IFRS 9 flings loan-loss provisions haphazardly higher

Under the standard, cash piles for bad loans were expected to ramble. Just not quite so much

Bankers always expected the new IFRS 9 accounting standard to hurl loan-loss provisions higher, in something like a scattered arc. But they’ve been taken aback by just how dispersed and uneven the arc has turned out to be.

“Because of the way internal models are developed, some variability is to be expected, but this is a much bigger variability than I would have anticipated,” says Rastislav Kovacik, head of risk reporting at the Erste Group in Vienna, echoing the sentiments of many.

The industry had complained about how far afield loan-loss provisions might wander under International Financial Reporting Standards 9, the biggest refashioning of accounting in decades. Bankers had also objected to its blanketing of questionable loans with lifetime provisions – a very different and subjective approach from the 12-month, no-wiggle-room formula under IAS 39, the predecessor rule.

To no avail – IFRS 9 went into effect in January last year.

Still, bankers say they don’t quite get where the sheer breadth in loan-loss provisions comes from: by one study, the difference is a factor of four.

For the industry, the discrepancies are a competitive matter. No bank wants to set aside more for loan losses than its rivals do. As well, meeting higher provisions means going into retained earnings, possibly biting into capital.

A group of banks surveyed by the European Banking Authority experienced a 47-basis-point reduction in Common Equity Tier 1 (CET1) capital, on average, upon adoption of IFRS 9.

There are any number of possible culprits in the generation of motley loan-loss provisions under IFRS 9. But most appear to be tied to the calculating of lifetime losses on a given loan, an exercise requiring assumptions on things years in the future, and to something else tucked inside that: human discretion, which unsurprisingly has spawned an array of economic forecasts and less consistency on what loans might be deteriorating and therefore relegated to lifetime provisioning.   

Wrestling with Basel, under IFRS 9

A lot of variability seems to start in three areas. The first is the sensitivity of a loan to given economic factors; for instance, a mortgage affected by the local economy, or an overseas loan buffeted by geopolitics. The second is the criteria for deciding which loans might be deteriorating and would need to be moved from a 12-month to a lifetime expected loss. Last are the methods for determining the ‘probability of default’ (PD) and the ‘loss-given default,’ or how much a bank would lose when a loan goes south over the entirety of its term.

Many of those pliable factors are the very ones used in the Basel Committee on Banking Supervision’s (BCBS) advanced internal ratings-based (A-IRB) approach, which also stirred up wide variations in credit risk-weighted assets due to differences in methodologies. As the IFRS 9 model is based on Basel’s, it too is producing bumpy results.

But the Basel and IFRS 9 approaches differ in two big ways. First, the Basel risk models use a ‘through-the-cycle’ approach that measures average losses over rolling 12-month periods; IFRS 9 relies on ‘point-in-time’ estimates of loan losses based on the prevailing credit conditions at each quarter-end.

Because of the way internal models are developed, some variability is to be expected, but this is a much bigger variability than I would have anticipated

Rastislav Kovacik, Erste Group

Second, IFRS 9 requires that loans be bucketed into 12-month or lifetime loan-loss categories; Basel required only 12-month estimates. The use of the shorter-focused Basel models for longer horizons has created some of the variations.

“Since the foundation of IFRS 9 models are the Basel models, the methodologies for producing point-in-time forward-looking estimates will drive variability,” says a credit risk executive at an Asian bank in Singapore.

To illustrate the point, a Global Credit Data study highlighted the case of a hypothetical borrower in the UK manufacturing sector. A five-year unsecured loan to this company – assigned a 0.75% probability of default – was given a loss-provisioning charge of 0.5% at some lenders, but less than a basis point at others.

Even if these banks were taken as outliers, the study noted, most banks still calculated provision charges of between 5.9 basis points and 24.2 basis points – “which we consider still a significant difference, leading to the conclusion that banks’ provision charges vary by a factor four or more.”

Photo: Ulrich Roth

BIS tower, home to the Basel Committee

“When you test the models against the same reference portfolio, you find a large distribution of outcomes – meaning two different institutions would treat expected losses differently on the same defaulted loan,” says Richard Crecel, executive director of Global Credit Data.

IFRS 9 requires banks to predict the future loss of all assets at the time of origination or purchase, and set aside provisions for these assets. Under IAS 39, banks provisioned for assets only once they defaulted – possibly too late for the bank saddled with them.  

IFRS 9’s data requirements have already brought more conservatism to balance sheets. At the 36 largest European Union banks complex and illiquid Level 3 assets – loans difficult to value because they are barely traded – rose to €175.2 billion ($196 billion) at end-2018 from €140.2 billion the year prior. Commerzbank saw the largest increase, with Level 3 assets up 76% to €9.7 billion, followed by Crédit Agricole with a 74% jump to €8.4 billion. Coming in third, Barclays added 20% more, to sock away €21.2 billion.

