Expected Credit Loss (ECL) under IFRS 9: The Loss You Book Before It Happens

What expected credit loss means, how the ECL calculation works using PD, LGD and EAD, and how finance leaders can navigate IFRS 9 impairment compliance without drowning in models.

By Mukesh Thakur, FCA, ACCA

Founder and Practicing Partner, Exactitude International (EXI)

Somewhere in your accounts, right now, sits a loan that has never missed a payment. The borrower is solvent. This month’s instalment landed on time. And IFRS 9 requires you to sit down and record a loss on it anyway.

If that sentence makes you uncomfortable, good. It should. That discomfort is the whole point of Expected Credit Losses, and it is also why ECL remains one of the most misunderstood, most argued-over, and most quietly mismanaged numbers on the balance sheet of any business reporting under IFRS.

I have lived with this shift from both ends. In 2009 I wrote a Taxmann guide to financial instruments when the world still ran on the old incurred-loss rules. Years later I guided a Mini Ratna public sector enterprise through its expected credit loss implementation under Ind AS 109, India’s mirror of IFRS 9, and watched a seasoned leadership team confront the day-one reality of provisioning against exposures that had never missed a payment. The old regime carried one fatal flaw. ECL exists to fix it.

From incurred loss to expected loss: why IAS 39 had to go

The old model under IAS 39 let you book a loss only once there was objective evidence that one had occurred: a missed payment, a covenant breach, a restructuring. Prudent in theory. In practice it was too little, too late. In 2008, banks carried performing loans at full value right up until borrowers stopped paying, then absorbed the damage all at once. The accounting told you the fire had started only after the building had burned down.

ECL inverts that logic. Rather than waiting for a loss event, IFRS 9 asks you to recognise, from day one, a forward-looking and probability-weighted estimate of the credit losses you expect across the life of the exposure. Not the loss you fear in a worst case. Not the loss you have already suffered. The loss you expect, weighted by its probability, and discounted to today.

How ECL is calculated: PD, LGD, EAD and discounting

Strip away the jargon and an ECL estimate is built from four ingredients.

Probability of Default (PD). The likelihood that the borrower fails to pay, over a defined horizon. This is the engine of the whole calculation, and, as we will see, the hardest part to get right, because IFRS 9 needs it as a term structure over time, not a single number.

Loss Given Default (LGD). If default happens, what share of the exposure do you actually lose after collateral, guarantees and recoveries? A loan fully secured by liquid collateral may carry an LGD of 10 percent; an unsecured trade receivable from an insolvent customer can approach 100 percent. Recoveries take time, so LGD must also reflect the cost of that delay.

Exposure at Default (EAD). How much will be outstanding at the moment of default? For a simple amortising loan this follows the repayment schedule. For an overdraft, credit card or committed but undrawn facility, it is a behavioural question: borrowers in trouble tend to draw down precisely when you wish they would not.

Discounting. Expected shortfalls are cash flows in the future, so they are discounted back to the reporting date, using the original effective interest rate of the asset, not the current market rate. Teams routinely miss this and quietly overstate or understate the provision.

In compact form: ECL is the sum, across each future period, of PD times LGD times EAD, discounted at the original effective interest rate. Multiplying three estimates together has an unforgiving property: a 20 percent error in any one of them is a 20 percent error in the answer, and the errors compound.

The IFRS 9 three-stage model, and where judgement bites

IFRS 9 sorts every exposure under the general model into one of three stages. Stage 1 is performing: you provide for losses arising from defaults possible in the next twelve months. Stage 2 applies where credit risk has increased significantly since initial recognition: now you must provide for losses expected over the entire remaining life of the asset. Stage 3 is credit-impaired: lifetime losses again, and interest revenue is now recognised on the net carrying amount, after the loss allowance, rather than the gross amount, so the deterioration reaches your top line too.

Read that once more, because the whole controversy lives in one phrase: increased significantly. The standard deliberately refuses to hand you a bright line. It offers a rebuttable presumption at 30 days past due and a backstop, then asks for judgement: relative deterioration since origination, not absolute credit quality today. A BBB exposure that was AA at origination may need Stage 2 treatment; a B-rated loan that was B on day one may not. That is precisely where two competent finance teams, looking at the same book, can reach materially different provisions, and both can be defensible.

ECL worked example: the cliff between Stage 1 and Stage 2

Numbers make the stakes concrete. Take a five-year unsecured corporate loan of 1,000,000 in your reporting currency. The borrower is performing. Your models say the probability of default in the next twelve months is 2 percent, the probability of default at some point over the remaining life is 12 percent, and the loss given default is 40 percent. Ignore discounting for a moment to keep the arithmetic visible.

Position	PD applied	LGD	ECL allowance
Stage 1 (12-month ECL)	2%	40%	8,000
Stage 2 (lifetime ECL)	12%	40%	48,000

Same loan, same borrower, same payment record. The transfer between stages multiplies the provision six-fold.

Nothing about the loan changed except your assessment of its risk. The borrower has still never missed a payment. Yet the allowance jumps from 8,000 to 48,000 overnight, and the difference goes straight through profit or loss. This cliff is why staging is the single most consequential judgement in the entire framework, and why auditors and regulators probe it hardest. Get staging wrong and no amount of modelling sophistication downstream can save the number.

Why the ECL calculation is harder than the formula

If ECL were only PD times LGD times EAD, a spreadsheet would suffice. The complexity hides in what each input demands.

Lifetime PDs are a curve, not a number. For Stage 2 you need the probability of default in year one, year two, year three and beyond, conditional on having survived so far, applied period by period against a moving exposure. Most organisations have data to estimate a twelve-month PD. Very few have clean default histories long enough to build a lifetime term structure, so they extrapolate, and every extrapolation is an assumption an auditor will ask you to support.

Point-in-time versus through-the-cycle. Credit ratings and Basel regulatory PDs are largely through-the-cycle: smoothed averages across good years and bad. IFRS 9 demands point-in-time estimates reflecting conditions now and conditions forecast. Banks that assumed their Basel models could simply be reused discovered otherwise: regulatory PDs are floored, twelve-month only, and calibrated for capital, while regulatory LGDs are deliberately tilted to downturn conditions. Reusing them without adjustment is a methodological error, not a shortcut.

EAD is a behavioural forecast. For revolving facilities you must estimate how much of the undrawn limit will be drawn by the time of default, and over what behavioural life, which for credit cards routinely extends beyond the contractual notice period. This is one of the few places the standard itself accepts that contractual terms do not capture economic reality.

Small portfolios starve the statistics. A portfolio of twenty large corporate loans with one historical default in a decade gives you almost nothing to model with. Low-default portfolios push you toward external benchmarks, expert judgement and careful documentation of why the chosen proxy is reasonable and supportable.

Forward-looking information and scenario weighting: the part that keeps CFOs awake

ECL is not confined to today’s data. IFRS 9 requires reasonable and supportable forward-looking information: your view of GDP, unemployment, interest rates and asset prices, built into multiple probability-weighted scenarios. You are, in effect, being asked to forecast the economy and price that forecast into your accounts.

Why multiple scenarios rather than one best estimate? Because credit losses are nonlinear. A mild downturn does modest damage; a severe one does disproportionate damage, as collateral values and default rates deteriorate together. The average of the outcomes is worse than the outcome of the average, so running only your base case systematically understates the provision. A simple illustration, using the Stage 1 loan from earlier:

Scenario	Weight	Scenario ECL	Contribution
Upside	20%	5,000	1,000
Base case	60%	8,000	4,800
Downside	20%	25,000	5,000
Probability-weighted ECL			10,800

The weighted answer is 35 percent higher than the base case alone. The downside scenario, at only a 20 percent weight, contributes more than the base case.

In calm years this exercise is merely uncomfortable. In volatile ones, a pandemic, a rate shock, a regional conflict, it becomes far harder, and the management overlays that teams bolt on to capture what the models cannot see turn into a governance minefield of their own. Overlays are legitimate, sometimes essential. Undocumented overlays that conveniently move the number in one direction, quarter after quarter, are how ECL estimates lose credibility.

The simplified approach and the provision matrix for trade receivables

For most non-financial companies, mercifully, IFRS 9 offers a simplified approach for trade receivables, contract assets and lease receivables: skip the staging debate entirely and book lifetime ECL from day one, typically through a provision matrix built on historical loss rates by ageing bucket, adjusted for forward-looking conditions.

Ageing bucket	Receivables	Loss rate	Lifetime ECL
Not yet due	5,000,000	0.5%	25,000
1 to 30 days past due	1,500,000	2%	30,000
31 to 60 days past due	600,000	5%	30,000
61 to 90 days past due	300,000	10%	30,000
Over 90 days past due	200,000	30%	60,000
Total	7,600,000		175,000

An illustrative provision matrix. The loss rates are the judgement: historical experience, adjusted for what you expect ahead, and segmented where customer groups behave differently.

