IFRS 9 Model Validation Services for Banks

IFRS 9 model validation is the independent, evidence-based assessment of the Expected Credit Loss calculation chain required under IFRS 9 paragraphs 5.5 and B5.5. Unlike legacy IAS 39 incurred-loss validation, IFRS 9 validation is forward-looking by construction. It must confirm not only that historical calibration holds, but that the point-in-time PD term structure, lifetime LGD, exposure bucketing, SICR triggers, and probability-weighted macroeconomic overlays continue to produce ECL outputs that are unbiased, reproducible, and appropriate for the current and projected economic environment.

A complete IFRS 9 validation engagement covers six layers: conceptual soundness, input data quality, model implementation, outcomes analysis, ongoing monitoring, and governance documentation. These six layers are aligned with the Basel Committee on Banking Supervision paper BCBS 239 on risk data aggregation, the Federal Reserve SR 11-7 principles on model risk management, and the GPPC Credit Risk Modelling Group guidance on ECL implementation. A defensible IFRS 9 programme also addresses FASB CECL convergence points for institutions with dual reporting obligations.

The Probability of Default term structure is the anchor of any IFRS 9 ECL model. FineIT validates PD by reconciling the point-in-time estimate used for stage one and stage two ECL against the through-the-cycle master scale via a Vasicek single-factor transformation, using the institution's own systemic factor loading. Calibration is tested with binomial and Hosmer-Lemeshow goodness-of-fit at grade level, rank ordering with Receiver Operating Characteristic AUC and Kolmogorov-Smirnov, and population stability with the Population Stability Index. Term-structure coherence is checked by backtesting the cumulative PD curve against realised cumulative default rates over the full lease, loan, or facility life.

Loss Given Default is validated at the facility level rather than the obligor level, in line with IFRS 9.B5.5.28. FineIT reviews the bifurcation between secured and unsecured cash flow expectations, the haircut applied to real-estate, movable, and financial collateral by jurisdiction, the cure rate assumption, and the downturn adjustment. Backtesting compares realised recoveries to modelled recoveries across vintages, with attention to the tail of resolution times and the treatment of restructured exposures.

Exposure at Default is validated separately for drawn and undrawn components. For revolving products, credit conversion factors are benchmarked against realised utilisation at default, with product-level segmentation. For term products, the amortisation schedule is reconciled to contractual and behavioural cash flows. Stage one and stage two EAD reconciliation to the general ledger is performed as a controls test, and any unexplained variance above a materiality threshold is documented as a model limitation with remediation timeline.

Staging and SICR triggers are tested against the quantitative thresholds defined in IFRS 9.5.5.9 and the backstop thresholds in IFRS 9.B5.5.17. Stage migration is backtested across rating grades, with transition matrix stability monitored over multiple economic quarters. Qualitative triggers covering watch-list status, forbearance, and collateral coverage are reviewed for internal consistency, and borrower-level overrides are sample-tested against the governance memo.