IFRS 9 and CECL Expected Credit Loss Provisioning

Before 2018, a bank only booked a loss on a loan once something had actually gone wrong — a missed payment, a covenant breach, a bankruptcy filing. That is called an incurred-loss model, and regulators hated it after 2008, because banks kept lending right up until the crisis hit and only then, all at once, wrote down billions. The reserve showed up exactly when it was too late to matter.

IFRS 9 (the international standard) and CECL (its US cousin, Current Expected Credit Losses) replace that with a forward-looking rule: reserve for losses you expect, starting on day one, before any default has happened.

The staging idea

IFRS 9 sorts every loan into one of three stages. Stage 1 covers performing loans with no signs of trouble: the bank reserves only for losses expected over the next 12 months. Stage 2 covers loans whose credit risk has "significantly increased" since origination — still performing, but worse than when the loan was made: the bank must now reserve for losses over the loan's entire remaining life, not just a year. Stage 3 is loans that are actually in default: full lifetime loss recognition, similar to the old incurred-loss trigger. CECL skips the staging and simply requires lifetime expected loss from day one for everything, which is simpler but front-loads more reserve into the first year of every loan.

Computing the number

The core calculation is the same building block used everywhere in credit risk: expected loss equals probability of default times loss given default times the exposure at default, summed over the relevant horizon.

$$ECL = \sum_{t} PD_t \times LGD_t \times EAD_t$$

In words: for each period, multiply the chance the loan defaults in that period by how much the bank would lose if it did, by how much exposure is outstanding, and add it all up — over 12 months for Stage 1, or over the whole remaining life for Stage 2 and 3.

Worked example

A bank holds a $10 million commercial loan, 5 years remaining, currently Stage 1. Its 12-month PD is 1.5%, LGD is 40%, and EAD is the full $10 million.

1. Stage 1 reserve. $ECL = 0.015 \times 0.40 \times 10{,}000{,}000 = 60{,}000$, i.e. the bank reserves $60,000 against this loan.
2. Six months later the borrower's industry hits a downturn and its internal rating is downgraded two notches — a significant increase in credit risk, so the loan moves to Stage 2. Lifetime PD (cumulative default probability over the remaining 4.5 years) is now estimated at 18%, and LGD is unchanged at 40%.
3. Stage 2 reserve. $ECL = 0.18 \times 0.40 \times 10{,}000{,}000 = 720{,}000$, i.e. $720,000 — twelve times the prior reserve, without a single missed payment.

What this means in practice

That jump is the entire point and the entire controversy. A bank's earnings can swing sharply in a downturn purely from stage migration, before any cash is actually lost, which makes provisioning volatile and forces banks to build economic-forecast models (unemployment, GDP, house prices) directly into their loss estimates. Quants and credit analysts who build these models spend as much time on the "significant increase in credit risk" trigger — which is judgment-heavy and varies by bank — as on the PD and LGD inputs themselves.