Paper: arXiv 2610.05926
Authors: Arno Botha, Henko Crewe, Marcel Muller, Janette Larney
Abstract
The use of run-off triangles (ROTs) is a common industry practice in estimating the loss given default (LGD) risk parameter when predicting credit losses in banking. We benchmark this industry practice using credit card data against a more sophisticated (though classical) regression-based approach, which is able to leverage various types of input variables in producing loan-level LGD-estimates. This regression-based approach can demonstrably recover the typical characteristics of the ‘U-shaped’ empirical LGD-distribution, which the ROT-based approach cannot do. First, we critically review the ROT-based approach and identify multiple demerits using data-driven diagnostics. We then estimate a two-stage regression-based LGD-model and favourably assess the model performance of each component (or ‘stage’). Finally, we aggregate the LGD-estimates produced by each approach over time, and compare each time series to the mean empirical loss rate over time. The ROT-based aggregates diverge substantially from the empirical rate over most time periods, whilst the regression-based aggregates follow the empirical trends much closer. These results underscore the greater prediction accuracy of the regression-based LGD-model, relative to the ROT-based one. By implication, the former approach is probably better than the latter ROT-based approach when estimating the LGD under the IFRS 9 accounting framework, which prioritises accuracy.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper demonstrates strong empirical rigor through its comparative analysis of LGD models using credit card data. It also employs a reasonable level of mathematical sophistication with regression-based approaches and GLMs. The novelty comes from the direct comparison of industry-standard ROTs with more advanced regression techniques for LGD estimation.
Research Flowchart
flowchart TD
A["Research Goal: Compare ROT vs Regression for Credit Card LGD Modeling"] --> B{"Key Methodologies"};
B --> C["Run-Off Triangles (ROTs) Approach"];
B --> D["Two-Stage Regression-Based LGD Model"];
C --> E["Data Input: Credit Card Default Data"];
D --> E;
E --> F["Computational Processes: Model Estimation & Aggregation"];
F --> G{"Key Findings/Outcomes"};
G --> H["ROT: Cannot recover U-shaped LGD, aggregates diverge from empirical rate"];
G --> I["Regression: Recovers U-shaped LGD, aggregates follow empirical trends, higher prediction accuracy, better for IFRS 9"];
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes two variants of a run‑off triangle (ROT) based LGD model and a classical two‑stage regression‑based LGD model for credit‑card data.
Source sentence
We propose two variants of a ROT-based LGD-model to serve as our baseline.
Assumes 5
- The loss severity component follows a Gaussian distribution with constant variance. model
Source sentence
We assume that the response variable follows a Gaussian distribution with constant variance 𝜎2, i.e., 𝑌𝑖∼𝑁(𝜇𝑖, 𝜎2).
- Development factors are stable across default cohorts and over time. model
Source sentence
The ROT-approach relies on the assumption that development factors are stable across default cohorts.
- A VIF‑threshold of 10 is used for variable selection. statistical
Source sentence
Accordingly, and given the relatively large size of our own dataset, we adopt a VIF‑threshold of 10 based on industry experience, and deselect variables accordingly.
- The logistic link function is used for the write‑off risk (binary) model. model
Source sentence
In this logit model, the inverse link function 𝑔−1 is the logistic (sigmoid) function (1 + 𝑒−𝑥)−1 of a given 𝑥.
Show 1 more
- The dichotomisation step maximises the Generalised Youden Index to obtain an optimal cut‑off. model
Source sentence
This dichotomisation step substantially improves the ability of the models to recover the empirical LGD-distribution, as will be shown later. Moreover, this dichotomisation step is based on maximising the Generalised Youden Index 𝐽𝑎(𝑐) across possible cut‑off values 𝑐∈[0, 1].
Data
- credit card portfolio data other unknown
Source sentence
We benchmark this industry practice using credit card data against a more sophisticated (though classical) regression‑based approach.
Costs
Not applicable
Finds 3
- ROT‑based aggregates diverge substantially from the empirical loss rate, while regression‑based aggregates closely follow empirical trends. predictive-accuracy
Source sentence
The ROT-based aggregates diverge substantially from the empirical rate over most time periods, whilst the regression-based aggregates follow the empirical trends much closer.
