In the mean-variance framework, stock returns are assumed to be normally distributed. But what does the historical record tell us? Why are portfolio managers wary of using mean and variance estimates over decades and prefer shorter time periods?