Minimum variance portfolio
The point on the efficient frontier with the smallest achievable dispersion: weights chosen so that the assets' movements cancel each other out as fully as possible.
Formula
\min_{w} \; w^{\top}\Sigma w \quad \text{subject to} \quad \sum_i w_i = 1Sigma is the covariance matrix of returns. Expected returns do not enter the problem at all, which is what makes it sturdier than the other optimisations.
How to read the number
It is the only point on the frontier that needs no return forecast — an estimate of joint movement is enough.
When the metric lies
The solution tends to collect in a handful of the quietest assets and turn into a concentrated bet. Without a cap on weights one name easily comes to dominate.
Also known as: min variance portfolio