After bruising EU model review, banks ask: ‘Why bother?’

Post-Trim changes erode capital savings from internal models while raising their running costs

  • The ECB’s “targeted review of internal models” at 65 large EU banks has led to a 12% aggregate increase in statutory capital, with ABN Amro, Commerzbank and ING among the hardest hit.
  • Changes to models made in the wake of the review have eaten into the savings available from using internal approaches to RWA calculations and raised the costs of the models. 
  • As a result, a number of banks are questioning whether in-house approaches are worth persisting with. Some banks in Germany that were thinking of using internal models have now changed their minds, according to a consultant.
  • Lenders argue their models’ capital outputs will now be too high as regulators were excessively “conservative” during the review – for example, when tackling loss given default and the incremental risk charge.

In the film Groundhog day, TV weatherman Phil Connors gets stuck in a time loop, reliving the same day over and over again. For one bank executive, a five-year regulatory review of banks’ internal models felt just like the film. Risk managers had to provide a mountain of data and work through hundreds of calculations, with requests from the European Central Bank (ECB) seemingly going over old ground as the years went on.

“They could have put the [capital] add-ons in place with tenfold less intrusiveness,” says the head of models at a mid-sized European Union bank. “Banks only need to be taught a lesson once. We don’t need it 15 times.” 

A project that was meant to ensure capital models devised by large banks are fit for purpose and consistent with regulations has left many lenders exasperated and questioning the value of using internal models at all.

The so-called targeted review of internal models, or Trim, did not only pile extra work on certain teams at banks. It has resulted in a €275 billion ($324 billion) or 12% aggregate increase in risk-weighted assets (RWAs) across 65 banks, according to an April report by the ECB. Disclosures by ABN Amro, Commerzbank and ING show they were among the hardest hit.

At many banks, the changes to models prompted by Trim have eroded the savings available from using in-house approaches to capital calculations, while at the same time making internal models costlier to maintain.

Lenders argue that the models’ outputs will now be too high because regulators were unnecessarily “conservative” in their assessments, be it in their calibration of certain key parameters such as loss given default (LGD) or in their pick of distributions to test model choices.

Models already require significant investment – in quantitative analysts to build, backtest and maintain them, model risk managers to oversee and validate them, and systems and processes to capture the necessary data.

Non-stop work

Trim was a gargantuan exercise involving members of the ECB, national supervisors and an army of consultants meeting risk managers at 65 large EU banks in 200 on-site meetings.

As part of the review, one mid-sized EU bank had to run between 800 and 1,000 calculations to test the outputs of its internal models, according to its head of market and credit risk.

The head of models at another mid-sized EU lender says: “The ECB requested more information than we have ever provided and then on top of that we had to provide an additional dataset for our national regulator.

“We had teams that were really working non-stop.”

“We have to justify all our investments to shareholders, and our management board is not in its element when discussing stochastic processes,” quips the head of credit and market risk at a mid-sized EU bank.  

Trim, which started in 2016, has already discouraged some banks in Germany that were considering switching to internal models from doing so, according to a consultant based in the country.

Although there are some banks that see value in the exercise, others are disillusioned and suspect the ECB of bias in favour of raising capital levels.    

“Looking at the outcome of Trim, I get the impression that the ECB is very proud of the additional billions in RWAs they imposed on the banks,” says the head of model risk at the mid-tier EU bank. “The increase is even bigger than we anticipated from the implementation of Basel IV – to me, there is kind of an issue there.”

When the latest update to the Basel standards was finalised, the European Banking Authority (EBA) estimated that the average global systemically important bank in Europe would end up with a 15.2% rise in the minimum required Tier 1 capital. The extensive revision of Basel III, which has prompted banks to dub the latest version Basel IV, comes into force in January 2023.

The head of models at a large EU bank echoes the sentiment of his counterpart at the smaller lender, calling the jump in capital caused by Trim a “self-fulfilling prophecy”. 

“Not worth it” 

The stated purpose of the landmark review was to check the rigour of internal models and their alignment with regulations, as well as make sure there is no undue divergence between the RWA outputs of banks whose models and portfolios are similar.  

The ECB and national supervisors examined models across credit risk, counterparty credit risk and market risk. More than 90% of the €275 billion increase in statutory capital was due to undercapitalisation of credit risk. Aggregate RWAs for the other two risks were also nudged up.

The Germany-based consultant says the headline sum may hide the true scale of the capital hike because RWAs have been raised at a time when long-run measures of risk are falling

“Long-run averages of actual defaults and losses have been on the decline for a long time before Trim in most European economies,” he explains. 

He adds that not all EU banks have fully implemented all model changes related to what is known as the EBA’s IRB repair’ programme that “brings additional conservatism with it”. The project centres on internal models used to calculate own funds requirements for credit risk under the internal ratings-based, or IRB, approach.  

