FRTB: banks grapple with hard-to-model risks

Swiss, UK bank efforts to comply with regulators’ risks-not-in-VAR rules may be undone by transition to FRTB

Image of risk factors

  • Basel policymakers met this week to discuss revisions to the identification and capitalisation of non-modellable risk factors (NMRFs), a key plank of the updated market risk framework.
  • A patchwork of approaches currently exist to tackle factors that fall outside of banks’ internal models. UK and Swiss banks are subject to a formal risks-not-in-VAR (RNIV) framework, used to apply capital add-ons for these factors.
  • Dealers subject to RNIV have worked in recent years to reduce these add-ons, by sourcing data sufficient to place these factors into their internal models and otherwise petitioning regulators to permit their inclusion.
  • However, the currently drafted NMRF rules introduce a highly conservative methodology for these factors that goes far beyond those laid down for RNIV.
  • This means banks’ gains tackling RNIV charges may be all but erased with the transition to FRTB.

This week, the Basel Committee on Banking Supervision convened its policy development group in Madrid, with a lengthy to-do list.

The publication of Basel III in December marked the end of a years-long effort to finalise post-crisis reforms of the regulatory capital framework – accompanied by a prolonged lobby battle by the industry. But for one aspect of the rules – the revised market risk framework – the battle is just beginning.

Critics claim Basel has yet to fully address concerns with two elements of the new regime, commonly known as the Fundamental Review of the Trading Book: the calibration of the profit-and-loss attribution test – intended to serve as the gateway to the internal models approach – and the capital charges connected to the knottiest of market exposures, known as non-modellable risk factors (NMRFs). Both are understood to be under discussion this week – though the committee declined to comment.

Basel’s blueprint for identifying and capitalising NMRFs has come under sustained fire from an industry that says it will increase market risk capital requirements far above current levels. Back in 2016, a survey conducted by a group of industry associations predicted that 29% of total market risk capital charges under FRTB would be attributable to NMRF, on average.

Bankers fear the ultimate impact could be much worse, however: “The studies we have at present probably underestimate the likely NMRF charge,” says Katherine Wolicki, head of regulatory strategy in global risk analytics at HSBC. “Certain assumptions were made in the early quantitative impact studies about what was modellable, but when you look at this in more depth you see that the real price observation criteria can be particularly tricky.”

One bank with a chunk of its trading portfolio concentrated in less liquid markets told their NMRF charge could account for as much as 50% of their total market risk-weighted assets.

The International Swaps and Derivatives Association, which had a hand in the 2016 survey, says that impressing on regulators the capital burden of NMRFs is a 2018 priority. In January, the association presented Basel’s market risk group, tasked with finalising FRTB, with technical data drawn from its membership to raise awareness of the various issues the NMRF component raises.

As an organisation, we didn’t want to go into FRTB running hot in the RNIV space – we needed to fix our data and enhance our methodologies where possible

Director in trading risk at a European bank

“NMRF remains a big area of focus for FRTB internal models implementation due to its large capital contribution. The industry remains committed to providing additional data and solutions to address the issues in the NMRF framework,” says Panayiotis Dionysopoulos, head of capital at Isda.

Basel is expected to release papers clarifying the NMRF and profit-and-loss attribution test frameworks before the end of the first quarter of this year. Unwilling to wait, some banks have been trying to find ways to reduce their expected inventory of NMRFs – by looking for clues to how these slippery factors are dealt with under current market risk rules.

Near and VAR

When FRTB was first floated by policymakers, certain stakeholders assumed hard-to-model exposures would be accommodated in a similar fashion as so-called risks-not-in-VAR (RNIV), a sub-category of market risk included among national watchdogs’ Pillar 2 frameworks to address those factors inadequately captured by firms’ internal models.

The UK and Switzerland have established formal RNIV frameworks, where these omitted risk factors are identified and capitalised as add-ons to the modelled market risk total. The UK’s watchdog, the Prudential Regulation Authority, developed its framework in Supervisory Statement 13/13, first issued in 2013, as part of its implementation of the European Union’s Capital Requirements Directive, which hands national authorities the discretion to tackle “material deficiencies” in internal model outputs. (See box: The unmodellables).

