IFRS 9 ECL: separate stage migration from model changes
An IFRS 9 allowance increase can come from weaker borrowers, a different economic outlook, a larger loan book, or a changed model. Separate those mechanisms before calling the change a deterioration in collections. The useful output is a stage-and-driver bridge, with the unexplained remainder visible.
Define the accounting population
IFRS 9 provides the impairment framework; IFRS 7 supplies the related disclosure context. This article concerns loans using the general IFRS expected credit loss approach. Trade receivables can use a simplified approach. US GAAP CECL does not use the same three-stage architecture, and regulatory classifications need their own definitions.
Record gross carrying amounts and allowances for each disclosed stage. As the IFRS 9 project summary explains, a significant increase in credit risk changes the loss horizon from twelve-month to lifetime expected losses under the general approach; credit-impaired assets need separate treatment. Twelve-month expected loss is not simply cash shortfalls expected during the next twelve months. It relates to lifetime shortfalls associated with possible defaults during that horizon.
Hypothetical migration bridge
Suppose a bank starts with allowances of 8, 16 and 26 across stages 1, 2 and 3, totaling 50 million. Its closing allowance is 68. The note identifies 6 from new lending, 9 from migration and remeasurement, 5 from revised forecasts, minus 4 from write-offs, and 2 from currency and other movements. The bridge is 50 + 6 + 9 + 5 - 4 + 2 = 68.
The 18 increase is therefore not an 18 increase in realized defaults. Neither does the bridge establish that model changes are harmless. The forecast adjustment may reflect real economic information before borrowers miss payments. Describe both the accounting mechanism and the business interpretation, without treating one as a substitute for the other.
A worksheet that preserves uncertainty
Create columns for opening exposure, opening allowance, transfers in, transfers out, new originations, repayments, write-offs, model changes, management overlays and closing balances. Add the allowance rate for each comparable portfolio. Keep exposure movement separate from allowance movement; transferring 100 of loans is not transferring 100 of losses.
Where disclosures combine migration and remeasurement, retain the combined row. Do not manufacture a pure migration estimate by subtracting unrelated totals. Mark whether figures include purchased credit-impaired loans, guarantees, undrawn commitments or foreign exchange. An off-balance-sheet provision can otherwise be incorrectly compared with only balance-sheet loans.
For acquired portfolios, record the acquisition date and initial measurement basis before interpreting a transfer as deterioration in loans originated by the reporting bank.
Test the explanation next period
Observe whether the stage 2 share increases, whether borrowers return to stage 1, and whether stage 3 inflows follow the forecast deterioration. Compare changes in risk criteria and cure periods. A bank can change the reported migration pattern through methodology even when the borrowers have not changed by the same amount.
Read overlay releases alongside model recalibration. If an overlay falls by 7 while modeled losses rise by 7, the flat total does not mean assumptions stayed constant. Ask which risks moved into the model and which disappeared. Public notes may not permit independent probability-of-default or loss-given-default reconstruction; state that boundary explicitly.
FAQ
Does a higher stage 2 balance prove default is imminent?
No. Stage 2 signals a significant increase in credit risk under the applicable policy, not a declaration that every loan will default. Inspect the policy, portfolio mix and later migration before drawing a borrower outcome conclusion.
Can two banks be compared using one allowance ratio?
Only cautiously. Collateral, portfolio age, geography, loss horizons and accounting framework affect the numerator. Compare matched portfolios first and show missing dimensions. A more granular bridge is useful even when it prevents a confident ranking.