estimated returns, low estimated risk or low correlations. Those are often the assets where estimation error is largest. This is why people call MVO an "error maximizer".

This gives the main limitations. Outputs can be highly concentrated, with a few assets getting large weights and others zero. Weights can be unstable from one period to the next, which causes turnover. MVO is a single-period model and ignores liabilities, taxes and illiquidity unless you add them. It assumes returns are normal and risk is captured by standard deviation, so it misses skewness, fat tails and tail risk.

There are practical fixes. Constraints such as maximum and minimum weights, or no short sales, force diversification but can hide poor inputs. Resampled MVO (Michaud) runs many simulations of inputs, optimizes each one and averages the weights. It gives more diversified and stable portfolios, but it is a black box and has no strong theory behind it. The Black-Litterman model starts from market-equilibrium returns implied by market-cap weights and adds the investor's views, scaled by confidence in them. Without views, you hold the market portfolio. It gives more intuitive, diversified weights. Other fixes include shrinkage of estimates, using other risk measures, and reverse optimization, which backs out the returns implied by a given set of weights.

Key rules to remember

Portfolio expected return: E(Rp) = Σ wi × E(Ri); Weights sum to 1. Use the same time unit for all inputs.
Two-asset portfolio variance: σp² = w1²σ1² + w2²σ2² + 2 w1 w2 ρ12 σ1 σ2; Standard deviation is the square root. Lower correlation gives lower risk.
Covariance and correlation: Cov(1,2) = ρ12 × σ1 × σ2; MVO needs a covariance matrix. The number of inputs grows quickly with the number of assets.
Efficient frontier rule: Efficient if no portfolio has higher E(R) at equal or lower σ; Portfolios below the frontier are dominated.
Black-Litterman starting point: Equilibrium returns come from reverse optimization of market-cap weights; Views then tilt these returns, scaled by confidence in each view.

How to solve Mean-Variance Optimization and Its Limitations questions

Use this method for any MVO question, whether it asks you to interpret a frontier, critique a result or recommend a fix.

1Read the client's objectives and constraints first. Note risk tolerance, horizon, liquidity needs and any limits on weights or short sales.
2Identify the inputs given: expected returns, standard deviations, correlations. Check they are in the same time unit and currency.
3If asked to compute, find portfolio return and variance with the formulas, showing each step. Then compare risk-return trade-offs.
4If asked to pick a portfolio, choose the efficient portfolio that meets the client's risk limit or return target. Dominated portfolios are never the answer.
5If asked to evaluate output, look for concentration, extreme weights, unusual inputs and instability. Link each to input sensitivity or estimation error.
6Name the limitation using the exact term: input sensitivity, concentration, single-period, normality, or ignoring liabilities and illiquidity.
7Recommend a specific fix that matches the problem: constraints, resampling, Black-Litterman, shrinkage or a different risk measure.
8State the answer in the fewest words that match the command word. Give one reason tied to the case.

Quickest way: Diagnose, then match the fix

When to use it: Use for item-set questions asking which limitation or remedy applies, when you have about a minute per question.

Spot the symptom: large weights in few assets, weights that jump after small input changes, or portfolios that ignore the client's liabilities.
Match the cause: extreme weights point to estimation error and input sensitivity; instability points to sensitivity; ignored liabilities point to asset-only single-period design.
Match the fix: constraints for quick diversification, resampling for stability, Black-Litterman for combining equilibrium with views.
Eliminate options that overstate what a fix does. No fix removes estimation error completely.

Common mistakes in Mean-Variance Optimization and Its Limitations

Saying MVO is wrong because the math is flawed.
Students confuse the model with its inputs.
Fix: Say the math works but outputs are very sensitive to estimated inputs, especially expected returns.
Picking a portfolio below the efficient frontier because it matches the client's risk exactly.
Focus on risk match, not dominance.
Fix: For the same risk, choose the portfolio with higher return on the frontier.
Claiming resampling removes estimation error.
Resampling gives smoother, more diversified weights, so it sounds like a cure.
Fix: Say it reduces instability and concentration but still relies on the original inputs, and it lacks a strong theoretical basis.
Thinking Black-Litterman starts from the investor's views.
The word views is memorable.
Fix: It starts from equilibrium returns implied by market-cap weights, then adjusts for views by confidence. With no views you get the market portfolio.
Treating constraints as a free fix.
Constraints produce neat, diversified weights.
Fix: Note that constraints, not the inputs, may then drive the result and can mask poor estimates.
Ignoring the client's context in a recommendation.
Students treat MVO as pure theory.
Fix: Tie the choice to the client's objectives, horizon, liabilities and constraints, and mention any ignored by a single-period asset-only model.

