Core Path · L6.3.07

Survival models for lifetime PD

14 minHazardKaplan-MeierCensoringCompeting risksDiscrete-time hazardLifetime PD

By the end you will be able to

  1. 1Estimate monthly default hazards and a Kaplan-Meier survival curve with a confidence band from loans watched for different lengths of time (M-D13: Survival analysis).
  2. 2Treat early repayment as a competing risk and explain why the cumulative incidence of default is lower than one minus the Kaplan-Meier curve.
  3. 3Fit a discrete-time hazard model by month on book, score band and vintage, and judge what it can and cannot say about the economy.
  4. 4Turn monthly hazards into lifetime, marginal and conditional PDs for an IFRS 9 calculation (M-C14: Survival and marginal default probabilities).

Why it matters

L3.3.14: Retail credit portfolios gave each score band a one-year PD and each vintage a booked default curve. A lifetime provision needs more: the chance that a loan written today defaults in its second or third year. The obvious count misleads. Of the 31,643 personal loans written since 2023Q1, 1,383 have defaulted, 4.37%. But most were written too recently to have finished their story, and by month 36 nearly a third repay early, which ends their chance to default. Done properly, the 36-month PD is 7.15% if no one could repay early and 6.29% allowing for those who do.