- The grand MAE for the ROT‑based LGD estimates is 64.37%, indicating large prediction error. performance
Source sentence
The grand MAE of 64.37% suggests that the ROT-based LGD-estimates 𝐺𝑣,𝑚 systematically fail to approximate the realised loss rates across all (𝑣, 𝑚), with substantial average errors.
- The regression‑based two‑stage model can recover the typical U‑shaped empirical LGD distribution, which the ROT approach cannot. predictive-accuracy
Source sentence
This regression‑based approach can demonstrably recover the typical characteristics of the 'U‑shaped' empirical LGD‑distribution, which the ROT‑based approach cannot do.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 19 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (43)
- A cost of capital approach to determining the LGD discount rate — → 2503.23992
- A procedure for loss-optimising the timing of loan recovery under uncertainty (2021) — link
- Simulation-based optimisation of the timing of loan recovery across different portfolios (2021) — link
- Defining and comparing SICR-events for classifying impaired loans under IFRS 9 — → 2303.03080
- Deriving the term-structure of loan write-off risk under IFRS 9 by using survival analysis: A benchmark study — → 2603.11897
- The Management of Consumer Credit: Theory and Practice (2010)
- Credit Risk Management: Basic Concepts (2009) — link
- Credit Risk Analytics: Measurement Techniques, Applications, and Examples in SAS (2016)
and 35 more
- Foundations of linear and generalized linear models (2015)
- ROC Curves for Continuous Data (2009)
- Information Theory and an Extension of the Maximum Likelihood Principle (1998) — link
- Applied Logistic Regression (2000) — link
- Regression Diagnostics: Identifying Influential Data and Sources of Collinearity (1980) — link
- International Convergence of Capital Measurement and Capital Standards: A Revised Framework, Comprehensive Version (2006) — link
- The Basel Framework (2019) — link
- International Financial Reporting Standard ( IFRS ) 9: Financial Instruments (2014) — link
- Methods and models of loss reserving based on run-off triangles: A unifying survey (2006)
- Non-optimal prediction by the chain ladder method (1997) — link
- The actuary and IBNR (1972)
- The Stanard-Bühlmann reserving procedure: A practitioner’s guide (2002)
- Time Varying and dynamic models for default risk in consumer loans (2010)
- What do we know about loss given default? (2004)
- Bank loan recovery rates: Measuring and nonparametric density estimation (2010) — link
- Benchmarking regression algorithms for loss given default modeling (2012) — link
- Predicting loss given default ( LGD ) for residential mortgage loans: A two-stage model and empirical evidence for UK bank data (2012) — link
- Improvements in loss given default forecasts for bank loans (2013) — link
- Incurred but not reported claim reserves (1934)
- Distribution-free calculation of the standard error of chain ladder reserve estimates (1993)
- A robustification of the chain-ladder method (2009)
- The Critical Model Parameter: LGD (2013)
- Credit risk: LGD estimation based on chain-ladder method (2015)
- Default weighted survival analysis to directly model loss given default (2018)
- Development of an impairment point in time probability of default model for revolving retail credit products: South African case study (2021) — link
- A General Qualitative Definition of Robustness (1971) — link
- A method for modelling varying run-off evolutions in claims reserving (1994)
- Variable selection–a review and recommendations for the practicing statistician (2018) — link
- An introduction to ROC analysis (2006) — link
- The impact of pre-selected variance inflation factor thresholds on the stability and predictive power of logistic regression models in credit scoring (2015) — link
- Explaining prediction models and individual predictions with feature contributions (2014) — link
- Divergence measures based on the Shannon entropy (1991) — link
- Comparing two tests used for diagnostic or screening purposes (1998) — link
- Maximization of the sum of sensitivity and specificity as a diagnostic cutpoint criterion (2008) — link
- Youden Index and the optimal threshold for markers with mass at zero (2008) — link
3 of 43 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2601 linked papers.