Paper: arXiv 2502.14479
Authors: Arno Botha, Tanja Verster, Roland Breedt
Abstract
The lifetime behaviour of loans is notoriously difficult to model, which can compromise a bank’s financial reserves against future losses, if modelled poorly. Therefore, we present a data-driven comparative study amongst three techniques in modelling a series of default risk estimates over the lifetime of each loan, i.e., its term-structure. The behaviour of loans can be described using a nonstationary and time-dependent semi-Markov model, though we model its elements using a multistate regression-based approach. As such, the transition probabilities are explicitly modelled as a function of a rich set of input variables, including macroeconomic and loan-level inputs. Our modelling techniques are deliberately chosen in ascending order of complexity: 1) a Markov chain; 2) beta regression; and 3) multinomial logistic regression. Using residential mortgage data, our results show that each successive model outperforms the previous, likely as a result of greater sophistication. This finding required devising a novel suite of simple model diagnostics, which can itself be reused in assessing sampling representativeness and the performance of other modelling techniques. These contributions surely advance the current practice within banking when conducting multistate modelling. Consequently, we believe that the estimation of loss reserves will be more timeous and accurate under IFRS 9.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced statistical techniques (beta regression, multinomial logistic regression) within a multistate framework, indicating high mathematical density. It is strongly data-driven, using residential mortgage data and evaluating model performance, which aligns with high empirical rigor.
Research Flowchart
flowchart TD
A["Research Goal"] --> B["Data & Inputs"]
B --> C["Model Development"]
C --> D["Performance Evaluation"]
D --> E["Key Findings"]
subgraph A ["Research Goal"]
A1["Model loan lifetime<br>default risk term-structure<br>under IFRS 9"]
end
subgraph B ["Data & Inputs"]
B1["Residential Mortgage Data"]
B2["Macroeconomic Variables"]
B3["Loan-Level Features"]
end
subgraph C ["Methodology"]
C1["Markov Chain"]
C2["Beta Regression"]
C3["Multinomial Logistic Regression"]
end
subgraph D ["Computations"]
D1["Custom Diagnostics Suite"]
D2["Model Comparison"]
end
subgraph E ["Outcomes"]
E1["Successive Models Outperform Previous"]
E2["Enhanced IFRS 9 Loss Reserve Accuracy"]
end
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 a data‑driven comparative study of three modelling techniques for loan default risk term‑structures, introducing novel diagnostics for model assessment.
Source sentence
we present a data-driven comparative study amongst three techniques in modelling a series of default risk estimates over the lifetime of each loan, i.e., its term-structure.
Assumes 2
- The baseline Markov chain is assumed to be time‑homogeneous and stationary. model
Source sentence
a simple time-homogeneous and stationary Markov chain was fit as a baseline transition-type model
- Transition probabilities can be modelled as functions of macroeconomic and loan‑level input variables. model
Source sentence
the transition probabilities are explicitly modelled as a function of a rich set of input variables, including macroeconomic and loan-level inputs
Data
- Residential mortgage portfolio alternative proprietary
Source sentence
We conduct our comparative study using a data-rich portfolio of residential mortgages, as provided by a large South African bank.
Costs
Not applicable
Finds 4
- The multinomial logistic regression (MLR) model achieves the highest predictive accuracy, improving the AD‑statistic by 64.1% on average over the Markov chain. predictive-accuracy
Source sentence
the MLR-model improved it even more by 64.1% on average.
- The beta regression (BR) model also outperforms the Markov chain, improving the AD‑statistic by 58.8% on average. predictive-accuracy
Source sentence
the BR-model improved the AD-statistic on average by 58.8%
- Across all transition types, the MLR models have the smallest average discrepancy (AD) values, indicating closest agreement with actual rates. predictive-accuracy
Source sentence
the MLR-models agree the closest with reality since they have the smallest ¯𝑟AD(𝑟1,𝑟2)-values
- Each successive modelling technique (Markov → BR → MLR) outperforms the previous one. performance
Source sentence
our results show that each successive model outperforms the previous
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 35 related papers). Coverage is partial, so this is neither replication nor refutation.
