Third-Party Research & Methodology Only

This section shares summaries of third-party academic research and descriptions of quantitative models. The content represents the findings of the original researchers, not the opinions or recommendations of Foxholm Financial. Foxholm Financial does not publish hypothetical or backtested performance metrics on its quantitative research pages. All content is restricted to methodology, signal construction, factor logic, and risk architecture. SEC rules require that investment advisers not present misleading performance data, and our methodology-only approach reflects that standard and the firm's fiduciary obligations.

Mean-Variance Optimization

Optimization Method Portfolio Construction Foundational Theory

Mean-variance optimization (MVO) is the mathematical framework for building portfolios that maximize expected return for a given level of risk, or minimize risk for a given level of expected return. It is the foundation of Modern Portfolio Theory and remains one of the most influential ideas in finance.

Developed by Harry Markowitz in 1952, MVO formalized a simple but powerful insight: the risk of a portfolio depends not just on how risky each individual asset is, but on how those assets move in relation to one another. By combining assets whose prices do not move in lockstep, an investor can reduce the overall portfolio's risk without sacrificing expected return. This idea, known as diversification, is at the heart of the framework.

Definition

Mean-variance optimization is a quantitative method that selects portfolio weights to achieve the most favorable tradeoff between expected return (the "mean") and risk (measured by variance, or equivalently, standard deviation). The optimizer takes a set of inputs, applies mathematical constraints, and produces the combination of asset weights that sits on the efficient frontier.

Core Concept

Given a set of assets, MVO seeks to identify the portfolio weights intended to maximize expected return for a chosen level of risk, or minimize risk for a chosen level of expected return.

The key insight is that portfolio risk is not simply the weighted average of individual asset risks. Because assets are not perfectly correlated (they do not always move together), combining them can produce a portfolio with less total risk than any single holding alone.

How It Works

The optimization process requires three categories of inputs and produces one output: a set of efficient portfolio weights.

Inputs

Expected returns vector: A forecast of the average return for each asset over the investment horizon. This is often the most difficult input to estimate accurately and the one to which results are most sensitive.
Covariance matrix: A table that describes how each pair of assets moves together. The diagonal entries represent each asset's own variance (risk), while the off-diagonal entries represent the covariance between pairs. This matrix captures the diversification benefit of combining assets.
Constraints: Rules the optimizer must follow. Common constraints include a budget constraint (weights must sum to 100%), a no-short-selling constraint (no negative weights), and upper or lower bounds on individual positions.

Output

The optimizer produces a set of portfolio weights, one for each asset, that define an optimal portfolio. By varying the target return or risk level and re-running the optimization, a series of optimal portfolios can be traced out. This collection of portfolios forms the efficient frontier.

The Efficient Frontier

The efficient frontier is the curve of all portfolios that offer the highest expected return for each level of risk. Any portfolio that lies below the frontier is suboptimal because it is possible to achieve either more return for the same risk, or less risk for the same return, by reallocating weights.

The shape of the frontier depends entirely on the inputs. With only two assets, the frontier is a simple curve. With many assets, it becomes a complex surface in higher-dimensional space, though it is typically plotted in two dimensions with risk (standard deviation) on the horizontal axis and expected return on the vertical axis.

When a risk-free asset is available, the optimal strategy involves holding a combination of the risk-free asset and the "tangency portfolio," the portfolio on the efficient frontier with the highest Sharpe ratio (the most excess return per unit of risk). This line, drawn from the risk-free rate through the tangency portfolio, is called the capital market line.

Practical Example

Consider a simple three-asset portfolio consisting of U.S. stocks, international stocks, and bonds. An investor supplies expected returns, standard deviations, and correlations for each asset. The optimizer then finds the weights that produce the best risk-return tradeoff.

Asset	Expected Return	Standard Deviation	Equal-Weight Allocation	MVO Allocation
U.S. Stocks	8.0%	16.0%	33.3%	45%
International Stocks	7.0%	18.0%	33.3%	15%
Bonds	4.0%	5.0%	33.3%	40%
Portfolio Expected Return			6.3%	6.4%
Portfolio Standard Deviation			10.2%	8.7%

In this example, the optimized portfolio achieves a slightly higher expected return while taking meaningfully less risk than the equal-weight portfolio. The optimizer accomplishes this by tilting toward the low-correlation bond allocation, which reduces overall portfolio variance. The exact weights depend on the assumed correlations between assets.

Known Limitations

Limitations to Keep in Mind

Extreme sensitivity to inputs. Small changes in expected return estimates can produce large swings in the recommended portfolio weights. An asset's allocation might jump from 5% to 50% based on a fraction-of-a-percent change in its expected return. This makes the output unreliable when inputs are uncertain, which they almost always are.
Estimation error in expected returns. Expected returns are notoriously difficult to forecast. Historical averages are noisy estimates of future returns, and small errors in these estimates get amplified by the optimizer. As Richard Michaud famously argued, MVO tends to "maximize the effect of estimation errors."
Corner solutions. Without constraints, the optimizer often produces extreme portfolios that concentrate heavily in a small number of assets or take large short positions. These "corner solutions" are mathematically optimal but practically unusable, which is why real-world implementations always add constraints.
Assumes returns follow a normal distribution. The framework uses variance (or standard deviation) as the sole measure of risk. This treats large gains and large losses symmetrically and underestimates the frequency of extreme events (fat tails). In reality, financial returns often exhibit skewness and heavier tails than a normal distribution would predict.
Static, single-period framework. Standard MVO optimizes for a single time period and does not account for how portfolios should evolve over time, changes in market conditions, or the cost of rebalancing.

Extensions and Alternatives

Because of these well-known limitations, researchers and practitioners have developed several extensions that build on the MVO framework while addressing its weaknesses.

Black-Litterman model: Instead of relying solely on historical returns, the Black-Litterman approach starts with the market equilibrium (the returns implied by current market capitalizations) and blends in an investor's own views using Bayesian statistics. This produces more stable, intuitive portfolios and reduces the sensitivity to return estimates.
Robust optimization: These methods explicitly account for uncertainty in the inputs by optimizing for the worst case within a range of plausible estimates. The resulting portfolios sacrifice some expected performance but are far less sensitive to estimation errors.
Shrinkage estimators: Rather than using raw sample estimates of the covariance matrix, shrinkage methods blend the sample covariance with a structured target (such as a diagonal matrix or a single-factor model). This reduces estimation noise and produces more stable portfolio weights. The Ledoit-Wolf shrinkage estimator is among the most widely used.
Resampled efficiency: Introduced by Richard Michaud, this technique runs the optimizer many times on simulated variations of the inputs and averages the resulting weights. This smooths out the instability of single-run optimization.

Academic Origin

Harry Markowitz introduced mean-variance optimization in his 1952 paper "Portfolio Selection," published in The Journal of Finance. The paper proposed that investors should evaluate portfolios based on their expected return and variance of return, rather than selecting individual securities in isolation. This was a foundational departure from the prevailing approach, which focused on picking stocks one at a time without a formal framework for considering how they interact within a portfolio.

Markowitz expanded the framework into a full book, Portfolio Selection: Efficient Diversification of Investments, published in 1959. His work earned him a share of the 1990 Nobel Memorial Prize in Economics, alongside William Sharpe and Merton Miller. The framework he established remains the starting point for virtually all modern portfolio construction methods, even those designed to improve upon its original formulation.