Estimated study time: 45 minutes
Content:
Mean-variance optimization (MVO) is the mathematical foundation of modern portfolio theory, originating with Harry Markowitz's 1952 paper. The core insight is that investors care about both expected return and variance (risk), and that diversification reduces portfolio variance without proportionally reducing expected return. The efficient frontier represents the set of portfolios that offer the maximum expected return for a given level of risk, or equivalently, the minimum risk for a given expected return.
Inputs to MVO are expected returns, variances, and covariances (or correlations) for all assets in the opportunity set. These inputs are the primary source of MVO's practical weakness: small changes in expected return estimates cause large changes in portfolio weights, making the optimization highly sensitive to estimation error. This "error maximization" problem means MVO tends to produce concentrated portfolios that are optimal in-sample but perform poorly out-of-sample.
Several techniques are used to improve MVO robustness. Resampling averages optimal portfolios across many simulated input sets to produce more stable, diversified results. The Black-Litterman model combines a market equilibrium prior (from CAPM) with the investor's own views using a Bayesian framework, producing more balanced portfolios than unconstrained MVO. Adding constraints — such as maximum position sizes, minimum diversification requirements, or turnover limits — also improves practical performance.
The efficient frontier shifts under different assumptions about the risk-free rate and borrowing. The Capital Market Line (CML) connects the risk-free asset to the tangency portfolio — the point on the efficient frontier with the highest Sharpe ratio. All investors combining risky and risk-free assets optimally hold the tangency portfolio combined with risk-free borrowing or lending. The CAPM extends this to an equilibrium where the tangency portfolio equals the market portfolio.
At Level 3, candidates must also understand asset-only versus asset-liability MVO. Asset-only optimization minimizes portfolio return variance, which is appropriate for investors with no liability structure. Asset-liability MVO minimizes surplus volatility — the variance of assets minus liabilities — which is relevant for pension funds, insurance companies, and any investor with a defined liability stream. Adding a liability benchmark changes the efficient frontier: assets that are highly correlated with liabilities are now "lower risk" even if they have high standalone volatility, because they hedge the liability risk.
Key Terms:
Quiz Questions:
Q1. A portfolio manager is applying mean-variance optimization to build a multi-asset portfolio. She notices that the optimization produces a 70% allocation to one asset class. The most likely cause of this concentration is:
A) The optimizer correctly identified that one asset class dominates on a risk-adjusted basis B) Mean-variance optimization tends to amplify estimation errors, producing extreme weights C) The risk-free rate assumption is too low, causing overweighting of higher-yield assets D) The covariance matrix is under-estimated, causing the optimizer to ignore diversification benefits
Answer: B — MVO is sensitive to small errors in input estimates. Small overestimates of one asset class's expected return lead the optimizer to allocate heavily to that asset — this is the "error maximization" problem. The Black-Litterman model and resampling are used to mitigate this.
---
Q2. A pension fund manager wants to minimize the risk of funding shortfalls. She is choosing between asset-only MVO (minimizing portfolio variance) and asset-liability MVO (minimizing surplus variance). Which approach is more appropriate?
A) Asset-only MVO, because it minimizes the volatility of the portfolio the trustee is responsible for B) Asset-liability MVO, because the relevant risk for the pension fund is the variance of assets relative to liabilities C) Asset-only MVO, because liabilities are the responsibility of the plan sponsor, not the manager D) Both are equally appropriate; the choice depends on the liability discount rate
Answer: B — For an investor with a defined liability structure (pension obligations, insurance reserves), the relevant risk is surplus volatility — not portfolio volatility alone. An asset with high standalone risk but high correlation with liabilities actually reduces surplus risk. Asset-liability MVO captures this correctly.
---
Q3. The Black-Litterman model addresses a key weakness of standard MVO by:
A) Restricting maximum position sizes to prevent concentration B) Combining CAPM equilibrium returns with investor views to produce more balanced portfolios C) Using historical returns instead of forecasted returns as optimizer inputs D) Replacing variance with downside deviation as the risk measure
Answer: B — Black-Litterman starts with an equilibrium prior (implied by market cap weights and CAPM) and blends in investor views with uncertainty. The result is a set of expected returns that leads to well-diversified portfolios rather than the extreme concentrations from unconstrained MVO.
---
Q4. An analyst places Asset A (expected return 8%, standard deviation 15%) and Asset B (expected return 6%, standard deviation 10%) on an efficient frontier. He then introduces a risk-free asset yielding 3%. The optimal portfolio for a risk-averse investor who can lend at the risk-free rate is:
A) A combination of Asset A and the risk-free asset B) A combination of Asset B and the risk-free asset C) The tangency portfolio combined with the risk-free asset, where the tangency portfolio has the highest Sharpe ratio D) The minimum-variance portfolio, regardless of the risk-free rate
Answer: C — When a risk-free asset is available, the Capital Market Line dominates the efficient frontier. All investors should hold the tangency portfolio (the risky portfolio with the highest Sharpe ratio) combined with either lending (risk-free investment) or borrowing, depending on their risk tolerance.
---
Q5. A fund manager is using resampling to improve the stability of her MVO portfolios. Resampling involves:
A) Running the optimizer 1,000 times with different risk aversion parameters and averaging the results B) Simulating many sets of inputs drawn from the estimated distribution of parameters and averaging the resulting optimal portfolios C) Constraining the optimizer to hold all assets in proportion to their market capitalization D) Replacing the expected return inputs with consensus analyst forecasts
Answer: B — Resampling generates many simulated input sets by drawing from the estimated distribution of expected returns, variances, and covariances. Each simulated set yields an optimal portfolio. Averaging across these portfolios produces a resampled efficient frontier with more stable, diversified weights.
---