Severity over quantity. Drivers of supervisory capital add-ons in internal ratings-based models

Banks use their internal models to estimate capital requirements in a risk-sensitive way, subject to a set of rules laid down in banking regulation. However, these models are not flawless as the usage of models suffers from imperfections, such as oversimplifications or wrong assumptions. As a result, risks may be underestimated. This is particularly troublesome, where models are used to assess risks to banks’ solvency. In this paper we address an important gap in the literature with regard to such model risk. We find that a small set of high severity deficiencies in models is responsible for the majority of the counterfactual RWA burden imposed by supervisory capital add-ons. We trace the underlying non-compliances to a subset of CRR articles that mostly govern the handling of IRB-relevant data by banks. Our results help improve the supervision of IRB-banks by proposing more efficient use of scarce supervisory resources. JEL Classification: G21, G28, G29