On the Use of Intercepts as Performance Measures

Intercepts have long been used as measures of abnormal performance in finance. Jensen's alpha made its debut in 1968, but more recently researchers have used intercepts from the Fama-French Three-Factor model and others as ex post performance measures. A positive intercept is taken as evidence that the stock (or portfolio) exceeded expectations, and conversely for a negative intercept.We find that even conditional on the premise that the model used is valid, the intercept's ability to detect abnormal performance is critically dependent on accurately estimating the slope(s). Unfortunately, the same unsystematic perturbation to stock returns that causes abnormal performance typically distorts this estimate of the slope, in which case the intercept can become misleading, in some cases so misleading that it will move in the direction opposite that of the perturbation. This paper quantifies the distortion in the intercept due to errors in estimating the slope(s). It also identifies necessary and sufficient conditions for the intercept to be so misleading as to move in the direction opposite that of the shock, estimates the frequency of these conditions, and provides hypothetical and actual examples. Finally, we offer an alternative approach that alleviates the problem