HOW BANKING WORKS · CREDIT RISK THROUGH TIME 45
A loan does not become risky only on the day it defaults
A borrower can lose customers, suffer a fall in income, breach a covenant, become highly leveraged or fall behind on payments long before the final legal event called default. If the bank waited for complete failure before recognising deterioration, the accounts would report yesterday’s optimism for too long.
Expected credit loss, or ECL, is the discipline of estimating credit loss before the final default arrives. It asks the bank to combine what it knows about the borrower, the exposure, likely recovery and the future environment into a probability-weighted view of loss.
This article begins Batch 12 under How Banking Works: credit risk through time.
Expected loss is a forecast of damage, not a declaration that the borrower will fail
A bank can recognise expected loss on a loan that is still paying on time. The estimate reflects uncertainty rather than certainty.
credit exposure × probability of adverse outcome × loss after recovery = expected damage, adjusted for timing and scenarios.
This is a conceptual map, not the full accounting formula. Under IFRS 9, the measurement uses probability-weighted outcomes, reasonable and supportable forward-looking information and the time value of money.
The official accounting reference is IFRS 9 Financial Instruments.
Three quantities help make the mechanism visible
- Probability of default: how likely is a defined default event over the relevant horizon?
- Exposure at default: how much might the borrower owe when default occurs, including expected future drawings where relevant?
- Loss given default: how much of that exposure is likely to be lost after recoveries, collateral, guarantees and costs?
Banks can use more complex methods, but these three ideas provide a strong public mental model.
A simple expected-loss example
Imagine a S$1 million loan. In a simplified scenario, the bank estimates a 2 per cent probability of default and expects to lose 40 per cent of the exposure if default occurs after recoveries.
S$1,000,000 × 2% × 40% = S$8,000 expected loss.
That does not mean the bank believes it will definitely lose S$8,000 on this specific loan. Across a portfolio, probability-weighted expected losses provide a way to recognise the cost of credit risk before final outcomes are known.
The estimate changes when the borrower changes
If income falls, leverage rises or arrears appear, the probability of default can increase. If collateral falls in value, loss given default can worsen. If a borrower draws more of a committed facility, exposure at default can increase.
ECL therefore is not a number fixed at loan origination. It is updated as evidence changes.
Forward-looking information matters because credit risk is not purely historical
A borrower may have paid perfectly for five years and still become vulnerable if unemployment, rates, property prices or industry demand deteriorate sharply.
Expected-loss accounting therefore incorporates forward-looking economic information where it is reasonable and supportable rather than assuming the recent past will repeat unchanged.
Scenario weighting prevents one forecast from pretending to be certain
A bank can model several economic paths—base, upside and downside, for example—and weight the expected losses across them. The purpose is to avoid reducing the future to one deterministic line.
Scenario design itself creates model risk. If downside scenarios are too mild or probabilities are unrealistic, expected losses can still be understated.
IFRS 9 uses a staged deterioration model
In broad terms, performing exposures begin with a shorter-horizon expected-loss allowance. When credit risk has increased significantly since initial recognition, the measurement expands to lifetime expected credit loss. Credit-impaired assets remain in a more severe category with different interest-recognition mechanics.
The exact accounting tests require the standard and applicable implementation guidance. The public idea is what matters here: as credit risk worsens, the accounting loss horizon expands before final default.
Significant increase in credit risk is not the same as default
A borrower can become materially riskier without having defaulted. The bank looks for evidence such as rating deterioration, worsening payment behaviour, covenant stress, weaker cash flow or adverse sector conditions.
The purpose is early recognition. Waiting for legal default would defeat the mechanism.
Past-due status is evidence, but not the whole credit story
Days past due are useful because they are observable. But a borrower can deteriorate before missing a payment. Conversely, a temporary administrative delay does not automatically mean the borrower has suffered deep economic impairment.
Good credit monitoring combines payment behaviour with cash flow, leverage, industry, collateral, management and other evidence.
Collateral reduces loss severity, not necessarily default probability
A well-secured borrower can still default. Collateral mainly changes what the bank expects to recover after default rather than making the borrower incapable of missing payments.
This is why ECL separates borrower failure probability from recovery severity conceptually.
Read Why Collateral Does Not Repay a Loan.
Guarantees can change recovery and sometimes the risk path
An enforceable guarantee can improve expected recovery if the guarantor remains strong. But a guarantor exposed to the same economic shock can weaken at the same time as the borrower.
Expected-loss models should therefore recognise correlation rather than assuming every secondary protection remains perfect during stress.
Expected loss changes bank earnings before cash loss is final
When the bank increases its loss allowance, the provision expense generally reduces current profit. Retained earnings therefore grow more slowly or fall.
This is how worsening credit quality moves into the capital system before write-off.
Read Bank Equity and Retained Earnings for the loss-absorbing layer.
Expected loss and regulatory capital solve different jobs
| Layer | Main job |
|---|---|
| Expected credit loss / provision | Recognise expected deterioration through accounting |
| Capital | Absorb unexpected and realised loss while preserving viability |
The two interact, but one does not replace the other.
A provision can be adequate in aggregate and wrong loan by loan
Portfolio models estimate expected outcomes across many borrowers. Some loans will default unexpectedly; others will perform better than expected. The aggregate estimate can still be useful even when individual outcomes differ.
The discipline is therefore statistical and portfolio-based as well as borrower-specific.
Model overlays exist because models can miss structural change
A new pandemic, geopolitical shock or sudden policy change may not resemble historical data. Banks can apply expert overlays where model output does not capture emerging risk adequately, subject to governance and evidence.
Overlays can improve realism. They can also become a place where management bias enters. Strong governance records the reason, amount, evidence and exit condition.
Under-provisioning flatters profit today and weakens capital tomorrow
If the bank recognises too little expected loss, current profit appears stronger. Retained earnings look larger. Capital ratios can appear healthier.
When defaults finally arrive, the bank then has to recognise larger losses later, potentially during the same stress that has already weakened funding and markets.
Early recognition is therefore part of resilience, not just accounting conservatism.
Over-provisioning has costs too
Excessively pessimistic assumptions can suppress earnings and capital unnecessarily and may distort pricing or lending decisions. The objective is not maximum provision. It is a supportable estimate of expected loss.
Good banking therefore needs evidence, calibration and independent challenge rather than simply “more conservative” numbers.
Stress testing and ECL ask related but different questions
ECL measures accounting expected loss under the applicable reporting framework. Stress testing deliberately pushes the bank through severe adverse scenarios to assess resilience.
A stress test can therefore produce losses far beyond the accounting base case.
Read Bank Stress Tests.
The World Return is repayment evidence
A credit model represents the future at loan approval. ECL updates that representation as the world returns evidence: salary, sales, arrears, collateral values, industry conditions and actual repayment.
Expected credit loss is therefore not only a calculation. It is the bank admitting that the future promised at origination must be re-measured against the future actually arriving.
Four misconceptions to remove
| Misconception | Better model |
|---|---|
| “Expected credit loss means the borrower is expected to default.” | It is a probability-weighted estimate of credit damage, not a binary prediction. |
| “A loan paying on time needs no expected-loss allowance.” | Expected loss can exist even when the borrower remains fully performing. |
| “Collateral removes expected loss.” | Collateral can reduce loss severity but cannot guarantee repayment or stable value. |
| “The ECL number is objective fact.” | It depends on data, models, scenarios, assumptions and governance. |
A mastery test
- Why does expected-loss accounting recognise risk before final default?
- What do probability of default, exposure at default and loss given default represent?
- Why does forward-looking information matter?
- How can increasing ECL affect retained earnings and capital?
- Why is ECL different from stress testing?
If those answers connect, expected credit loss becomes visible as banking’s attempt to account for deterioration while there is still time to see it—before the final default turns uncertainty into fact.