In the Spotlight: Understanding expected credit losses – what metrics might help? - Banking

Publication date: 14 Nov 2019

In the Spotlight: Understanding expected credit losses - Banking

At a glance

IFRS 9’s expected credit loss (ECL) model for measuring impairment provisions has now been in place for over a year. However, the market’s understanding of what ECLs mean is still developing. In this publication, we give insights into what ECL is and is not, indications of why it might differ across banks and portfolios, and our suggestions of what metrics can be useful in understanding and comparing ECL provisions.

IFRS 9 – Finding the baby in the bath water

We all expected IFRS 9 to be complicated and there’s little doubt that, on that front, it delivered. What about the rest, though? Wild volatility? Procyclicality? In many parts of the world, an economic downturn seems nearer by the day and yet there’s been little, if any, meaningful movement in banks’ performing allowances. Since transition, some have gone up. Others down. Many embraced IFRS 9 (in certain cases, so much so that they’ve begun incorporating it into credit decisioning), while some users ignored it, and attempts to compare and analyse thus far have created more questions than answers. With all that in mind, hopefully it’s clear why some might worry about surprises down the road.

Let’s start with a bit of history. Before 2018, IFRS employed an ‘incurred loss’ model, whereby loan impairments were only reflected in financial statements once there was evidence that they’d been incurred. Of course, during the financial crisis, that meant that lenders, investors, regulators and others were caught off guard when losses spiked. The solution? Forward-looking expected credit losses (ECLs) to give an earlier signal when things are taking a turn for the worse. What’s that really mean? Very possibly, not what you expect.

For starters, as with any expected loss model that probability-weights a number of possible outcomes, projections will rarely equal the actual losses realised. That’s especially true for Stage 1 which calculates ECLs only over 12 months and, depending on the lives of the underlying loans, that difference could mean missing an important part of the story. Putting that and the idyllic notion of perfect foresight aside, ECLs are good as a directional indicator of management’s assessment of underlying credit risk. Certainly, an earlier warning than the incurred loss approach.

Unlike traditional measures that help us to understand what’s already happened (think net write-offs), ECLs give us insight into management’s take on what’s yet to come. Therein lies both their value and intended purpose – an earlier warning of troubles ahead.

Still, it’s important to emphasise what ECL is, and what it isn’t. For instance, knowing that a majority of losses are being measured on a 12-month (rather than lifetime) basis is important context. Ditto that in many cases they’ve been designed during a period when economic conditions have been benign, meaning that shortcuts that are immaterial now might not be someday. Additionally, data and models are based largely on the experience around past performance and expert judgements of expectations going forward (since the future might be different from the past). Of course, there are other fun twists like idiosyncratic borrower behaviour and the fact that, when all hell breaks loose (remember where we started), cliff effects will prevail. So, what to do? Were the naysayers right? Should traditional measures like net write-offs be favoured instead of these new forward-looking estimates? No. If the past has taught us anything, it’s that we need to pay close attention to both. Here are a few things to keep in mind as you do.

ECLs often differ across different portfolios, banks and over time. Reasons for this include:

Underlying portfolios – a bank’s portfolio is specific to its strategy, and some concentrate on certain borrower types (such as wholesale, retail, sub-prime and so on). Some are predominantly domestic or regional, others global. No two banks are truly alike and, for that reason, the risks underlying their portfolios, changes in those risks, and the resulting ECLs are sure to be different too.

Management’s assessment of the risks – ask three banks about the credit riskiness of a given loan, and you’re likely to get three different answers. These differences in perception and strategy around credit risk lead to differences in ECLs, since management’s assessment of these risks and changes in them (for example, including whether to apply 12-month or lifetime losses) are key inputs. Remember, ECLs are entity-specific in that they’re based on management’s own assessment rather than a market view.

Expectations about the future – IFRS 9 requires banks to consider multiple forward-looking scenarios, weighting the results by their relative probabilities. Improvements and deteriorations in economic conditions, political events, trade wars – you name it. Of course, those determinations are highly judgemental and thus are also likely to vary from one bank to another.

Modelling approaches – there’s often more than one way to model things. Consider again forward-looking economic scenarios. Many banks use 3 scenarios, others 4, 5 or more. Some use Monte Carlo simulations to consider far more. All can be perfectly acceptable, yet each could yield a different result – even before considering the complexities of scenario selection and weighting. In other words, even for two banks with identical economic outlooks, similar portfolios and the same number of scenarios, ECLs could be very different – severity of downside and scenario weightings being key.

Credit risk management – of course, there’s also always differences which arise from each bank’s approach to credit risk management and recovery.

With so many possible differences and an accounting standard that’s intentionally a ‘broad church’, how best to understand and compare provisions? Short answer – it isn’t easy. Here’s what we’d focus on:

Metric Example What it answers Limitation
Write-off coverage ECL / annualised net write-offs

Net write-offs / Stage 3 exposure
How adequately is the bank provided, compared to write-offs? Write-off policies might differ by bank, thus impacting the coverage ratios.
Stage composition % of overall exposures in each of Stages 1, 2 and 3 What is the credit quality of the portfolio by stage? Criteria for moving to Stage 2 and default definitions might differ by bank, thus affecting the classifications.
Stage 2 duration Average # of days in Stage 2 before cure or default How responsive and effective are credit risk management activities? Criteria for moving to Stage 2 might vary, limiting comparability.
Overlays ECL / Gross exposure (typically calculated for performing Stage 1 / 2, and for non-performing Stage 3 separately) How adequately is the bank provided, relative to its exposures? Exposures might be defined and presented differently.


Metric Example What it answers Limitation
Multiple scenario sensitivities ECL if each scenario were the only one How sensitive is the ECL under different economic scenarios? These are not a forecast of the actual ECL in a chosen scenario.
Stage 2 composition Breakdown of Stage 2 exposures by current, 1–29 days past due, and 30+ days past due How forward-looking are the criteria for moving to Stage 2 (for example, as opposed to relying on days past due)? While this provides information about the breakdown within Stage 2, it doesn’t necessarily shed light on what first caused them to get there.
Stage 2 duration Average # of days in Stage 2 before cure or default How responsive and effective are credit risk management activities? Criteria for moving to Stage 2 might vary, limiting comparability.
Overlays % or $ of ECL not substantiated purely by modelled outcomes What portion of the provision is not model-driven? Overlays might exist for a variety of reasons (such as data limitations, unmodellable events etc). A high level of overlay can call into question the effectiveness of the models, yet equally their absence might mean that risks haven’t been adequately captured.

Of course, endless others exist and none alone is a silver bullet. With the varying limitations and differences described above, the trick isn’t to look at one metric, but to consider all to the extent that underlying information is disclosed. These aren’t the only things we’d look at either. Others like net write-offs, stress-testing results and so on can provide important context when available. Piecing it all together, the picture might just be clearer than you think.

Where do I get more details?

For more information, please contact Christopher Wood (, Sandra Thompson ( or Mark Randall (

  • Add Selected section
  • OR
  • Add Full document
  • Print Selected section
  • OR
  • Print Full document
  • Copy link
  • Add to my favourites

Warning, this action will download the whole document into PDF format.


Warning, this action will add the whole document to my documents.

Back to top