Final FRTB internal model rules get mixed reviews

Bankers divided on whether changes to two key tests will ease ‘penal’ capital charges

The BIS, home to the Basel Committee on Banking Supervision

Photo: Ulrich Roth

The Basel Committee on Banking Supervision has lowered the hurdles for two key modelling tests in its market risk capital standards. But bankers are divided on how far the final amendments will improve the Fundamental Review of the Trading Book, with some warning that a failure to fully take on board industry criticism will still make certain aspects of the tests too difficult to pass.

“We welcome the Basel Committee’s revisions of the FRTB standards, particularly on non-modellable risk factors and the profit-and-loss (P&L) attribution test, which were key areas of focus during the consultation phase,” says Panayiotis Dionysopoulos, head of capital at the International Swaps and Derivatives Association.

“The overall impact needs to be fully assessed, with a specific focus on areas where no changes have been made. But elsewhere, the changes appear to address many of the shortcomings of the current rules, which would have disadvantaged banks’ trading book activities,” he adds.

On January 14, Basel published final amendments to the 2016 FRTB standards, after reopening them for consultation in March 2018. Banks must calculate capital by either using their own internal models or a regulator-determined standardised approach, with the latter generating higher capital requirements.

In the final standard, Basel has relaxed thresholds in the P&L attribution test, which trading desks must pass to use the internal models approach (IMA), as well as the thresholds for a further test to prove whether risk factors are modellable. Banks had lobbied Basel to ease both tests as they would have resulted in significant increases in capital requirements, but there are divided views on where the final version has landed.

“It is a welcome improvement, but is far from what the industry was hoping for,” says a senior modelling expert at a global investment bank.

The P&L attribution test measures the accuracy of risk model estimates of P&L across trading desks. In the 2016 version of FRTB, if a desk failed the test, it was immediately switched from IMA to the standardised approach. Basel’s consultation paper, published in March last year, introduced an “amber zone” to act as an intermediate phase, with capital for a desk in the amber zone still calculated under IMA but with a surcharge.

It is a very good outcome with the calibration of the zones in the P&L attribution test

Adolfo Montoro, Deutsche Bank

Initial findings from banks suggested the amber zone would be “almost useless” in acting as an intermediate phase, because slight perturbations in good models caused desks to jump straight from the green to the red zone in two of the statistical approaches being considered for the P&L attribution test: the Kolmogorov-Smirnov (KS) method and the Chi X method. The behaviour was also observed in a third method, the Spearman correlation, to a lesser extent.

Industry associations including Isda and the Global Financial Markets Association urged Basel to reset the thresholds after assessing the new methods using data from real trading desks.

In the end, Basel has chosen to drop the Chi X method and use only the KS method for determining the distance between the P&L values of the front-office pricing and risk models. The Spearman correlation method will be used to determine whether the P&L values of the front-office pricing and risk models move in the same direction.

Wider parameters

Parameters for the amber zone in both the KS and Spearman correlation methods have been pushed back and widened.

“It is a very good outcome with the calibration of the zones in the P&L attribution test,” says Adolfo Montoro, director in the market risk management and risk methodology team at Deutsche Bank. “The thresholds have been loosened. In our paper we put forward a solid analytical business case to increase the zones’ thresholds slightly more, but I think they have extended quite materially. They have expanded the size of the green zone and materially expanded the amber zone, increasing the P&L attribution leniency.”

The senior modelling expert recognises the new parameters are an improvement, but says they fall short of what the industry had been hoping for, which was for the amber zone to be set roughly where the new red zone starts.

A capital manager at a UK investment bank says the final calibration appears “penal”, but notes he is still undertaking assessments of the new parameters. “Though the thresholds have been fixed for P&L attribution, it remains unclear whether the tests will be effective and see desks passing.”

“It will be very important to observe [changes to the amber zone parameters] in practice – once banks have built their models – to see if the calibration is accurate,” says an industry source. “This is an area where we believe final calibrations can only be made after we have seen the data.”

Seasonal changes

A second test in the IMA, used to determine whether banks can sufficiently model risk factors, has also been dialled back. Risk factors deemed non-modellable (NMRF) in this test must be capitalised separately with a stressed capital surcharge.

The previous framework required banks to have at least 24 real price observations of a risk factor over a 12-month period, with no longer than a one-month gap between two observations.

Banks have long complained that the seasonal nature of trading meant a high number of risk factors would be classed as NMRFs due to there being gaps of longer than a month in trading activity, which would subsequently increase bank capital under IMA.

Now banks can select one of two criteria to assess whether a risk factor is modellable. The first requires banks to have at least 24 real observable prices over 12 months, with no 90-day period having fewer than four observations.

If a risk factor is unable to meet those requirements, banks can still model the risk factor if it has a total of 100 real price observations over 12 months. The response of bankers towards the changes is varied.

“We think the 100 observations, but more particularly the 24 and four-in-90 observations, will make a big difference to modellability, which is positive and helpful,” says the capital manager at a UK investment bank.

I think this is an attempt to avoid people saying the one-month gap is causing the issues, whilst not dropping the original standard. It will address seasonality for only a few cases

Senior modelling expert at a global investment bank

Four observations over a period of 90 days would allow for a longer window of time between two separate observations. A consultant working at a European bank says: “We believe a lot of risk factors that used to be considered potentially non-modellable will become [modellable] with these new criteria.”

But the senior modelling expert at the global investment bank says the changes will not resolve the problem of seasonality. Industry associations and several investment banks have recommended using the approach laid out in the first criteria, but with three observations over the 90-day period.

“We didn’t expect it to solve much, but we did expect some flexibility before the standard was published,” says the senior modelling expert. “I think this is an attempt to avoid people saying the one-month gap is causing the issues, whilst not dropping the original standard. It will address seasonality for only a few cases.”

The senior modelling expert believes the second test will only relieve the seasonality problem for the most liquid risk factors and is therefore too high a bar.

“The 100 observations is a form of addressing seasonality in liquid risk factors, but it is an extreme one. The Basel Committee has basically gone, ‘Fine, [if] you think that [risk factor] is liquid, then show it.’ But 100 seems dramatic – it should be more than 24, but not [so] much bigger,” the expert adds.

Insufficient evidence

The senior modelling expert believes the Basel Committee did not go further in alleviating the seasonality issue because the standard-setter says it didn’t receive much evidence to support industry claims.

“It has been really challenging for the industry to pull together compelling evidence on the risk factor eligibility test,” says Jacob Rank-Broadley, a director at technology vendor Refinitiv. “The industry was basically being asked to get a huge breadth of data, boil it down and then run some relatively simple tests. It is much easier to do that in relatively small quantities. It is also quite difficult when a lack of data is your problem to evidence a lack of data objectively. Apart from saying ‘we are struggling to find that’.”

Basel has also inserted a rule within both criteria for passing the modellability test that no two observations should be from the same day. The intention is to avoid banks potentially gaming the test.

“We got some questions from some of our clients saying, ‘Can I have 13 observations on one day and then one observation per month and still pass?’” says Rank-Broadley. “Technically, with the old rule, you would have been able to pass. We didn’t find the scenarios existed in reality. All of those hypothetical scenarios just didn’t come up very often.”

Rank-Broadley says eliminating this loophole is a useful step, because banks might otherwise have explored whether they could use it, before finding out there was no benefit in doing so, given that the scenario is so rare.