2014-05: In Search of Statistically Valid Risk Factors

An influential observation has the potential to render a model unsuitable for estimation with an OLS regression. This is well known in the statistics literature; see, for example Chatterjee and Hadi (1986). However, the use of standard measures for detecting and managing influential observations is not common practice in empirical finance. In this article, we examine a leverage point, which is an outlier in an independent variable, from an economic perspective since it represents an actual market event. It is also interesting and subtle from a statistical perspective: it materially deflates the volume of regression confidence ellipsoids but has a limited effect on the regression coefficients, thereby enabling pure noise to masquerade as statistical significance.