A Holistic Model Validation Framework for Current Expected Credit Loss (CECL) Model Development and Implementation

Abstract

The Current Expected Credit Loss (CECL) revised accounting standard for credit loss provisioning is the most important change to United States (US) accounting standards in recent history. In this study, we survey and assess practices in the validation of models that support CECL, across dimensions of both model development and model implementation. On the development side, this entails the usual SR 11-7 aspects of model validation; however, highlighted in the CECL context is the impact of several key modeling assumptions upon loan loss provisions. We also consider the validation of CECL model implementation or execution elements, which assumes heightened focus in CECL given the financial reporting implications. As an example of CECL model development validation, we investigate a modeling framework that we believe to be very close to that being contemplated by institutions, which projects loan losses using time-series econometric models, for an aggregated “average” bank using Federal Deposit Insurance Corporation (FDIC) Call Report data. In this example, we assess the accuracy of 14 alternative CECL modeling approaches, and we further quantify the level of model risk using the principle of relative entropy. Apart from the illustration of several model validation issues and practices that are of particular relevance to CECL, the empirical analysis has some potentially profound policy and model risk management implications. Specifically, implementation of the CECL standard may lead to under-prediction of credit losses; furthermore, coupled with the assumption that we are at an end to the favorable phase of the credit cycle, this may be interpreted as evidence that the goal of mitigating the procyclicality in the provisioning process that motivated CECL may fail to materialize.