Paper: arXiv 2601.20533

Authors: Jianwei Peng, Stefan Lessmann

Abstract

Survival analysis has become a standard approach for modelling time to default by time-varying covariates in credit risk. Unlike most existing methods that implicitly assume a stationary data-generating process, in practise, mortgage portfolios are exposed to various forms of data drift caused by changing borrower behaviour, macroeconomic conditions, policy regimes and so on. This study investigates the impact of data drift on survival-based credit risk models and proposes a dynamic joint modelling framework to improve robustness under non-stationary environments. The proposed model integrates a longitudinal behavioural marker derived from balance dynamics with a discrete-time hazard formulation, combined with landmark one-hot encoding and isotonic calibration. Three types of data drift (sudden, incremental and recurring) are simulated and analysed on mortgage loan datasets from Freddie Mac. Experiments and corresponding evidence show that the proposed landmark-based joint model consistently outperforms classical survival models, tree-based drift-adaptive learners and gradient boosting methods in terms of discrimination and calibration across all drift scenarios, which confirms the superiority of our model design.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper combines longitudinal behavioral joint modeling with discrete-time hazard formulations and calibration to tackle non-stationary data drift in retail credit risk. It offers an exceptionally thorough empirical framework evaluating sudden, incremental, and recurring drift scenarios on large-scale Freddie Mac loan data.

Research Flowchart

  flowchart TD
    A[Goal: Improve Survival Analysis for Credit Risk under Data Drift] --> B[(Freddie Mac Mortgage Datasets)]
    B --> C[Simulate Data Drift:<br>Sudden, Incremental, Recurring]
    C --> D[Dynamic Joint Modelling Framework:<br>Longitudinal Markers + Discrete-Time Hazard]
    D --> E[Model Refinement:<br>Landmark One-Hot Encoding + Isotonic Calibration]
    E --> F[Performance Evaluation vs.<br>Classical, Tree, & Boosting Models]
    F --> G[Outcome: Landmark Joint Model Outperforms<br>in Discrimination & Calibration Across All Drift]

Cited References (38)

  1. Machine learning for credit scoring and loan default prediction using behavioral and transactional financial data (2025) — link
  2. Survival analysis methods for personal loan data (2002)
  3. Prognosis research strategy ( PROGRESS ) 3: prognostic model research (2013)
  4. A survey on machine learning for recurring concept drifting data streams (2023)
  5. Default risk and income fluctuations in emerging economies (2008)
  6. Learning drifting concepts: Example selection vs. example weighting (2004)
  7. A survey on concept drift adaptation (2014)
  8. Forecasting and stress testing credit card default using dynamic models (2013)
and 30 more
  1. Time to default in credit scoring using survival analysis: a benchmark study (2017)
  2. Mitigating concept drift in data streams: an incremental decision tree approach (2024) — link
  3. Survival analysis and competing risk models of hospital length of stay and discharge destination: the effect of distributional assumptions (2007)
  4. Improved insights on financial health through partially constrained hidden Markov model clustering on loan repayment data (2018)
  5. An Akaike information criterion for multiple event mixture cure models (2015)
  6. Macro-economic factors in credit risk calculations: including time-varying covariates in mixture cure models (2019)
  7. Dynamic survival models with varying coefficients for credit risks. (2019)
  8. Multihorizon discrete time survival models (2022)
  9. Risk management using model predictive control (2015-08-31) — link
  10. Approaches for modelling the term-structure of default risk under IFRS 9: A tutorial using discrete-time survival analysis (2025) — link
  11. Combining dynamic predictions from joint models for longitudinal and time-to-event data using Bayesian model averaging (2014)
  12. Reflection on modern methods: dynamic prediction using joint models of longitudinal and time-to-event data (2021)
  13. Dynamic prediction of an event using multiple longitudinal markers: a model averaging approach (2025)
  14. Joint models for longitudinal and discrete survival data in credit scoring (2023)
  15. Pencal : an R package for the dynamic prediction of survival with many longitudinal predictors (2025) — link
  16. One or two things we know about concept drift—a survey on monitoring in evolving environments. Part B : locating and explaining concept drift (2024)
  17. CORTEX : composite overlay for risk tiering and exposure in operational AI systems (2025) — link
  18. An overview of concept drift applications (2016) — link
  19. Credit scoring with drift adaptation using local regions of competence (2022)
  20. Adaptive random forests for evolving data stream classification (2017)
  21. Learning regularized Hoeffding trees from data streams (2019) — link
  22. Calibrating machine learning approaches for probability estimation: A comprehensive comparison (2023)
  23. Where are we with calibration under dataset shift in image classification? (2025) — link
  24. Dynamic survival prediction combining landmarking with a machine learning ensemble: Methodology and empirical comparison (2021)
  25. The Cox model (2000)
  26. XGBoost : A scalable tree boosting system (2016) — link
  27. Adaptive learning from evolving data streams (2009)
  28. Credit scoring with macroeconomic variables using survival analysis (2009)
  29. Tackling data and model drift in AI : Strategies for maintaining accuracy during ML model inference (2023)
  30. Longitudinal data analysis (2006)

0 of 38 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.