Why Maximum-Sharpe Portfolios Align with Mean-Variance Optimization

Summary

The document explains why maximizing a portfolio’s expected return divided by its volatility leads to weights proportional to the inverse covariance matrix multiplied by expected returns. Because the Sharpe ratio is unchanged when all weights are scaled by the same positive constant, this solution determines the portfolio’s direction but not its overall size. The scale factor is arbitrary in the unconstrained problem.

The answer relates the ratio problem to constrained optimization: fix portfolio risk, maximize expected return, or use a Lagrange multiplier to combine the objective and risk constraint. At an optimum, the gradients align, producing the same direction as the mean-variance objective. Constraints on weights or transaction costs can determine scale and complicate the optimization, though the answer characterizes the resulting problem as convex. The document offers a conceptual derivation rather than empirical evidence, and its stated result assumes the covariance matrix and expected-return vector define the unconstrained setup.

Key ideas

• The maximum-Sharpe portfolio has weights proportional to the inverse covariance matrix times expected returns.
• The Sharpe objective is invariant to multiplying all portfolio weights by a common positive scale.
• Fixing portfolio risk turns the ratio objective into maximizing expected return under a constraint.
• A Lagrange multiplier links this constrained formulation to mean-variance optimization.
• Weight constraints and transaction costs can determine scale and make the optimization more involved.

Tags

# Maximum Sharpe ratio and mean-variance optimization


# Maximum Sharpe ratio and mean-variance optimization












I want to understand why this holds: $argmax_w ( \frac{\mu^T w}{\sqrt{w^T\Sigma w}})=\Sigma^{-1}\mu $

I just found this post: Derivation of the tangency (maximum Sharpe Ratio) portfolio in Markowitz Portfolio Theory?

In the process you exchanged the optimization problem for the optimal tangency portfolio with the optimization problem for the mean-variance portfolio: $argmax_w (w^T\mu-\frac{1}{2}w^T\Sigma w )$

I want to understand why these optimization problems come to the same conclusion.

Thanks!

## Answer by Michael Isichenko (score 3, accepted)

https://quant.stackexchange.com/a/67871

Your first argmax is actually defined up to a constant multiplier: $argmax\left(\frac{\mu^Tw}{\sqrt{w^T\Sigma w}}\right)=\lambda\Sigma^{-1}\mu$, where $\lambda$ is an arbitrary portfolio size scale. In general, maximizing a scale-invariant ratio of the form $f(w)/g(w)$ can be done in conditional terms: $max(f)$ subject to $g=const$, or, using a Lagrange multiplier $\lambda$, $max(f-\lambda g)$. Formally, the optimality condition $\nabla(f/g)=0$ results in a collinearity of the gradients of $f$ and $g$ -- the same as in the conditional maximum with a Lagrange multiplier. The arbitrariness of the $\lambda$ scale disappears if there are constraints on the portfolio weights $w$ and/or transaction costs incurred when rebalancing the portfolio from its previous state. This involves a more complicated, but still convex optimization problem, c.f. this book.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.