Empirical excluded-factor model comparison and its complications

Abstract This paper replicates the work of Barillas and Shanken (2016) and extends upon it by applying an extensive pairwise comparison of seven prominent asset pricing models. This comparison shows some practical complications that arise with excluded-factor comparison. Firstly, the evidence indicates that it might be possible to influence the outcome of such comparisons significantly by adding weak factors. Moreover, the results of all pairwise comparisons combined are non-transitive, therefore the one-on-one comparisons are inconclusive. Furthermore, sampling variability is introduced through a moving block bootstrap method to examine the reliability of evidence based on excluded-factor regression alphas. Introducing sampling variation does not alter the significance or conclusion based on this excluded-factor evidence.