Marginal Likelihood based Factor Selection in the International Setting

The relevance for the identi_cation of priced risk factors on the international level has increased tremendously in the last couple of decades. We revisit the recent work of Barillas and Shanken (2018) and Chib et al. (2018) who introduce a marginal likelihood based factor selection methodology. We argue that the specification of the Barillas and Shanken (2018) priors of the alpha's (across the candidate models) implies a prior bias towards sparse factor models, and find simulation results indicate that the factor selection methodology tends to favour sparser factor models, as opposed to the factor model implied by the simulated DGP, excessively. We find we can drastically increase the precision of the factor selection methodology by increasing the spreads of the priors of the alpha's, and find the precision of the methodology to be robust in a setting with student-t distributed factors. We apply the factor selection methodology, using priors for the alpha's with increased spreads, to select priced risk factors out of a set of prominent global factors as proposed in the literature, and find our selected factor model outperforms several prominent factor models proposed in the literature in terms of relative pricing performance.