Model risk management: Valuation and governance of pseudo-models

Abstract

The standard estimation approaches and their implementation generally assume well-specified models. What is feasible and unfeasible when the models are misspecified is discussed. An adjustment method is introduced for forecasts based on misspecified models. The key roles of the object of interest and of the management of validation samples are highlighted. The approach is used to derive forecast intervals robust to misspecification, to quantify the required capital for model risk in prudential supervision, to measure treatment effects with pseudo-models, or to jointly manage a set of pseudo-models.