A note on minimum-variance models