Dealers subject to RNIV capital add-ons have worked hard in recent years to clamp down on these factors, and move as many esoteric risks into their internal VAR and stressed VAR models as possible, with varying results (see figure 1). This has been done both to please regulators, who want banks’ internal models to accurately gauge the risks in their trading portfolios, and internal treasury managers, who want to optimise capital.

“Efforts to reduce RNIVs date back to 2015, when it became clear that FRTB was coming and we knew the direction of travel,” says a director in trading risk at a European bank. “As an organisation, we didn’t want to go into FRTB running hot in the RNIV space – we needed to fix our data and enhance our methodologies where possible.”

However, as the market’s understanding of NMRFs has grown, so the list of similarities to RNIVs has shrunk. Indeed, the message received from three banks subject to RNIV that spoke to for this article is that their efforts to bring these factors into modellability will largely be undone by the transition to the NMRF framework.

Banks say they will have to develop new strategies to tackle Basel’s fresh take on these risks. “There remains a general view that RNIVs will morph into NMRF. But it’s a bit of apples and pineapples. Given the consequences of how the NMRF framework is currently written, the pineapple is much bigger than the apple,” says the European director in trading risk. 

The same camp

Superficially, RNIVs and NMRFs appear to describe similar risk factors. Under the UK framework, RNIVs are described as “any risks which are not adequately captured by…models”, including absent, or illiquid, risk factors such as “cross-risks, basis risks, higher-order risks, and calibration parameters”. These factors are also described as RNIV by Credit Suisse and UBS, banks subject to Swiss regulator Finma’s market risk framework.

NMRFs are similarly defined as those factors that lack adequate data to be included in an internal model. The FRTB text states that risk factors that do not have a history of continuously available real prices – connected to trading instruments that rarely, if ever, change hands – must be classified as NMRF.

Here’s where the two diverge, however. Regulators do not prescribe the scope of an RNIV, instead permitting banks to circumscribe these factors at their discretion – though they reserve the right to label additional factors RNIV as they see fit.

As such, today’s RNIV are a mixed bag of factors, says James Balfour, head of market risk measurement and modelling at Lloyds Banking Group.

“RNIVs can be in place for a few reasons. First, there are those risk factors that are not appropriate to be included in a VAR model, for example if they are very illiquid and there’s no valid proxy. Second, there are low-materiality risk factors that may be excluded from the VAR model if the costs of including them outweigh the benefits to the model, from both a risk management and capital perspective. Third, there are risk factors that can be included in the VAR model but where an implementation lag exists until model development, internal governance and regulatory approval has been completed for the model change.”

The natural seasonality of markets means we have instruments showing 100, 200 observations a year but that fail the gapping rule because in August they don’t trade

Director in trading risk at a European bank

Efforts to reduce total RNIV have generally focused on the latter variety. Lloyds’ own Pillar 3 report for the first half of 2017 states that a “number of risks captured as RNIVs were moved into the VAR model in 2016”, and that “plans are in place to transfer a material proportion of the remaining RNIVs during the second half of 2017”. This migration of factors from RNIV to the VAR and stressed VAR models would explain at least part of the £133 million ($186 million) reduction in RNIV from year-end 2016 to the first half of 2017.

In contrast, the classification of an NMRF is hardcoded into the FRTB text. Only risk factors supported by 24 or more “real prices” in a given year, with no more than a month between consecutive observations, can be said to have a history of “continuously available real prices”. All others are NMRF. Under the RNIV framework, on the other hand, banks can make their own case for a factor to be deemed modellable and present this evidence for scrutiny by the regulator – which can then decide whether or not it can be included in the VAR models.

As a result of Basel’s prescriptiveness, hundreds of risk factors currently considered “liquid” by banks may find themselves thrown in the NMRF bucket. Once-named RNIVs moved into modellability may also be rejected under FRTB.