“This could mean that the conservative direction in which the ECB has taken models is not fully visible,” the consultant concludes. 

For banks, the conservatism calls into question the future of internal models. 

“The amount of effort to maintain internal models is enormous, and if it does not mean big savings, maybe senior managers will think about going back to the standardised approach,” says another consultant, Elias Bustillo, who specialises in risk and capital management for financial institutions at Alvarez & Marsal.

This is even more of a risk, he adds, when the overhaul of Basel III takes effect, because it will limit savings from internal models by introducing a floor on their outputs. The floor will cap internally calculated capital at 72.5% of RWAs that are based on standardised approaches in 2028 when it is fully phased in. 

“The output floor will penalise banks in the Netherlands, for example, because they have lots of mortgage portfolios with very low risk,” Bustillo says. “Just 25% savings against the standardised approach in these cases is not worth it.” 

The question of whether Trim will have an impact on the popularity of internal models is addressed by the ECB in an FAQ webpage accompanying its final report on Trim. 

“With Trim, we did not intend to persuade or dissuade banks from using internal models, but rather to assess how adequate the models in use were,” the bank writes. “Nonetheless, Trim, together with the changes envisaged through the finalisation of the Basel III standards, may have consequences for where and how internal models are used by banks.”

LGD conservatism

Banks say the output of internal models will now be higher and closer to the upcoming Basel floor partly because regulators were excessively cautious in their review – for instance, when calibrating the LGD parameter.

The parameter, which asks banks to predict roughly how much they will recoup if a counterparty defaults on a loan, is notoriously difficult to calibrate, with cyclical adjustments necessary to consider how bad defaults would get at the height of an economic crisis compared with a period of economic stability.

The global head of risk models at the large EU bank points to charts 10 and 11 in the Trim report, which show the distribution of the average LGD at the assessed banks before the review and after it – that is, after a simulation of the impact of resulting supervisory measures (see charts). 

“You can clearly see that the distribution of LGD has dramatically shifted upwards and [for large corporates for the unsecured portfolio] the range below the median is a lot smaller than before,” says the head of models. “It seems like a floor has been systematically introduced to LGDs. This is, of course, a way to reduce RWA divergence, but it’s a very blunt instrument.”

Charts 10 and 11 from the ECB’s ‘Targeted review of internal models project’ report
TRIM SIs’ denotes the “significant institutions”, or the large EU banks, that were subject to a Trim investigation for large corporates or for financial institutions including the LGD model respectively

He adds that he has seen “relatively little evidence” that banks have been systematically underestimating LGD.

The Trim report states that for credit risk models related to retail portfolios and lending to small and medium-sized enterprises, in 95% of the cases where LGD was reviewed, there was at least one “high-severity” finding in relation to this parameter. The report also notes that calculation of realised LGD was a frequent cause of compliance issues.

The ECB’s conservative approach to LGD was on display in its treatment of Spanish banks, according to Bustillo, who is based in Madrid. Since Spanish banks ended up with a high number of non-performing loans after the financial crisis, they often sold these portfolios at below-market rates to quickly improve their capital ratios – often with encouragement from regulators. The ECB, however, took the selling prices as an indication of high risk of the lenders’ portfolios. 

“The ECB says that if you sold the portfolio at a certain price, this must be the LGD. They don’t take into account that the banks needed to improve their capital ratios quickly,” Bustillo says. 

“I think that some level of conservatism is good, but maybe this is too much. And at the end of the day, this can mean that the models are not really internal models, due to all the imposed floors.” 

For its part, the ECB can point to the 2013 Capital Requirements Regulation, which contains this line on internal models: “Where methods and data are considered to be less satisfactory, the expected range of errors is larger [and] the margin of conservatism shall be larger.”

Risk.net understands that the banks that complained about having a floor put under their model outputs faced a choice between that or losing the right to use internal models altogether. The ECB presented them with the choice because it found too many weaknesses in their in-house approaches. 

Abundance of caution

Another example of what banks see as regulators’ extreme prudence is their stance on probability of default (PD), which feeds into the calculation of the incremental risk charge (IRC). This charge represents the risk of issuer default or the risk of a credit downgrade for unsecuritised credit products in banks’ trading books.

A senior risk expert at a large EU bank says his bank was instructed to assume a PD for tradable credit instruments such as bonds that was above zero and at least 1 basis point. 

“All of a sudden, somebody’s coming in and saying: ‘You know, you’ve been understating the default probability for high-grade sovereigns,’” he says. “That’s a big statement.”