Open to discretion

IFRS 9 requires banks to base their estimates of a loan’s lifetime expected losses on ‘reasonable and supportable’ information. But it’s left up to the banks to define that. The banks first have to come up with an economic forecast, then translate it into an estimate of expected losses.

Unless very narrowly defined, a forecast by necessity will encompass the global economy and the political winds in a number of countries. And today, that’s anyone’s guess: the US-China trade war, the fallout of Brexit, the prospect of recession, all are colouring the economic outlook.

And banks do not just produce one forecast, but many, typically between three and five, with the midpoint representing the consensus view. Banks then assign weights to the probability of each scenario.  

Louise Lindgren, Länsförsäkringar Bank

“You can have four scenarios and weight them equally, or place a higher weight on the normal business cycle and a lower weight on the more stressful business cycle,” says Louise Lindgren, chief risk officer at Länsförsäkringar Bank in Stockholm. “That weighting will have quite a big impact.”

In developing scenarios, banks apply credit judgment along with as much information as can be gathered to objectively support the assumptions.

“If there are emerging and more urgent downside situations that we believe the models mightn’t have captured because there mightn’t be any empirical precedence for that, then, as much as possible, we use objective estimations to quantify the impact and try to express that via management overlay, if necessary,” says the executive at the Asian bank.

Scotiabank, for instance, uses three scenarios generated by its economics department. It starts with a baseline scenario, then develops optimistic and pessimistic versions, along with initial probabilities. The scenarios and the weightings assigned to each go to an oversight committee for approval. Loan-loss estimates are based on economic factors that might be affected by geopolitical developments.

We use objective estimations to quantify the impact and try to express that via management overlay, if necessary

Executive at an Asian bank

It might, for example, assign a probability that the US-China trade war will escalate – a ready example of the impact political risk can have on banks’ ability to forecast the future.

“The possibility of a US-China trade war is certainly taken into account when we generate scenarios,” says Mark Engel, senior vice-president of risk analytics and cyber at Scotiabank in Toronto. “Our scenarios project factors like GDP, unemployment and commodity prices, and estimate the extent a US-China trade war would affect those factors.”

Similarly, banks look closely at local developments. Länsförsäkringar Bank, for instance, will ask its local branches for expectations on customers and regional business conditions. If a local government were thinking of limiting a bank’s ability to lend in say, energy, that would be factored in.

“If something like that were to take place, we would take that into account,” says a senior accounting executive at a large bank. “However, if our portfolio is short-dated – and a good bit of it is – then those kinds of changes wouldn’t happen overnight. If there was legislation, it would take time to take effect.”

After forecasting and weighting, banks then assign their loans a sensitivity to the relevant parts of the economy – which can also vary. A bank whose portfolio is largely mortgages, for instance, is less likely to be affected by US-China relations than a bank that makes manufacturing or project-finance loans.

It’s not easy to calibrate a point-in-time PD, especially for books of business where very little data is available

Jimmy Skoglund, SAS

In the current environment, banks with stronger macroeconomic sensitivity will tend to calibrate a lower point-in-time PD than banks with less sensitivity. That is, banks with strong sensitivity will take a more optimistic view during good economic times, and as a result their point-in-time PD would be lower. During stressful times, the situation is reversed: banks with a stronger macroeconomic sensitivity calibrate a higher point-in-time PD than banks with lower sensitivity.

This is especially pronounced in segments where data is lacking. All the banks participating in the Global Credit Data study had average point-in-time PDs that were nearly 50% lower than their through-the-cycle PDs for the hypothetical UK borrower in the manufacturing segment, indicating high sensitivity to macroeconomic conditions.

“This is where banks differ a lot. It’s not easy to calibrate a point-in-time PD, especially for books of business where very little data is available. The variability is due mainly to the macroeconomic calibration,” says Jimmy Skoglund, principal product manager at risk analytics provider SAS in Motala, in northern Sweden.

The second big driver of volatility is IFRS 9’s requirement to bucket loans into one of three categories, yet another subjective effort. Stage 1 assets, loans in good standing, need set aside only 12 months of expected losses. Stage 2 assets, where ‘significant deterioration’ has occurred, must get lifetime provisions. Stage 3 assets, the outright impaired, must have lifetime provisions and get a reduction in expected interest payments.

The human decision on which loans land in Stage 2 with its lifetime loss estimates from Stage 1 is a marked departure from the Basel model, where risk is determined with a formula over a 12-month period. 

Depending on the portfolios, the methodologies underpinning lifetime-loss estimations are quite different

Analyst at an Asian bank

To decide what goes in Stage 2, banks use watch lists and ratings downgrades and the bank’s own rate of default. For the same customer, one bank might determine a three-notch downgrade is required for a Stage 2 transition; another might say one is enough.

Even once a customer goes to Stage 2, there are vagaries in coming up with lifetime-loss estimates.

