# Mean-Variance Optimization: Markowitz's Framework for Portfolio Construction

Cite as: Gorak, R. (2026). Mean-Variance Optimization: Markowitz's Framework for Portfolio Construction. Tradicted. https://www.tradicted.com/research/markowitz-portfolio-1952/
Paper: Harry Markowitz — *Portfolio Selection*
Published in: Journal of Finance (1952)
Original: https://www.jstor.org/stable/2975974

Key finding: The E-V rule implies investors should hold diversified portfolios on the efficient frontier — minimizing variance for a given expected return or maximizing return for a given variance.

---
Mean-variance optimization selects portfolios by minimizing variance and maximizing expected return jointly. Markowitz (1952) proved portfolio variance equals V = Σ σ_ij X_i X_j in "Portfolio Selection." Maximizing expected return alone never implies diversification. The E-V rule implies diversification for a wide range of μ_i, σ_ij combinations.

## What the Study Found

Markowitz (1952) proves maximizing expected return always concentrates all weight in a single security. The E-V rule implies diversification for a wide range of μ_i, σ_ij. Two equal-variance portfolios, when combined, produce a compound portfolio with lower variance than either original — except when returns are perfectly correlated. Sixty railway securities are less diversified than an equal-size cross-industry portfolio. Within-industry covariances exceed cross-industry covariances.

## Methodology

Markowitz (1952) presents a theoretical model, not an empirical study — no dataset or sample period applies. Inputs are μ_i (expected return) and σ_ij (covariance) per security. Portfolio weights X_i ≥ 0 sum to 1, with no short sales permitted. The 3- and 4-security cases are solved geometrically; the N-security result is stated analytically.

## Key Statistics

| Metric | Finding | Context |
|---|---|---|
| Portfolio variance | V = Σ_i Σ_j σ_ij X_i X_j | Double-weighted covariance sum across all N securities |
| Portfolio expected return | E = Σ_i X_i μ_i | Weighted sum of individual security expected returns |
| Two-portfolio diversification | Variance of compound < variance of either original | Holds unless returns are perfectly correlated |
| Inadequate diversification example | Sixty railway securities | Less diversified than equal-size cross-industry portfolio |
| Efficient set structure | Series of connected line segments | Applies to 3-, 4-, and N-security cases |

## Why This Matters

The E-V framework gave portfolio construction a mathematical foundation separate from stock-picking heuristics. Markowitz's proof that cross-industry diversification reduces variance in ways within-industry diversification cannot remains the basis for institutional asset allocation and risk budgeting. The efficient frontier gives practitioners a precise vocabulary for the risk-return trade-off.

## FAQ

### What is mean-variance optimization in portfolio construction?

V = Σ σ_ij X_i X_j is minimized for a given expected return. Markowitz (1952) showed this traces an efficient frontier — a series of connected parabola segments in E-V space. Each point represents a portfolio no other dominates on both return and variance.

### Why doesn't maximizing expected return produce a diversified portfolio?

Markowitz (1952) proves that maximizing R = Σ X_i R_i always places all weight in the single highest-return security. No diversified portfolio is ever preferred under this rule. Only adding variance minimization produces diversified holdings.

### What does research show about cross-industry vs. within-industry diversification?

Markowitz (1952) states that sixty railway securities are less diversified than an equal-size portfolio spanning railroad, utility, mining, and manufacturing. Firms within the same industry have higher pairwise covariances than firms in dissimilar industries. Reducing V requires low-covariance securities, not merely many.

### How does the efficient frontier apply to institutional portfolio construction?

Markowitz (1952) defines the efficient set as a series of connected line segments in the N-security case. Minimum variance sits at one end; maximum expected return at the other. Each point corresponds to a specific weight vector X_i computed from estimates of μ_i and σ_ij.

## Source

Markowitz, H. (1952). Portfolio Selection. *Journal of Finance*, 7(1), 77–91.

[Read the full paper →](https://cowles.yale.edu/sites/default/files/files/pub/mono/m16-all.pdf)