“We see the migration of things that were modellable going into the NMRF framework. We’re talking about standard stuff like credit default swaps,” says the director in trading risk. “The natural seasonality of markets also means we have instruments showing 100, 200 observations a year but that fail the gapping rule because in the month of August, for instance, they don’t trade.”

Worse, banks infer from the FRTB text that NMRF should be classified at a more granular level than RNIV are at present. Take the example of interest rate risk factors, where Basel identifies specific points on each relevant yield curve, including the 0.25, 0.5, 3, 10, and 30 year points as separate risk factors. This implies a non-modellable yield curve would have to be similarly divided into a collection of NMRFs, with each referencing a specific point.

Common practice under the existing RNIV framework, meanwhile, is to consider a hard-to-model yield curve as a single, combined risk-not-in-VAR. This subtle difference threatens to load huge capital costs on to dealers, because no correlation or diversification offsets between NMRFs are permitted. Each has to be capitalised separately, and the entire inventory added together to arrive at the total NMRF charge. The 2016 industry associations’ survey estimated the aggregate NMRF charge across its sample of banks would be 4.3 times that for RNIV.

Under the above example, capital would have to be put aside for a trading portfolio’s sensitivity to each point on the non-modellable curve. A short exposure to the 0.25 year point could not be offset by a long exposure to the 0.5 year point under this methodology, despite the fact that movements in the curve at these points would be highly correlated.

“The way FRTB defines NMRF diversification appears to be significantly stricter than under the current RNIV framework, as individual tenors on individual curves may need to be considered as separate non-modellable risk factors and capitalised separately. If that interpretation is taken it could be very punitive from a capital perspective, especially for some directionally hedged portfolios,” says Lloyds’ Balfour.

Put simply, banks may find singular RNIVs transform into multiple NMRFs under the new framework and be forced to hunt out additional data points to eliminate these.

Capital concerns

Identification is one difference; capitalisation is another. RNIV can be capitalised in one of two ways: using the same VAR metrics as those applied to risk factors fully incorporated in their internal models, or through application of a stress test.

The former methodology is used for RNIV where “sufficient data are available”, says the Prudential Regulation Authority: including those factors awaiting regulatory approval to enter the internal model, and those a bank has elected to exclude for their own reasons, such as factors that may be cost-additive to move into the model. The latter approach need only be applied to those lacking the data necessary to populate a VAR metric.

In practice, this means the capital implications of RNIV vary on a case-by-case basis. Often capital savings will be gained from moving an RNIV into the internal model. However, sometimes these gains can be offset – or even exceeded – by their inclusion.

“The capital charges can cut both ways,” the director in trading risk says. “In some instances, we get better data and enhance our methodologies – and move RNIVs into our models. But these risks will now appear in our VAR and stressed VAR numbers. Depending on diversification and portfolio changes, the RNIV component of our market risk-weighted assets may go down, but the VAR and stressed VAR could go up.”

The asymmetric interplay of RNIV and internal model capital has, historically, given banks the option of gaming the classification of their risk factors to produce the most favourable capital outcome. The loose identification criteria for RNIV has similarly allowed firms to pick and choose what counts in the RNIV framework.

“There are cases where banks have highlighted some risks as immaterial to their regulators, however upon challenge by the regulators the banks have then been pushed to implement RNIV models, or improve their RNIV model,” says a risk consultant and former head of market risk management at several international banks.

In contrast, all NMRFs must be capitalised using a stress scenario. This must be calibrated to be as conservative as that applied for modelled risks – specifically, an expected shortfall loss calibrated to a 97.5% confidence level – over a period tailored to a prescribed liquidity horizon for that factor.

Clarity is key – but not if the result is rules that are so dogmatic they make the capital figures totally misleading

James Balfour, Lloyds Banking Group

This capitalisation methodology, together with the “no diversification benefit” rule and highly granular assessment procedure, means that the charges for NMRF are expected to greatly exceed those for RNIV – and prevent the sort of capital gaming between modelled factors and RNIV that can occur under the current framework.

