Statistical model infrastructures are often developed using a piecemeal approach to model building, in which different components are developed and validated separately. This type of modeling framework has significant limitations at each stage of the model management life cycle, from development and documentation to validation, production, and redevelopment. We propose an empirical framework, spurred by recent developments in the implementation of Generalized Structural Equation Modeling (GSEM), which brings to bear a modular and all-inclusive approach to statistical model building. We illustrate the “game changing” potential of this framework with an application to the stress testing of credit risk for a representative portfolio of mortgages; we also extend it to the analysis of the allowance for credit loss under the novel Current Expected Credit Loss (CECL) accounting regulation. We illustrate how GSEM techniques can significantly enhance every step of the modeling framework life cycle. We also illustrate how GSEM can be used to combine various risk management projects and tasks into a single framework; we specifically illustrate how to seamlessly integrate stress testing and CECL (or IFRS9) frameworks and champion, and challenger, modeling frameworks. Finally, we identify other areas of model risk management that can benefit from the GSEM framework and highlight other potentially fruitful applications of the methodology.
Can We Take the “Stress” Out of Stress Testing? Applications of Generalized Structural Equation Modeling to Consumer Finance
WP 21-01 – Financial firms, and banks in particular, rely heavily on complex suites of interrelated statistical models in their risk management and business reporting infrastructures.