Accounting research has extensively debated the modelling of Earnings Management (EM) for all industries but the baking industry. We provide a first analysis of the validity, strengths and weaknesses of existing loan loss provisioning (LLP) models and develop an extensive framework for modelling components. We rely on US data from 2000-2018 and apply prevalent test procedures that examine the extent of measurement errors, extreme performance, omitted-variable biases and predictive power of each of the models. The results indicate that established modelling can be optimized with regard to measurement errors, omitted-variable biases and predictive power. In particular, we find that including net charge offs is less important while a non-performing loan component is indispensable. In addition, our results reveal that LLP models are less prone to measurement errors related to one-step vs. two-step modelling, while more advanced estimation approaches are not linked to better prediction power.
Supplementary notes can be added here, including code and math.