Mean-variance optimisation

The mathematical procedure that produces the efficient frontier from expected returns, volatilities and correlations.

Its documented failure mode is that it behaves as an error maximiser: the assets with the most overestimated returns receive the largest weights, so instead of diversifying estimation error away, it amplifies it.

Practical fixes include constraining weights, shrinking the estimates toward a common mean, and starting from market equilibrium and expressing views as deviations.

Its mathematics is correct. All of the trouble comes from the inputs.