Efficient frontier
The set of portfolios each of which offers the highest expected return for its own level of dispersion. Anything below the line loses on both dimensions at once.
How to read the number
The point of the exercise is not the specific weights but the finding that adding an asset weakly related to the rest pushes the frontier upward, even when that asset looks dull on its own.
When the metric lies
The frontier is built on expected returns, and those are taken from the past. An error in the inputs is passed into the weights with amplification: the optimiser loads up on whatever asset happened to be estimated highest.
Also known as: mean variance frontier