Center for Risk Management Research

Theory predicts that aggregate volatility ought to be a priced risk factor. In an influential study with more than 1000 citations on Google Scholar, Ang, Hodrick, Xing, and Zhang (2006) propose an ex post factor, FVIX, intended as a proxy for aggregate volatility risk. Their test validating FVIX relies on an OLS regression

of portfolio excess returns on FVIX and other independent variables over the data period February 1986 - January 2001. October 1987 is an outlier, in which FVIX exhibits a 26-sigma deviation. The inclusion of this outlier results in a reduction of the regression standard error by more than a factor of two, creating the appearance of statistical signicance when none is present. We explain how standard statistics

can be used to assess the suitability of a dataset for OLS regression.