Paper: arXiv 2404.17008
Authors: Arno Botha, Tanja Verster, Roelinde Bester
Abstract
A novel procedure is presented for finding the true but latent endpoints within the repayment histories of individual loans. The monthly observations beyond these true endpoints are false, largely due to operational failures that delay account closure, thereby corrupting some loans. Detecting these false observations is difficult at scale since each affected loan history might have a different sequence of trailing zero (or very small) month-end balances. Identifying these trailing balances requires an exact definition of a “small balance”, which our method informs. We demonstrate this procedure and isolate the ideal small-balance definition using two different South African datasets. Evidently, corrupted loans are remarkably prevalent and have excess histories that are surprisingly long, which ruin the timing of risk events and compromise any subsequent time-to-event model, e.g., survival analysis. Having discarded these excess histories, we demonstrably improve the accuracy of both the predicted timing and severity of risk events, without materially impacting the portfolio. The resulting estimates of credit losses are lower and less biased, which augurs well for raising accurate credit impairments under IFRS 9. Our work therefore addresses a pernicious data error, which highlights the pivotal role of data preparation in producing credible forecasts of credit risk.
Complexity vs Empirical Score
- Math Complexity: 3.0/10
- Empirical Rigor: 8.5/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper is methodologically grounded with statistical references but relies on a relatively simple optimization procedure rather than dense mathematics, while demonstrating the procedure on real-world datasets and quantifying improvements in risk predictions.
Research Flowchart
flowchart TD A["Research Goal:<br>Identify true repayment endpoints<br>and trailing zero balances"] --> B["Data Source:<br>Two South African<br>credit datasets"] B --> C["Key Methodology:<br>TruEnd-procedure<br>Define 'small balance' threshold"] C --> D["Computational Process:<br>Detect & remove false<br>trailing zero observations"] D --> E["Outcome:<br>Cleaned loan histories"] E --> F["Analysis:<br>Survival modeling on<br>corrected time-to-event data"] F --> G["Key Findings:<br>1. Reduced credit loss estimates<br>2. Less biased IFRS 9 impairments<br>3. Improved risk timing accuracy"]
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 the TruEnd-procedure, an optimisation framework to identify true loan endpoints by defining a small‑balance threshold.
Source sentence
Accordingly, we contribute both such an 𝑓and a broader optimisation framework that systematically evaluates the costs/benefits of any small-balance definition in isolating the latent TZB-period. This novel framework is called the TruEnd-procedure, which finds the best small-balance definition
Assumes 3
- A weight w∈[0,1] must be assigned to down‑scale the non‑TZB mean when applying the procedure. model
Source sentence
In applying the TruEnd-procedure on a loan portfolio of size 𝑁, a value must first be assigned to the weight 𝑤∈[0, 1] in Eq. 7 for down‑scaling the non‑TZB mean 𝑀2
- Monthly account balances are used as the control variable for screening candidate thresholds. data
Source sentence
Regarding limitations, the TruEnd-procedure currently uses monthly account balances 𝑩 as the control variable in screening candidate thresholds 𝑏
- Trailing zero‑valued balances are false observations caused by operational failures that delay account closure. data
Source sentence
The monthly observations beyond these true endpoints are false, largely due to operational failures that delay account closure
Data
- Mortgage dataset alternative commercial
Source sentence
Mortgage accounts (𝑁= 653, 331) were observed from January 2007 up to December 2022
- Credit‑card dataset alternative commercial
Source sentence
credit card accounts (𝑁= 297, 164) were observed from May 2007 up to May 2023
Costs
Not applicable
Finds 5
- The optimal small‑balance threshold is ZAR 300 for mortgages and ZAR 400 for credit cards. statistical
Source sentence
the optimal 𝑏∗-value is ZAR 300 for the mortgage dataset, and ZAR 400 for the credit card dataset
- TZB cases affect about 23 % of mortgage accounts and can extend the false history by ≥18 months, with a mean TZB balance of ZAR 7.45. statistical
Source sentence
TZB‑cases are not only quite prevalent in affecting 23% of mortgage accounts,
