Approaches for modelling the term-structure of default risk under IFRS 9: A tutorial using discrete-time survival analysis

2025-07-21

AI Summary

Under IFRS 9, existing loan-level probability of default (PD) models often lack dynamic responsiveness to macroeconomic shifts throughout the loan lifecycle. To address this limitation, this paper proposes a discrete-time survival analysis framework for PD modeling. Methodologically, it innovatively integrates time-dependent covariate modeling, logistic regression, and systematic model diagnostics. A reusable, standardized workflow for data quality assessment and model validation is introduced, accompanied by an open-source R package. The resulting framework significantly enhances the time-varying sensitivity, transparency, and robustness of PD estimation—enabling timely expected credit loss (ECL) measurement. Moreover, the standardized modeling architecture provides financial institutions, model validators, and regulatory authorities with a technically rigorous yet operationally practical reference grounded in both statistical theory and real-world implementation requirements.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Probabilistic Circuits and Graphical Models
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Under the International Financial Reporting Standards (IFRS) 9, credit losses ought to be recognised timeously and accurately. This requirement belies a certain degree of dynamicity when estimating the constituent parts of a credit loss event, most notably the probability of default (PD). It is notoriously difficult to produce such PD-estimates at every point of loan life that are adequately dynamic and accurate, especially when considering the ever-changing macroeconomic background. In rendering these lifetime PD-estimates, the choice of modelling technique plays an important role, which is why we first review a few classes of techniques, including the merits and limitations of each. Our main contribution however is the development of an in-depth and data-driven tutorial using a particular class of techniques called discrete-time survival analysis. This tutorial is accompanied by a diverse set of reusable diagnostic measures for evaluating various aspects of a survival model and the underlying data. A comprehensive R-based codebase is further contributed. We believe that our work can help cultivate common modelling practices under IFRS 9, and should be valuable to practitioners, model validators, and regulators alike.

Problem

• Modeling dynamic default risk under IFRS 9 standards
• Estimating accurate lifetime probability of default (PD)
• Evaluating survival models with reusable diagnostic measures

Innovation

• Uses discrete-time survival analysis technique
• Provides reusable diagnostic measures
• Includes comprehensive R-based codebase