IFRS quantitative finance series
IFRS 9 ECL Backtesting: What Banks Should Test
A single comparison of total provision with realised write-offs cannot validate an IFRS 9 framework. Losses emerge over different horizons, portfolios change, recoveries arrive late and reporting estimates incorporate information that was available at each historical date.
Test components and decisions
Backtesting should cover PD calibration, discriminatory power, LGD amount and timing, EAD drawdowns, staging entry, cures, scenarios and overlays. Component tests explain why total ECL differs from outcome and reduce the chance that offsetting errors create a misleading pass.
Use vintage-consistent evidence
A fair test freezes the model, data and information set available at the reporting date, then follows exposures through the relevant horizon. Outcome windows must match twelve-month or lifetime objectives and account for incomplete recovery cycles.
Translate findings into action
Threshold breaches should lead to diagnosis, not automatic recalibration. The response may be monitoring, data correction, segmentation change, parameter redevelopment or temporary adjustment. Validation conclusions should be connected to financial-reporting materiality.
Quantitative expression
Implementation controls
- Freeze historical vintages and scenario inputs.
- Separate calibration, ranking and stability tests.
- Track remediation to closure.
- Escalate material bias to accounting governance.