Worked examples

Example 1

An optimizer gives a portfolio of 70% emerging market equity, 30% global bonds and nothing else. A small increase in the emerging market return estimate changes weights to 95% and 5%. Identify the limitations shown and recommend one fix.

Show the solution

The portfolio is concentrated in just two asset classes, with zero weight in all others. The shift to 95% and 5% shows even greater concentration and instability.
The large shift from a small input change shows input sensitivity and weight instability.
Cause: expected returns are estimated with error, and the optimizer exploits assets whose inputs look best.
Fix: apply resampled MVO, which averages optimal weights across simulated inputs and gives more stable, diversified portfolios. Weight constraints would also limit concentration.

Answer: The output shows concentration and input sensitivity from estimation error. Use resampled MVO (or add weight constraints) to get more stable, diversified weights.

Example 2

Portfolio A has expected return 8% and standard deviation 10%. Portfolio B has 8% and 12%. Portfolio C has 10% and 12%. All are feasible. Which are on the efficient frontier, and what is the variance of a portfolio with 60% in an asset with σ = 10% and 40% in an asset with σ = 20% if the correlation is 0.5?

Show the solution

Compare A and B: same return, B has more risk, so B is dominated by A.
Compare B and C: same risk, C has higher return, so B is dominated by C.
Compare A and C: C has more return and more risk, so neither dominates the other. Among the three given portfolios, A and C are the non-dominated ones and B is dominated.
Efficiency relative to the full feasible set cannot be confirmed from these three alone. A feasible portfolio not listed could still dominate A or C.
Variance = 0.6²×0.10² + 0.4²×0.20² + 2×0.6×0.4×0.5×0.10×0.20.
0.36×0.01 = 0.0036; 0.16×0.04 = 0.0064; 2×0.6×0.4×0.5×0.10×0.20 = 0.0048.
Total variance = 0.0036 + 0.0064 + 0.0048 = 0.0148.
Standard deviation = √0.0148 = 12.17%.

Answer: Among the three portfolios, A and C are non-dominated and B is dominated. Whether A and C lie on the efficient frontier of the full feasible set cannot be confirmed from these three alone. Portfolio variance is 0.0148, a standard deviation of about 12.17%.

Exam tips

Command words matter. If asked to "identify", name the limitation briefly. If asked to "justify", add one reason linked to the case.
Always connect a limitation to a cause, usually estimation error in expected returns, and then to a specific fix.
When a calculation is asked, show each step. A correct number typed alone earns full credit, but working protects you in a multi-part item.
For Black-Litterman, remember the sequence: equilibrium returns, then views weighted by confidence, then optimization.
In asset allocation cases, mention what MVO ignores for this client, such as liabilities, illiquidity or taxes.

Mean-Variance Optimization and Its Limitations in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Mean-Variance Optimization and Its Limitations: frequently asked questions

What are the main limitations of mean-variance optimization?

Outputs are very sensitive to inputs, especially expected returns. Portfolios are often concentrated and unstable over time. It is a single-period model that uses standard deviation as risk, so it ignores skewness, fat tails, liabilities and illiquidity unless adjusted.

How does resampled efficient frontier work?

You simulate many sets of inputs around the original estimates and optimize each set. You then average the resulting weights to build the resampled portfolios. This gives more diversified, stable weights, but it does not remove estimation error and lacks a strong theoretical basis.

What is the Black-Litterman model in simple terms?

It starts with equilibrium expected returns implied by market-cap weights. You then add your own views, each weighted by your confidence. The result is a set of adjusted returns that gives more intuitive, diversified optimal weights than raw MVO.

How do I use the efficient frontier for asset allocation?

Build the frontier from your inputs, then choose the efficient portfolio that fits the client's risk tolerance, return needs and constraints. Check the result for concentration and sensitivity before you recommend it.