This may be the Basel Committee’s intention, of course. Portfolios referencing hard-to-model factors are being made more costly in order to nudge banks out of the business of bearing complex, illiquid risks – the sort that can tear through capital buffers.

“There is a clear incentive for banks and other market participants to improve the liquidity of these risk factors,” says Chris Cormack, managing director at MP Capital Advisory, a consultancy specialising in portfolio modelling and risk methodologies. “This potential consequence could drive a genuine improvement in value passed to all market participants, including retail customers. The impact on structured trades and hybrid [instruments] from the restrictions in NMRF is predicted to drive a significant reduction in the demands of these products.”

Initiatives are afoot, however, to allow these very products to survive under the new framework. Data pooling solutions, to facilitate the collection and distribution of pricing data that would enable more risk factors to qualify for inclusion in internal models, are under development by Bloomberg, Markit, DTCC and others.

However banks adapt, they cannot escape the fact that FRTB does away with an existing framework that gives regulators and firms flexibility in assessing and charging for complex risks, and replaces it with one where the rules are rigidly enforced and which loads on capital charges that may lead some firms to question whether an internal model approach is worth it.

Lloyds’ Balfour admits the Basel Committee has a delicate balancing act to maintain: “Guidance that is open to interpretation has the advantage of allowing banks and regulators to take a pragmatic interpretation. However if guidance is too loose, it can lead to excessive and ineffective debate as to what is really meant by it. When you consider that one of the key drivers of FRTB is getting more consistency across different banks on how they are capitalising for trading book risks, I’d say clarity is key – but not if the result is rules that are so dogmatic they make the capital figures totally misleading when compared to the underlying risks faced by the banks.”

The unmodellables

Estimating market risk exposure is a multi-layered exercise. Under the current Basel market risk regime as implemented by national regulators, Pillar 1 regulatory capital is calculated by internal-models firms using an array of proprietary tools and specially designed risk management frameworks.

Risk positions that are richly evidenced by historical market data are assessed using internal value-at-risk and stressed VAR models. An add-on to account for issuer default and credit migration risk within a dealer’s trading portfolio of bonds and credit default swaps is calculated using a bespoke incremental risk charge (IRC) model.

Esoteric risks – those rarely traded, or impossible to trade, or existing outside of general market risk categories – are captured using different methodologies in different jurisdictions.

US dealers apply a specific risk charge to cover event and idiosyncratic risks connected to debt, equity and securitisation positions. Some of these risks are modelled using tailored specific-risks models; others are assigned a standardised risk weight. To reduce specific risk charges, US banks need regulatory approval to move these into their existing internal models.

For European dealers, the Capital Requirements Directive, which represents the single market’s interpretation of Basel standards, hands competent authorities discretion to tackle “material deficiencies” identified in a firm’s internal model approach, such as risks not appropriately captured, through the application of capital add-ons. Member state regulators have interpreted this directive in various ways.

This mandate served as the catalyst for the UK’s RNIV framework, which the Swiss regulator, Finma, is said to have drawn inspiration from for its own iteration.

Pillar 3 regulatory disclosures shed light on how different UK banks sort their RNIVs. HSBC, for instance, includes “gap risk” inherent in non-recourse margin loans and “de-peg risk” connected to pegged and heavily managed currencies among those RNIV capitalised through the application of a stress scenario. Certain basis risks that are partially observable, such as Libor tenor basis, are classified as RNIV and capitalised using VAR metrics.

Finma sets its own parameters for RNIV and has the authority to approve or reject its supervisees’ set framework for identifying and capitalising these risks. Swiss dealers cleave closely to their UK peers in RNIV quantification. For example, UBS employs a mix of VAR measures and stress test approaches to capitalise individual RNIVs and uses statistical methods to aggregate them in order to produce a group-level, 10-day 99% VAR estimate of its entire RNIV portfolio.

  • LinkedIn  
  • Save this article
  • Print this page  

You need to sign in to use this feature. If you don’t have a 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 indvidual account here: