Unless the changes in prices of securities followed a joint normal distribution (which was regarded as ruled out, because it would imply, as Samuelson had noted, a non-zero probability of negative prices), mean-variance analysis seemed to rest on a specific form of the "utility function" characterizing the relationship between the return, y, on an investor's portfolio and his or her preferences. Mean-variance analysis seemed to imply that investors' utility functions were quadratic: that is, they contained only terms my andy2. Special cases aside, a rational investor with a utility function whose mathematical form differed from this would not be guided by the expected mean and variance alone.

Donald MacKenzie, An Engine, Not a Camera, loc. 588