2 papers in the archive
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 (69)
- Defining and comparing SICR-events for classifying impaired loans under IFRS 9 — → 2303.03080
- 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
- Modelling the term-structure of default risk under IFRS 9 within a multistate regression framework [Source Code] (2025) — link
- Credit risk scorecards: developing and implementing intelligent credit scoring (2005)
- Consumer Credit Models: Pricing, Profit and Portfolios (2009) — link
- Credit risk analytics: Measurement techniques, applications, and examples in SAS (2016)
- Credit scoring and its applications (2002) — link
and 61 more
- Credit Risk Management: Basic Concepts (2009) — link
- Markov Chains (1997) — link
- Applied Logistic Regression (2000) — link
- An introduction to categorical data analysis (2007) — link
- An introduction to statistical learning: with applications in R (2013) — 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
- Regulation (EU) No 575/2013 of the European Parliament and of the Council of 26 June 2013 on prudential requirements for credit institutions and investment firms and amending Regulation (EU) No 648/2012 Text with EEA relevance (2013) — link
- PS6/23: Model risk management principles for banks (2023) — link
- SS1/23: Model risk management principles for banks (2023) — link
- Statistical classification methods in consumer credit scoring: a review (1997) — link
- Review of the literature on credit risk modeling: development of the past 10 years (2010) — link
- Classification methods applied to credit scoring: Systematic review and overall comparison (2016)
- Time Varying and dynamic models for default risk in consumer loans (2010)
- Impairment of Financial Instruments under IFRS 9 (2018) — link
- IFRS 9 : Expected Credit Losses (2014) — link
- Credit risk term-structures for Lifetime Impairment Forecasting: A practical guide (2017) — link
- Estimating Lifetime Expected Credit Losses Under IFRS 9 (2016) — link
- It’s about time: Using discrete-time survival analysis to study duration and the timing of events (1993)
- It’s d 'e j `a vu all over again: Using multiple-spell discrete-time survival analysis (1995) — link
- Estimation of the allowance for doubtful accounts by Markov chains (1962) — link
- The use of exponentially-smoothed transition matrices to improve forecasting of cash flows from accounts receivable (1978) — link
- Note—On the Cyert-Davidson-Thompson Doubtful Accounts Model (1981) — link
- Testing the adequacy of Markov chain and mover-stayer models as representations of credit behavior (1985) — link
- Credit rating dynamics and Markov mixture models (2008) — link
- Forecasting losses on a liquidating long-term loan portfolio (1995) — link
- Estimating Loss Given Default Based on Beta Regression (2020) — link
- Generalized beta regression models for random loss-given-default (2011) — link
- Loss Given Default Modeling: A Comparative Analysis (2013) — link
- A zero-adjusted gamma model for mortgage loan loss given default (2013)
- Markov chain models for delinquency: Transition matrix estimation and forecasting (2011) — link
- A credit risk model for large dimensional portfolios with application to economic capital (2006) — link
- Markov regenerative credit rating model (2017) — link
- The predictive accuracy of Sukuk ratings; Multinomial Logistic and Neural Network inferences (2015) — link
- Multinomial logistic regression and spline regression for credit risk modelling (2018) — link
- A transitions-based framework for estimating expected credit losses (2014) — link
- The good, the bad and the impaired: A credit risk model of the Irish mortgage market (2016) — link
- Bank loan recovery rates: Measuring and nonparametric density estimation (2010) — link
- The probability approach to default probabilities (2007) — link
- Tutorial in biostatistics: competing risks and multi-state models (2007) — link
- Statistical inference about Markov chains (1957) — link
- Calibration of PD term structures: To be Markov or not to be (2007) — link
- Beta regression for modelling rates and proportions (2004) — [link](https://doi.org/10. 1080/0266476042000214501)
- Improved estimators for a general class of beta regression models (2010) — link
- Incorporating heterogeneity and macroeconomic variables into multi-state delinquency models for credit cards (2018) — link
- Intensity models and transition probabilities for credit card loan delinquencies (2014) — link
- Too big to fail: large samples and the p-value problem (2013) — link
- Information Theory and an Extension of the Maximum Likelihood Principle (1998) — link
- Variable selection–a review and recommendations for the practicing statistician (2018) — link
- Measuring Skewness: A Forgotten Statistic? (2011) — link
- Beta regression in R (2010)
- Coefficients of determination for multiple logistic regression analysis (2000) — link
- Conditional logit analysis of qualitative choice behavior (1972) — link
- An introduction to ROC analysis (2006) — link
- A review of spline function procedures in R (2019) — link
- Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach (1988) — link
- Approaches for modelling the term-structure of default risk under IFRS 9: A tutorial using discrete-time survival analysis — → 2507.15441
- Unavoidable model risk in Expected Credit Loss models under IFRS 9 and CECL standards (2024) — link
- Improving credit risk assessment in uncertain times: Insights from IFRS 9 (2025) — link
- The Statistical Analysis of Compositional Data (1982) — link
2 of 69 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.