US credit risk modellers prepare for life after IRB

Stress tests and economic capital calculations may not carry the same weight as Basel ratio

If the rumours are to be believed, US regulators are preparing to end the use of internal models to calculate capital requirements for credit risk.

Some banks are already moving on. One senior risk modeller says chief risk officers are reluctant to sustain the levels of investment needed to maintain the internal ratings-based (IRB) approach for credit risk. “They are saying: ‘If regulators have decided that the models don’t work, then we don’t have to use them, and we shouldn’t use them.’”

He thinks they may come to regret that decision. The senior risk modeller, who currently works at an advanced approaches bank, recalls asking some former colleagues at a lender that only uses standardised approaches if they still calculated value-at-risk for credit portfolios. “No,” came the reply. “We’ve abandoned all these models.” He then asked if the bank knew its one-in-50-year or one-in-100-year exposure for specific portfolios, “and no-one could give me an answer”.

Brett Ludden, Sterling Point

When the Basel Committee on Banking Supervision first permitted banks to use internal models in 2004, under the Basel II framework, it sowed the seeds for a flourishing ecosystem of credit risk models and the collection of data needed to run them.

However, there are growing fears that US regulators’ draft of the Basel III framework, which could be published by the end of March, will no longer allow IRB to be used. This raises questions about the future of credit risk models – and the people who build and manage them. Banks are expected to cut back on credit modelling, and there is already talk of a commensurate reduction in headcount. As the risk modeller puts it, credit risk quants will have to “totally reinvent themselves”.

Brett Ludden, co-head of financial services at advisory firm Sterling Point and a former senior credit risk manager at Capital One – which uses its own credit risk models, also known as advanced approaches – echoes that sentiment: “The transition to standardised will definitely eliminate a substantial volume of modelling and analytical work, and may even reduce effort in the data management segment.”

Advanced in retreat

Credit risk models are not going to disappear completely. Banks still need them to make decisions about economic capital – and regulators may yet require the largest banks to continue running internal models, even if those models are not used to set capital requirements. Some banks are already thinking about how to create efficient credit risk modelling that meets both regulatory and internal requirements.

Global Credit Data (GCD) – a loss data pooling initiative designed to help banks estimate the two key inputs for IRB: probability of default (PD) and loss-given-default (LGD) – is one of the flowers that bloomed under Basel II. Its executive director, Richard Crecel, has yet to see any signs of alarm among US banks.

“We are even onboarding new banks from the US,” he says. “That tells me either they’ve not noticed that they might not be obliged to calibrate internal models any more in the near future, or – which I think is more interesting – they are using the data and the initiative for other purposes than prudential capital calculations.”

He adds that even standardised approaches may require some of the data that has been collected for IRB models.

The transition to standardised will definitely eliminate a substantial volume of modelling and analytical work, and may even reduce effort in the data management segment

Brett Ludden, Sterling Point

However, some credit risk modellers are more pessimistic. A senior capital manager at an advanced approaches bank says it will be “very difficult” to persuade business lines to invest in the data and processes needed to produce capital numbers that will not be “remotely binding” if IRB is abolished in the US.

“The focus is going to be on the most binding metric, which is going to be standardised with no models,” says the capital manager. “That is what’s going to drive your day-to-day decision-making.”

Courtney Davis, a principal in the risk practice at consultancy Deloitte, says boards are already trying to prioritise investments and work out “how much headcount they really need” for Basel compliance purposes. He adds that it is not a straightforward decision, and that there will be wide variation between banks: “Institutions are taking different approaches, whether it’s trying to build things in-house or leaning more heavily on third-party vendors.”

No substitute

IRB is far from the only function in US banking that credit risk models can be used for. They are still needed for the current expected credit loss (CECL) accounting framework and for the annual comprehensive capital analysis and review (CCAR), which is used, in turn, to calculate regulator-mandated stress capital buffers. “By and large we recycle IRB data for other purposes,” says a treasurer at an advanced approaches bank. “There’s some know-how in there that is used to train the models for CCAR. There’s some information in there that is used for CECL-type analysis.”

Davis says banks have already been seeking synergies in their risk data architecture to bring down the cost of compliance: “It’s fair to say that institutions, as much as they can, are leveraging Basel models, data and so on to enable stress-testing.”

Ludden also expects CECL to drive further investment in credit risk modelling. In particular, the lifetime loan loss provisioning required under CECL can lead to greater balance sheet volatility during economic downturns, so banks will want to have a good analytical handle on credit loss forecasts. He adds that “strategic advantages exist for banks who identify credit risk trends first”.

Richard Crecel, GCD

Crecel says GCD’s members are using its pooled data “more and more” – not only to project provisions, but to accurately price the sales of non-performing loans.

The senior risk modeller says some banks’ credit risk teams have grown to accommodate the needs of CCAR and CECL, but CCAR is not a straight substitute for IRB because the methodologies are very different. Stress-testing concerns capital depletion under scenarios hand-picked by the US Federal Reserve. By contrast, advanced approaches are built on an individual bank’s overall expected credit loss distribution, and are not connected to specific scenarios for economic growth or unemployment.

“For CCAR, each projection you are making is your expected loss in that environment,” says the risk modeller. “But the reality is it could be higher, it could be lower – there’s statistical error around it.”

In theory, it might be possible to adapt CCAR to reproduce the kind of analysis derived from IRB by running what the risk modeller calls a “CCAR Monte Carlo simulation”. This would project losses over the nine quarters captured in the Fed’s stress tests for 10,000 different economic scenarios to give the full distribution. However, the modeller believes this would be “unrealistic, at least with current computing technology. Maybe in 10 years you will be able to do it”.

