Deriving the term-structure of loan write-off risk under IFRS 9 by using survival analysis: A benchmark study

Battle Count: 52

This paper is primarily focused on credit risk modelling and regulatory compliance (IFRS 9) rather than quantitative trading. Its relevance to trading is indirect: accurate LGD/ECL estimation affects bank capital requirements and credit spreads, which can inform fixed-income trading strategies and credit derivative pricing. The survival analysis techniques could potentially be adapted for modelling default timing in corporate bond portfolios, but the paper's direct application is in banking risk management and loss provisioning.

The methodology involves multiple stages: data structuring into counting-process format, fitting discrete-time hazard GLMs, conditional inference survival trees, Tweedie compound Poisson GLMs, dichotomisation via Generalised Youden Index optimisation, and comprehensive benchmarking with time-dependent diagnostics. While individual components use standard R packages (glm, partykit, cplm), the overall pipeline requires careful data preparation, hyperparameter tuning, and understanding of survival analysis concepts. The open-source codebase significantly reduces implementation burden.

The paper provides an open-source R codebase maintained by Gabru et al. (2026) on Zenodo (DOI: 10.5281/zenodo.18982288). The methodology is described in tutorial-style detail with explicit R function calls (glm(), ctree(), cpglm()). However, the underlying mortgage data from a large South African bank is proprietary and not publicly available, limiting full end-to-end reproducibility. The appendix provides detailed data structure illustrations, variable descriptions, and hyperparameter settings.