Loan loss elasticity benchmark for current expected credit loss and expected credit loss

Abstract

The benchmarking of current expected credit loss (CECL) estimates presents a persistent challenge for financial institutions, regulators, and auditors alike: one that is particularly acute for smaller institutions whose limited modelling capabilities place them at a material disadvantage relative to the multi-variable approaches employed by larger banks. Models involving extensive variables can be complex and difficult to validate, and a transparent, replicable benchmark would provide the institutions, supervisors, and auditors with a reliable instrument for detecting over or under-provisioning. In order to address this problem, this paper proposes and validates a credit loss elasticity model grounded in a Cobb-Douglas specification, in which current credit losses are modelled as a function of outstanding loan balances raised to an estimated elasticity coefficient. A log-linear regression framework is applied to a panel dataset of US banks with assets exceeding US$300m, with robustness tests conducted across bank size, charter type, geographic region, and time period, and with macroeconomic variables — gross domestic product growth, unemployment rate, and real estate index — incorporated to assess the model’s stability under varying conditions. The central finding of this analysis is that loan loss elasticity is stable over time, including during the inaugural year of CECL implementation in 2023, and that the model performs consistently across bank segments and timeframes. This stability yields several important implications. The elasticity model introduces a novel benchmarking instrument that is accessible to institutions with limited modelling resources, equipping auditors and regulators with a practical means of identifying deviations from normative provisioning behaviour. The model serves solely as a benchmarking tool; it cannot substitute for a compliant CECL or ECL estimation model, as it does not capture the portfolio-specific risk characteristics required by regulatory standards. By providing a transparent, replicable external reference point against which institutional CECL and ECL estimates may be evaluated, the model supports the broader goals of financial system transparency and comparability. There are, however, acknowledged limitations: the model does not differentiate between loan vintages or credit quality, and negative credit loss entries are excluded owing to concerns about accrual manipulation. The implications of this exclusion for longitudinal benchmarking across credit cycles are discussed in the limitations section. This paper concludes that no single modelling approach can resolve the full complexity of CECL estimation, but that the elasticity model offers institutions a transparent, replicable, and theoretically grounded tool that meaningfully improves comparability across institutions; and that its simplicity is, in this context, a feature rather than a limitation. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.