Statistical Model Selection in Operational Risk

Abstract

While substantial amount of work in operational risk has been dedicated to fitting distributions to the loss data, comparatively little was written about model selection. Arguably, a good selection procedure is more important than fitting one since it can rectify a failure of fitting methodology by rejecting a bad model. This presentation applies the work of K.P. Burnham and D.R. Anderson, 2002 in the context of the operational risk capital modeling, including the use of information criteria and model averaging.