The Alphas of Asset Allocators

Abstract

The performance of asset allocation strategies is commonly evaluated using alpha. Alpha is a measure of selection, not timing, however, and thus has no basis for use in this manner. This article examines what happens when alpha is used to examine asset allocation strategies. The results show that alpha has remarkably good large sample properties. In an economic sense, it seems to identify ability and is not fooled by the absence of it, but in a statistical sense, it fails to identify ability. For the smaller samples encountered in practice, it can even be fooled into thinking that strategies using random-number generators have economic ability, though not statistical ability. In practice, most applications of alpha do not test for statistical significance, so there is a tremendous risk of false conclusions—particularly, finding ability where it does not exist. Alpha should therefore not be used to measure any form of timing ability. In doing so, the industry could, however unknowingly, be making itself look better than it really is.