Trading desks want regulators to face down the NMRF monster

Rule-makers in Australia and the European Union are open to changes to the unpopular FRTB test

Three guardians were charged with ensuring only accurate and reliable trading risk models could be used to calculate capital requirements under revamped bank prudential rules. Each guardian presents aspiring banks with tests they must pass to demonstrate the strength of their modelling.

One of the trio, however, is overzealous in its duty and wreaks havoc on all banks wanting to use their own models. As local rule-makers become receptive to curbing the powers of the guardians, some think regulators should give up trying to tame the wildest of the three, which will otherwise deter many banks from even attempting the internal models approach (IMA).

“It has become its own live monster by itself,” says a capital manager at a global investment bank. “I’d say we collect the existing elements in capital rules, see how they can be enhanced and see if that achieves what we want to achieve with this.”

The monstrous test is known as non-modellable risk factors (NMRFs). Banks wanting to use their own models to calculate capital requirements under the Fundamental Review of the Trading Book (FRTB) must assess if every risk factor within the model has enough real price observations to evidence its movements. If banks have too few data points for a risk factor, it is deemed non-modellable and must be capitalised with a stressed version of the IMA’s main capital metric, the expected shortfall.

The NMRF test has been controversial since the concept was first mooted by the Basel Committee on Banking Supervision more than a decade ago, with many illiquid risk factors often falling short, prompting banks to complain of unwarranted capital add-ons. Those arguments are now cutting through with policymakers.

Australia’s Prudential Regulation Authority was the first regulator to signal an aversion to the NMRF rules. In a corporate plan published in August 2024, the regulator instructed banks to prepare for the FRTB, but not the NMRF rules. Sources say that doesn’t necessarily mean the rules will be scrapped in Australia, but they could be revised. Apra hasn’t put forward any ideas for how the rules should be altered.

Meanwhile, the European Commission published for consultation on March 24 a report into future actions it could take with the European Union’s version of the FRTB. The EC has concerns its January 2026 start date could put EU banks at a disadvantage to US rivals that won’t be subject to the same rules. One option is to make temporary changes to the rules alleviating pain points, including the rules of NMRFs.

For some, the effort isn’t worth making. Instead, three risk experts argue regulators should retain existing qualitative assessments of missing risks in models as well as restoring older quality assurance tests. However, others think there are good reasons to introduce the NMRF regime, and it just needs fine-tuning.

“It’s a legitimate concept and a manageable problem,” says a market risk expert at a bank that is already operating under the FRTB. They add the NMRF concept is “fundamental to the IMA” and a “substantial part of the framework”. The bank in question has adopted the standardised approach rather than the IMA.

These hurdles are essentially prohibiting firms from applying for the IMA, even if they want to

Gregg Jones, Isda

Resolving NMRF burdens could be crucial to entice banks back to the IMA. Many have been avoiding the approach due to how expensive it is to develop and run the machinery. The cost is prohibitive if the IMA only produces small capital savings compared to the regulator-set standardised approach (SA). There are ways to reduce the number of NMRFs, but they also come at a price.

The NMRF rules are often cited as a culprit that narrows the gap in capital requirements between the two approaches. Of the 26 banks responding to a survey published by the International Swaps and Derivatives Association in July last year, 18 pointed to the NMRF rules as a deterrent to the IMA.

“Each part of the NMRF rules individually offers layers of conservatism and, in aggregate, it is a very conservative process, which makes it super challenging and costly for firms to pass,” says Gregg Jones, a senior director at Isda. “These hurdles are essentially prohibiting firms from applying for the IMA, even if they want to.”

The Basel Committee declined to comment for this story.

Created with good intentions

The reason the Basel Committee first engineered the NMRF rules was to ensure banks have adequate data to model their risks. People say the genesis of the monster comes from the UK Prudential Regulation Authority’s concept of assessing risks not reflected in banks’ old market risk capital models, which were built around the value-at-risk measure. The PRA would use the results of their risks not in VAR (RNIV) exercise to inform the setting of discretionary capital add-ons.

RNIV tasked banks and supervisors with judging the extent to which models were not fully reflecting risks. The market risk expert at the bank applying the FRTB says the way regulators approached this tended to differ globally, which is one reason why they want to see a reformed version of the NMRF rules. The FRTB’s innovation was to develop a universally recognised quantitative threshold that can apply in all jurisdictions for determining whether a risk factor is backed by appropriate data. The idea of creating a uniform approach to risks not in VAR was inherently a good one, says Eduardo Epperlein, global head of risk methodology at Nomura.

Eduardo Epperlein

“Risks not in VAR was more like a discovery process: you realise [there is] a missing time series or missing risk factors and you put an add-on,” says Epperlein. “Unfortunately, even though it had a good initial motivation to formalise risk not in VAR, you end up with a monster.”