Do not mistake simplified for trivial. The three weaknesses I encounter most often: loss rates lifted from history and never adjusted for forward-looking conditions; a single undifferentiated matrix applied across customer segments with very different risk, exporters and domestic buyers, government and private counterparties; and the quiet assumption that amounts not yet due carry zero risk. Each is easy for an auditor to challenge, and each is fixable in a week of honest analysis.

IFRS 9 ECL compliance: five practices that keep you defensible

So how does a finance leader stay both compliant and sane? A few principles I return to on every engagement.

Get your staging discipline right before you polish your models. The single biggest source of ECL error I see is not a flawed PD curve; it is inconsistent, undocumented, or quietly optimistic movement between stages. Define your significant-increase triggers, quantitative and qualitative, apply them consistently, and record why. The worked example above shows what is at stake in that one judgement.

Treat documentation as part of the number, not paperwork about it. An ECL figure you cannot explain is an ECL figure you cannot defend, whether to your auditor, your board, or your regulator. Every scenario weight, every overlay, every rebuttal of a presumption deserves a reasoned trail.

Govern the judgement, not just the model. Who signs off the macroeconomic scenarios? Who approves overlays, and on what evidence? Put credit risk and finance in the same room. ECL fails quietly when it becomes a black box owned by one team and understood by none.

Right-size the machinery. A corporate with trade receivables does not need a bank’s scenario engine, and a bank cannot hide behind a provision matrix. Proportionality is built into the standard; use it deliberately, and document why your level of sophistication fits your exposure.

Mind your disclosures. IFRS 7 credit-risk disclosures are where users, and reviewers, actually judge the quality of your estimate. When the IASB completed its post-implementation review of the IFRS 9 impairment requirements in 2024, it concluded the model is working as intended and needs no overhaul, a reassuring verdict in itself, but it singled out the areas still causing friction: how firms assess significant increases in credit risk, how they incorporate forward-looking and increasingly climate-related information, and the usefulness of their credit-risk disclosures. That is the examiner telling you where the marks are lost. Read it as such.

The mindset shift that separates good from merely compliant

Here is what I try to leave every CFO with. ECL is not an accounting penalty to be minimised. Handled well, it is the earliest warning system you own. The same discipline that produces a defensible provision, honest staging, forward-looking scenarios, integrated credit and finance data, is the discipline that tells you where your book is softening before the cash flow does. The businesses that struggle with ECL are the ones that outsourced it to a spreadsheet and a quarter-end scramble. The ones that master it turned a compliance obligation into a management instrument.

The loss you book before it happens is not a fiction the standard-setters invented to torment you. It is the standard forcing you to admit, in the cold language of the balance sheet, what your credit team probably already suspects. The only real question is whether you learn it on your own terms, with rigour and documentation behind you, or on the auditor’s.

Frequently asked questions on expected credit loss

What is expected credit loss (ECL) under IFRS 9? ECL is a forward-looking, probability-weighted estimate of the credit losses expected on a financial asset, recognised from the day the asset is first booked. It replaces the incurred-loss model of IAS 39, under which losses were recognised only after a default event had occurred.

What is the difference between 12-month ECL and lifetime ECL? 12-month ECL covers losses from defaults possible within the next twelve months and applies to Stage 1 performing assets. Lifetime ECL covers losses from defaults possible over the entire remaining life of the asset and applies once credit risk has increased significantly (Stage 2) or the asset is credit-impaired (Stage 3).

What counts as a significant increase in credit risk? IFRS 9 sets no bright line. It requires a relative assessment against the risk at initial recognition, using quantitative and qualitative indicators, with a rebuttable presumption that risk has increased significantly once payments are 30 days past due. Each entity must define, document and consistently apply its own triggers.

Does IFRS 9 ECL apply to trade receivables? Yes. Trade receivables, contract assets and lease receivables qualify for a simplified approach: lifetime ECL is recognised from day one, usually through a provision matrix of historical loss rates by ageing bucket, adjusted for forward-looking conditions. Simplified does not mean optional, and it does not mean historical rates can be used unadjusted.

Is Ind AS 109 the same as IFRS 9 for ECL? Ind AS 109, applicable to Indian companies, mirrors the IFRS 9 expected credit loss model in all substantive respects, including the three stages, forward-looking information and the simplified approach for trade receivables. Practitioners moving between the two frameworks will find the impairment requirements essentially identical.

Mukesh Thakur is a Fellow Chartered Accountant (FCA) and ACCA with over 22 years across audit, international financial reporting and CFO advisory. Author of the Taxmann Guide to Financial Instruments (2009), he is Founder and Practicing Partner of Exactitude International (EXI), advising organisations across India, the Middle East and the UK on IFRS and Ind AS adoption, expected credit loss implementation, finance transformation and governance.