MEAN-VARIANCE APPROXIMATIONS TO THE GEOMETRIC MEAN

This paper uses two databases to test the ability of six functions of arithmetic mean and variance to approximate geometric mean return or, equivalently, Bernoulli's expected log utility. The two databases are: (1) a database of returns on frequently used asset classes, and (2) that of real returns on the equity markets of sixteen countries, 1900–2000. Three of the functions of arithmetic mean and variance do quite well, even for return series with large losses. The other three do less well.