- Applying the optimal b* reduces the average account age by 3 months (to 97.9 months) and by about 18 months for terminated accounts. performance
Source sentence
The overall mean age reduces by 3 months to 97.9 when applying this 𝑏∗-policy, even though its effect is much more pronounced for terminated accounts – in which case, the mean age reduces by about 18 months
- The mean realised loss rate falls by 1.2 percentage points to 40.5 % after discarding TZB periods. performance
Source sentence
The mean loss rate therefore decreases by 1.2% points to 40.5% when applying this 𝑏∗-policy
- Discarding excess TZB histories improves the accuracy of predicted timing and severity of risk events without materially impacting the portfolio. predictive-accuracy
Source sentence
Having discarded these excess histories, we demonstrably improve the accuracy of both the predicted timing and severity of risk events, without materially impacting the portfolio
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 28 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper 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 (44)
- 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
- The loss optimisation of loan recovery decision times using forecast cash flows (2022) — link
- The TruEnd-procedure [source code] for treating trailing zero-valued balances in credit data (2024) — link
- Survival analysis: a self-learning text (2012) — link
- The elements of statistical learning: Data mining, inference, and prediction (2009) — link
- An introduction to statistical learning: with applications in R (2013) — link
- Credit Risk Analytics: Measurement Techniques, Applications, and Examples in SAS (2016)
and 36 more
- The Management of Consumer Credit: Theory and Practice (2010)
- Consumer Credit Models: Pricing, Profit and Portfolios (2009) — link
- Convex optimization (2004)
- 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
- PS6/23: Model risk management principles for banks (2023) — link
- SS1/23: Model risk management principles for banks (2023) — link
- G9/2022: Matters related to the credit risk models of banks (2022) — link
- It’s about time: Using discrete-time survival analysis to study duration and the timing of events (1993)
- Survival analysis and interpretation of time-to-event data: the tortoise and the hare (2018) — link
- Nonparametric Estimation from Incomplete Observations (1958)
- Survival analysis (2016) — link
- Time to default in credit scoring using survival analysis: a benchmark study (2017) — link
- Survival analysis and the credit granting decision (1992)
- Not if but when will borrowers default (1999) — link
- Credit risk term-structures for Lifetime Impairment Forecasting: A practical guide (2017) — link
- The interaction of the IFRS 9 expected loss approach with supervisory rules and implications for financial stability (2016) — link
- What do we know about Loss Given Default? (2004) — link
- Bank loan recovery rates: Measuring and nonparametric density estimation (2010) — link
- Improvements in loss given default forecasts for bank loans (2013) — link
- Survival analysis methods for personal loan data (2002) — link
- A proposed best practice model validation framework for Banks (2017)
- Forecasting and stress testing credit card default using dynamic models (2013)
- Time Varying and dynamic models for default risk in consumer loans (2010)
- Credit scoring with macroeconomic variables using survival analysis (2009) — link
- Retail credit stress testing using a discrete hazard model with macroeconomic factors (2014) — link
- Addressing probationary period within a competing risks survival model for retail mortgage loss given default (2017) — link
- Comparisons of linear regression and survival analysis using single and mixture distributions approaches in modelling LGD (2012) — link
- Modelling the Time to Write-Off of Non-Performing Loans Using a Promotion Time Cure Model with Parametric Frailty (2023) — link
- Making use of survival analysis to indirectly model loss given default (2018) — link
- Estimation of failure probabilities in the presence of competing risks: new representations of old estimators (1999) — link
- A survey of methods for time series change point detection (2017) — link
- Computer age statistical inference (2016)
- Survival Analysis in Credit Risk Modeling (2018)
- Climate stress testing for mortgage default probability (2024) — link
0 of 44 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.