The final Basel version of the FRTB published in 2019 establishes two thresholds, and risk factors need only meet one to be deemed modellable. The two are both over a one-year period: either 100 real-price observations, or 24 observations with no 90-day period having fewer than four prices.

The requirement for 24 real prices is connected to a key setting within the FRTB’s version of expected shortfall, according to the market risk expert at the bank that has adopted the FRTB.

The FRTB requires banks to scale the shock applied to risk factors based on the Basel Committee’s perception of how long it takes to exit the risk type. Banks must scale the shock to the number of days prescribed in the FRTB. A 10-day base horizon for exiting the position is established for banks to determine how large the shock to a risk factor is meant to be.

The raw implementation of the NMRF rules can lead to [add-ons of] 100%, or even more, of the equivalent VAR model. There needs to be a reality check: surely you can’t be missing all that risk

Eduardo Epperlein, Nomura

Banks must measure observations of price movements in risk factors over the last 12 months. The calendar year for each jurisdiction varies in the number of working days, but regulators often assume there are 250 trading days in a year. Ensuring there are at least 24 evenly distributed real price observations to risk factors over a year will therefore likely lead to one observation falling within the 10-day base horizon.

The market risk expert at the bank applying the FRTB says banks need an “anchor” for the shock banks are trying to refer to in their modelling, so regulators don’t want the base horizon observation period to be “empty”. Requiring 24 evenly spaced observations across 250 trading days should ensure there is at least one observation in every 10-day period.

Illusory precision?

Not everyone agrees the quantitative construction of the test is wholly necessary, however. Without extensive data, banks will make their own hypothesis as to a risk factor’s behaviour, which can be based on observed movements in closely related risks. The hypothesis could be wrong, but so could a model driven by observed data, because the past is not a reliable predictor of the future even for the same risk factor.

“Requiring spot-on precise historical series is pointless once you realise that the future is anyway different from the past,” says Carlo Acerbi, founder of consultancy Risknowledge. “Insisting that you can only model ‘modellable’ risk factors that pass the eligibility test betrays the wrong conviction of the regulator that there exists one right way to model, and this right model is the historical simulation of past events.”

The market risk expert at the bank applying the FRTB, however, argues the use of proxies can become subjective when there are no observations to keep a risk modeller “grounded”. They say the use of these proxies can lead to “significant variability and arbitrariness” in risk-weighted assets (RWAs) – the key input in determining risk-based capital requirements.

When the limitations of proxies, the need to ensure a meaningful 10-day observation period in expected shortfall, and the variability in the old RNIV method are all put together, “you can see why this FRTB innovation is a reasonable attempt to improve the status of these problems,” says the market risk expert.

Still, the total cost of NMRFs has shocked many banks that considered adopting the IMA. Risk managers complain that models that passed previous adequacy tests for years are suddenly found to have extensive missing risks.

Insisting you can only model ‘modellable’ risk factors that pass the eligibility test betrays the wrong conviction of the regulator that there exists one right way to model

Carlo Acerbi, Risknowledge

A recent impact study published by the Basel Committee on March 26 found NMRFs accounted for 21.3% of the overall capital requirement from the IMA, which also consists of the charge for modellable risk factors and a separate charge for the risk of securities issuers defaulting. Epperlein estimates the existing regime typically yields 10–20% extra capital charges from risks not in VAR.

“The raw implementation of the NMRF rules can lead to [add-ons of] 100%, or even more, of the equivalent VAR model,” says Epperlein. “So, there needs to be a reality check: surely you can’t be missing all that risk.”

Taming the NMRF

The FRTB’s journey to implementation has reached a turning point. Regulators seem receptive to the idea of changes and the NMRF rules are one area coming under review. Some are suggesting ways the rules could be reformed to make them cheaper, and so encourage wider IMA adoption.

Risk.net understands Isda is preparing a paper setting out ways the FRTB could be revised to boost the number of banks applying for IMA approval, which should include easing the NMRF rules. Isda has already suggested ways to make NMRFs more agreeable in the past. The industry group’s response to US draft proposals to implement Basel III – which were released for consultation in July 2023 but faced a backlash afterwards – included suggestions for lowering the capital impact from the rules.

Related

Tug of law: EC’s FRTB compromise plan can’t please everyone 

Leaked draft consultation unlikely to reconcile the global versus regional bank divide

“We are working with the working groups to address the issues with NMRFs in the FRTB,” says Panayiotis Dionysopoulos, head of capital at Isda. “All options are open and under discussion as part of the European Commission’s consultation on market risk that is currently ongoing.”

Published in January 2024, one suggestion in Isda’s response was for NMRFs to be capitalised in the main expected shortfall measure if they meet seven principles laid down in the FRTB for assessing the modellability of risk factors. Those principles set out essential characteristics for the data that banks use in internal models, including proxies being limited and sufficiently similar, and for the data to be updated at least monthly. Banks must assess the data for their modellable risk factors against the principles, and supervisors can determine a risk factor to be non-modellable if the principles are not applied. Isda recommended NMRFs that meet the principles should be included in the main expected shortfall measure, but with an additional liquidity surcharge.