The explanation given as to why the PD could not be lower than 1bp was the language in the 2019 ECB guide to internal models, the person notes. The guide states: “The IRC model must give a meaningful differentiation of risk, and accurate and consistent estimates of incremental default and migration risk. Therefore, the ECB understands that all annual PDs should be risk-sensitive and greater than zero for all obligors. In this context, the term ‘greater than zero’ is interpreted to mean greater than, or equal to, 1 basis point.”

When it comes to the central bank’s approach to the IRC, some banks also criticise its choice of a Student-t copula to test IRC calculations, for two main reasons. 

One is that this type of distribution is not suitable for all banks. 

“The modelling choices you make are dependent on your idiosyncratic characteristics as a bank,” says the head of models at the mid-sized EU bank. “If you try to have a reference that is one-size-fits-all, that can only be academic.”

Another reason is the difficulty of applying a Student-t copula.

“There were several difficulties in implementing this because it requires several important decisions to be made regarding parameterisation, and no-one [among supervisors] was talking about these details,” says the head of credit and market risk at the mid-sized EU bank.  

The modelling choices you make are dependent on your idiosyncratic characteristics as a bank. If you try to have a reference that is one-size-fits-all, that can only be academic
Head of models at a mid-sized EU bank

There may be a third reason – not mentioned by the banks – why the requirement to use a Student-t copula proved unpopular. This distribution has fatter tails than the Gaussian distribution, giving more weight to risky outcomes, and so it is more conservative. However, when the ECB told lenders to use a Student-t copula to test their IRC models, it was following its own advice in the guide to internal models.  

Market risk managers detect eagerness to err on the side of caution in supervisors’ approach to counterparty credit risk models, too. The senior risk expert at the large EU bank points to the treatment of Effective Expected Positive Exposure (EEPE) models, which measure changes over time in the amount a bank expects to lose, should a counterparty default on a trade. 

He outlines a hypothetical example, in which an interruption in cashflows on a trade causes its present value to drop, prompting an equivalent rise in margin demands from the bank to cover the change in exposure. 

So far as the EEPE model is concerned, “this brings about a ‘spike’ in exposure”, he says, since if the counterparty defaults, it assumes the collateral payment will not be made. In reality, he argues, “for daily margined trades, this risk is very short-lived – essentially, one day – but since EEPE, which is the quantity that enters the capital calculation, is by definition non-decreasing, the exposure will stay at the level of the spike, and will not drop back down”.

The non-decreasing nature of EEPE is meant to reflect the fact that portfolios will roll over – for instance, when a counterparty rolls a derivatives trade into another monthly or quarterly contract at expiry. But if there has been any previous change in trade flows that affect the contract’s present value, the dealer’s counterparty’s exposure remains at the same elevated level.

One of the peripheral benefits of such an approach, proponents point out, might be to incentivise banks to negotiate collateral agreements that allow for the netting of coupon payments close to or at the same time as variation margin calls are made, in order to avoid exposure spikes at the legal and operational level.

However, the banker notes that, while this would “go some way to eliminating the problem”, it would require a retooling of standardised documentation between counterparties. “It would be a big change for over-the-counter derivatives markets, and the implications are not fully clear at this stage,” he says.

Thanks, but no thanks

While Trim has reduced the benefits of using internal models, it has also made them more labour-intensive, dimming their appeal compared with standardised approaches.

For example, the ECB has told model validators to work harder to make sure models continue to be appropriate as underlying data changes. This should theoretically make the models more robust during crises, but banks say it has added to the complexity of developing and monitoring models and made them costlier to maintain.

Some banks have already made their choice in the wake of the Trim exercise.

The Germany-based consultant says he knows of some German lenders that were toying with the idea of using internal models, but, as the review of their larger peers progressed, they decided against it. 

The additional scrutiny from the ECB evident during Trim has also prompted them to find “creative” ways to avoid direct supervision by the institution, he adds. The banks’ tactics include restructuring their legal entities to fall beneath the size threshold that determines whether the ECB considers the bank “significant” and therefore requiring direct supervision. Significant banks are those with more than €30 billion in assets. 

But some think the doom and gloom about the review is overdone. 

Despite some quibbles, the senior risk expert at the large EU bank says that, overall, it was a useful exercise, adding: “It was an opportunity for us as well to look at what we are doing well and what maybe we are not doing so well.”

And a model risk expert at a large US bank warns that EU lenders should set their sights elsewhere. 

“I think the Trim review maybe overshoots a little bit, but it’s not an existential issue,” he says. “Whereas the Basel output floor may well be.” 

The ECB declined to respond to the criticisms. 

Additional reporting by Tom Osborn, editing by Olesya Dmitracova

Only users who have a paid subscription or are part of a corporate subscription are able to print or copy content.

To access these options, along with all other subscription benefits, please contact info@risk.net or view our subscription options here: http://subscriptions.risk.net/subscribe

You are currently unable to copy this content. Please contact info@risk.net to find out more.

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here