“Depending on the portfolios, the methodologies underpinning lifetime-loss estimations are quite different,” says the analyst at an Asian bank. “That is where methodologies could vary, and the portfolio type would vary. A bank with a lot of project finance loans sitting in Stage 2 would warrant a huge ECL [expected credit loss] increase.”

The third big driver of variability is the way banks derive point-in-time probability of default and loss-given default – both needed to come up with estimated losses, and both malleable by economic conditions. Banks have amended their existing credit-risk and stress-testing models to derive forward-looking default rates and losses. But the ingredients that go into the models – economic forecasts and the weights assigned to them – are quite different across banks.

As late as mid-2018, banks were still coming to grips with their IFRS 9 models.  

Scott Aguais

“The Number One culprit in the variability is that the methodologies to go from through-the-cycle PD to point-in-time PD are all over the place,” says Scott Aguais, managing director of an eponymous credit analytics company in London. “Some banks use simple multipliers to raise or lower point-in-time PDs relative to through-the-cycle, which is a very brute-force, inaccurate approach.”

Banks have not settled on a uniform practice for extracting point-in-time PDs from through-the-cycle ones. They take the internal ratings-based (IRB) models as their starting point, and then need to develop point-in-time PDs for the maturity of the loan. Most banks are still struggling with that.

“A big problem for IRB banks is to reconcile your IRB outcomes with IFRS 9 outcomes,” says Alexander Petrov, a credit risk executive at Nordea in Stockholm. “Not every bank can do that.”

In developing its IFRS 9 models, Scotiabank had to collect a considerable amount of data to ensure it had a reliable, historical record.

“With the [A-IRB] models, you just needed to know averages,” says Niall Whelan, vice-president of enterprise stress testing at Scotiabank in Toronto, “but here you’re interested in how things are fluctuating through time.”

As a clearer picture emerges of the factors driving variability, the question is what, if anything, could be done about it.

Waiting for convergence

Frustrated by the gaping differences, some have suggested regulators, with their access to the models and data used by banks, study the different approaches and come up with something less open to interpretation.

“If a regulator wanted to make changes, they should understand what drives the differences,” says Engel. “It would be important to understand how much of the differences were due to which aspects of the models. Regulators would be in a unique position to understand that.”

Given IFRS 9’s international nature, however, some say it falls to the BCBS itself to issue guidelines on the models used for IFRS 9.

“BCBS might want to come up with prescriptive rules around which methodologies might be favourable,” says the credit risk executive at the Asian bank.

The Basel Committee declined to comment.

Absent any move from Basel, though, national regulators might act locally. One, the Monetary Authority of Singapore, set minimum-loss allowances that could even them out: it ordered large banks to keep at least a 1% loss allowance on credit exposures.

If a regulator wanted to make changes, they should understand what drives the differences

Mark Engel, Scotiabank

But any more radical steps would effectively neutralise IFRS 9 and send banks back to the Basel regime, with its far looser rules on loan-loss provisions. For that reason, most people think IFRS 9 will remain as is.  

“Regulatory dampening of volatility would lead us essentially to Basel-expected loss estimates,” says Skoglund.

Changes to the models themselves have been proposed. Some Canadian banks have begun employing algorithms that smooth ‘ratings migration’ as portfolios are subjected to lifetime-loss estimations.

For some, though, the variability is in fact desirable and will give banks perspectives that will strengthen the sector as a whole. Stocking up on loan-loss provisions in periods of plenty makes sense to be prepared for the downturns. Some also say IFRS 9 gives them a clearer understanding of underlying risk, their own and others’.

“IFRS 9 provides more insight into the portfolio. It’s more granular than ordinary risk models for PDs and LGDs [loss-given defaults],” says Lindgren.

And many believe the variability in IFRS 9 will taper off as banks get used to it. Its creators think so, too.

“Taking into account that IFRS 9 was such a big accounting change for the banks, it’s going to take some time for the models to settle down. We will see convergence over time,” says Riana Wiesner, a technical staff member at the IFRS Foundation in Johannesburg.

And convergence will likely wash out some of the more extreme estimates.

“Variability is inherent within IFRS 9 and is therefore expected to some degree as the portfolio composition and credit quality mix is different across peer banks,” says Abhishek Jain, global head of wholesale impairment at HSBC in London. “The changes to the accounting standards typically present an opportunity to achieve convergence over time as improved disclosures and availability of comparative information across peer banks emerge.”

Jonathan Bingham, a partner in the banking practice at KPMG in London, agreed.

“Inconsistency will come out of the system over time,” he says. “The industry is trying to eliminate some of the inconsistency by sharing views on how they calculate their models.” 

But there is some inconsistency that is down to very different business practices, and probably won’t change.

“There is a degree of variability between banks that gets attributed to modelling, but there’s also a fair amount due to portfolios and business practices, like collections,” says Engel.

He adds, “To the extent that there are differences due to portfolio strategy and business practices, those should continue to persist.”