The idea of a liquidity surcharge already has a basis in the FRTB, which sets liquidity horizons for specific risk types based on assumptions for the number of days it would take an institution to exit the position. There are five buckets differing in the numbers of days, from 10 to 120. After the initial calculation of expected shortfall using all modellable risk factors, banks strip out the risks with the shortest liquidity horizon and perform the calculation again. The process is repeated until they perform the last calculation containing only those risks in the final fifth bucket. Banks then add the result of all the calculations afterwards.

Panayiotis Dionysopoulos

Photo: Isda

Isda suggests NMRFs meeting the modellability principles can be pushed up a bucket to fall within a longer liquidity horizon than they are assigned in the FRTB, with the final bucket acting as a cap.

Some suggest going further, so that the data principles alone replace the quantitative thresholds of the risk factor eligibility test as the primary means for determining whether a model is supported by adequate data.

“You can argue if you fail the data principles of getting good data, yes, you should have risks not in VAR,” says Nomura’s Epperlein. “So, we should go back to the original basic principles, and then you can retire the risk-factor eligibility test altogether because it’s serving very little purpose.”

A further recommended change Isda put forward in its consultation response was to lower a parameter representing diversification within the NMRF capital charge from 0.6 to 0.25. Isda’s response argues the current setting grants “inappropriately” limited recognition to the benefits of diversification. The consultation response estimates the change would shave off $40.1 billion from the amount of RWAs generated by the FRTB in the US. Isda estimates the version of the FRTB that US regulators proposed in July 2023 would have led to a total increase in RWAs of $277.8 billion.

The EC proposes a similar temporary relief measure in its consultation. Rather than lowering an existing parameter, however, it proposes introducing a new multiplier in the 35–45% range on the whole NMRF charge.

“[It’s] unlikely minor changes to the regime would make IMA attractive to us,” says a market risk manager at an investment bank active in emerging markets. Pointing to the EC’s recent proposal to slash the overall NMRF capital charge, they continue: “This is the sort of thing that might make us reconsider.”

I don’t think the NMRF should be kept at any cost; there is not necessarily something good to be saved here

Carlo Acerbi, Risknowledge

The EC also proposes to exclude newly issued bonds from the NMRF test. The trouble with newly issued debt is the very low likelihood banks will have enough observations to pass the test, since the bond would only just have become tradable. The problem is particularly acute for exposures to sovereign issuers, which regularly issue debt in the market and make up a sizable slice of banks’ trading books. Carving out new issuances from the test should further lower the NMRF capital requirement.

“For a new issuance, it can be very problematic, especially for very high-quality sovereign debt, which is very sensitive for every jurisdiction,” says Isda’s Dionysopoulos.

Trust the old guardian

Three sources, however, want to see the NMRF monster finally slain rather than tamed. Instead, they argue there are already established tests that could be relied upon or enhanced to adequately address the risks to models.

Two sources point to a long-established test of assessing the predictions of models against reality, known as backtesting. This is the only reliability test banks have to undertake under the old trading book capital rules. Each time actual one-day losses are larger than the forecast of VAR an exception is counted, which informs the setting of a multiplier on top of the output of VAR. The more exceptions, the higher the multiplier.

Related

BNP Paribas exec fears data drought from market’s IMA cuts 

Australian FRTB projects slow down amid scheduling uncertainty 

In the FRTB, banks using internal models must perform two backtests. The first assessment is done on all risks falling within the scope of a bank’s IMA perimeter. Similar to the old capital framework, the result of this test informs a multiplier. The second assessment is carried out on each trading desk that uses the IMA. If a trading desk experiences 12 exceptions or more over a one-year period, the desk loses permission to use the IMA and reverts to the SA.

“I don’t think the NMRF should be kept at any cost; there is not necessarily something good to be saved here,” says Acerbi of Risknowledge.

What matters, he believes, is that there is a good backtest in place. This means the model is tested correctly, along with its elements – namely, the data and assumptions that are fed into the model.

“If a backtest is there, you evaluate the result of the predictions of the model and if they are wrong, they may be wrong because of the data or because of the assumptions,” explains Acerbi.

The trouble with the two FRTB backtests is they may miss out gaps in models, which regulators are keen to pick up. Since the NMRF rule assesses every single risk factor, it provides a more granular assessment of modellability. However, the capital manager at the first European bank says an extension to backtesting can overcome this problem.

“If you can’t calibrate your model because your observations are not reflective, you should see it somewhere – something should overshoot somewhere,” says the capital manager at the global investment bank. “Maybe backtesting at group and bank level ignores many things at the asset class level, so maybe make backtesting more granular to spot those overshootings.”

Additional reporting by Nancy Qu. Editing by Philip Alexander