Definition
A mean-variance portfolio chooses weights using expected return and covariance to trade off the portfolio's mean against variance under stated constraints. The Markowitz problem is sensitive to estimation error, especially in expected returns and covariance matrices. The efficient frontier is a set of solutions for different target or penalty choices, not one universal portfolio.
How it is calculated
Minimize one-half portfolio variance w-transpose Sigma w minus lambda times expected return mu-transpose w, subject to weights summing to one.
| Symbol | Meaning | Unit |
|---|---|---|
| w | portfolio weight vector | fraction |
| Sigma | return covariance matrix | return squared per period squared |
| lambda | risk-return tradeoff parameter | return units |
| mu | expected return vector | return per period |
| 1 | ones vector | dimensionless |
Interpretation
The optimizer rewards expected return and penalizes covariance-based portfolio variance. That sounds tidy, but it asks the data to estimate many inputs before the weights are chosen. Small input errors can create large weight changes, especially without stabilizing constraints.
Assumptions and conventions
-
Expected returns, covariance, horizon and constraints are defined.
-
lambda >= 0 for the stated return reward.
-
Conditions ensure a finite attained optimum; singular covariance with unrestricted weights can otherwise make the objective unbounded.
-
The objective's risk-aversion convention is explicit.
-
Estimation error and turnover are considered in interpretation.
Weights are fractions; covariance is squared return units; lambda gives compatible objective units.
Limitations and common misreadings
-
The efficient frontier is conditional on estimates and constraints.
-
Minimum estimated variance is not necessarily minimum realized variance.
References
- Harry Markowitz. (1952). PORTFOLIO SELECTION*. The Journal of Finance. Whole article. doi:10.1111/j.1540-6261.1952.tb01525.x. Source