The senior capital manager points out that the data inputs needed for a CCAR are also rather different from those used to calculate an IRB approach. The main projections for the latter are PD and LGD, together with the exposure at default; by contrast, CCAR leans heavily on pre-provision revenue models.

“I would argue that those PD and LGD models that are used in advanced approaches are not useful any more in this world,” says the senior capital manager. “Banks will just adapt to standardised metrics.”

Whither economic capital?

One important unknown is how far banks and regulators will want IRB calculations to survive in some form for risk management purposes, even if they are no longer part of the regulatory capital framework. In theory, banks may still wish to use internal credit risk models to inform decisions around capital consumption and allocation – so-called economic capital calculations.

As long as IRB models are in use for regulatory capital, banks see limited value in running economic capital models as well, because these fulfil essentially the same function but do not feed into regulator-mandated capital ratios. If IRB were scrapped and banks considered the standardised approaches too crude and unrealistic to be useful tools for measuring and managing capital consumption and risk, the senior capital manager says there might be an opportunity to revive economic capital models. However, the case for doing so will still be dampened if regulators are unsupportive.

“For the past five-plus years, models and economic capital have not been part of discussions with the US regulators,” says the senior capital manager. “They are not focused on them, and that’s why a lot of people think they are going to do away with the advanced approaches. In that world, would supervisors now push for economic capital models again, and expect people to have their own models and methodologies internally?”

Related

Fed official confirms US targeting 2025 for Basel III adoption

A-IRB to lose credit risk reach under Basel III

He thinks the new framework is more likely to lean towards the stress capital buffer as the risk-based component of capital requirements.

However, the treasurer at the advanced approaches bank says the Fed will never completely ignore economic capital modelling, and draws a parallel with the so-called qualitative objection element in CCAR. This allowed the Fed to fail a bank not because of the losses it would suffer under the stress scenario, but because of poor testing methodology or data management. The objection was scrapped for most firms in 2019.

“The inherent examination that would otherwise contribute to the qualitative pass/fail does still exist,” says the treasurer. “It’s just not a public shaming. You still get matters requiring immediate attention and you could get consent orders, so it’s not like there’s no focus on that stuff. It’s just that, from a public domain perspective, the details aren’t there.”

The treasurer adds that although regulators may be less interested in the outputs of economic capital models, they are still paying attention to how modelling is conducted overall, especially for CCAR – and they want to know how risk model outputs feed into overall business decisions.

“It’s about whether you understand the vulnerabilities,” says the treasurer. “Are you adhering to best-in-class model governance? Are you applying conservatism in the absence of certainty?”

Perverse incentives

Tobias Becker, head of business development at risk technology vendor Quantile and a former capital specialist at Credit Suisse, says US regulators indicated last year that pro forma calculations for the internal model method for counterparty credit risk might need to continue, even if the Fed were to preclude the method from being fed into regulatory capital requirements. The method’s credit component is built on the IRB approach.

Becker says this requirement to maintain internal models would most likely be directed at the largest banks only. However, he worries about how well modelling will be conducted if the capital incentives are missing.

Tobias Becker, Quantile

“You have to show that you are doing more than just standardised calculations – which, on the one hand, makes sense,” says Becker. “But I think that’s a dangerous thing, because when the regulator tells you to do something, you will say: ‘OK, I have to do this.’ But you are not really investing any brain power in it. You are not really improving it – you are just trying to hit the bare minimum that your regulator tells you.”

Ludden says IRB’s removal would level the playing field between the advanced approaches banks and the smaller lenders on standardised approaches. However, like Becker, he warns it would not necessarily lead to sound risk management.

“Banks who are [currently] superior in their modelling capabilities will lose a competitive advantage,” says Ludden. “There may be unintended effects whereby banks change their credit policies, which could even lead toward riskier lending.”

IRB is far more granular than standardised approaches, in which two loans with inherently different risk profiles could end up in the same risk bucket: if the capital cost of both is the same, a bank might choose to issue a riskier loan with a higher spread to maximise its returns.

The treasurer at the advanced approaches bank tries to sound a positive note while reflecting on the work that has gone into IRB and the consequences if it were to be removed from the US capital framework: “It’s not been completely futile, but it would be unfortunate.”

Ratings substitute needed

Even if internal models for credit risk are removed from the US regulatory capital framework, there will still be room for risk managers to use their own rating scales in the standardised approaches. This is because the US does not permit banks to deploy corporate credit ratings compiled by external ratings agencies as part of the standardised approaches.

Under the standardised credit risk methodology completed by the Basel Committee on Banking Supervision in December 2017, corporate loans receive a risk weighting of 100%. However, in jurisdictions that forbid the use of external ratings, this can be lowered to 65% for companies deemed investment-grade. The banks themselves make this assessment, and must show the regulators how they reached their conclusions.

If US regulators implement the Basel III standardised approach faithfully, IRB inputs such as PD and LGD may well be needed as part of the process for assigning 65% risk weightings to high-grade corporates. GCD’s Richard Crecel says he expects banks to pool their data to show they have a consistent idea of where the boundaries lie between investment-grade and speculative credits.

“When trying to monitor the risk profiles and the efficacy of their internal rating scales, banks are happier to use consistent data such as this internal data collected by GCD,” he says. “It’s useful because banks anyway want – and have regulatory obligations – to compare themselves and to benchmark their risk metrics against market views.”

Additional reporting by Sharon Thiruchelvam and Tom Osborn

Editing by Daniel Blackburn