Japanese megabanks shun internal models as FRTB bites

Isda AGM: All in-scope banks opt for standardised approach to market risk; Nomura eyes IMA in 2025

All three Japanese megabanks caught in the country’s early roll-out of Basel III have shunned the internal models approach (IMA) for calculating market risk capital under new trading book rules, citing high costs, complexity and a lack of data.

The mass abandonment of the IMA is creating concerns around behavioural changes – such as herding into crowded trades – that could result from most banks using a single standardised approach instead. The lack of banks seeking to use the IMA comes as a surprise to one senior regulator in Japan.

“We didn’t have an exact number of banks in mind, but did not imagine a situation where one model ruled the world,” said Tomoki Tanemura, deputy director-general at the Bank of Japan.

Tanemura was speaking on a panel at the International Swaps and Derivatives Association’s annual general meeting in Tokyo today (April 17).

The so-called Fundamental Review of the Trading Book requires banks to calculate their market risk capital requirements by using either their own models – provided they pass a series of reliability tests – or a regulator-set standardised approach (SA). Mizuho, Mitsubishi UFJ and Sumitomo Mitsui, which became subject to the new requirements on March 31, all opted to use the SA. Estimates by Isda, specifically for the US market, suggested if banks purely adopted the SA, it could double capital requirements compared with the previous regime.

We didn’t have an exact number of banks in mind, but did not imagine a situation where one model ruled the world

Tomoki Tanemura, Bank of Japan

Japan’s megabanks aren’t the only ones to witness a wipeout of internal models as the FRTB takes effect. All five of Canada’s largest banks, which implemented the FRTB in January this year along with the rest of the capital reforms drafted by the Basel Committee on Banking Supervision, are currently only using the SA – though one is understood to be considering a switch to the IMA. Tanemura said he hopes Japanese banks will follow suit.

“With the industry’s initiatives coming in and more data, I hope that the monster will disappear and it may increase the number of IMA banks,” he added.

That looks set to be the case in June 2025, when the Japan Financial Services Authority will apply its own version of the Basel III capital reforms to securities houses.

“We are approaching Basel very faithfully, that’s what the JFSA wants us to do, but we are going to be [an] IMA bank,” said Eduardo Epperlein, global head of risk methodology at Nomura, speaking on the same panel at the Isda event.

However, Daiwa is set to stick with the SA when it becomes subject to the FRTB next year, according to people familiar with the bank’s plans.

Dark arts of the P&L test

Nomura’s pioneer status in part reflects the greater importance of its markets business, as well as its more active role in exotic products such as derivatives. Risk Quantum data from the fourth quarter of 2023 shows more than 36% of the bank’s total risk-weighted assets stemmed from market risk, compared with well below 10% for the three megabanks.

“IMA is a huge cost for a small benefit,” said one model risk manager at a Japanese bank on the sidelines of the event. “If you’re talking about a portfolio of corporate bonds, IMA is not so beneficial for that.”

The IMA under Basel III represents a significant departure from the previous Basel 2.5 regime, which saw banks use two value-at-risk measures – estimating maximum losses according to current market conditions and under stressed conditions – to calculate capital requirements. Under the FRTB’s IMA, risk is calculated using the more computationally intensive expected shortfall metric, which estimates the average distribution of returns beyond a certain confidence level.

Jacques Vigner, chief strategic oversight officer for global markets at BNP Paribas, warned banks face a tough choice between the standardised approach that does not align risk capital and exposures, or an internal model fraught with complexity.

“The standard approach is intended to align the real risk with the capital measurement, but here it’s not present,” said Vigner. “Or you go for the internal model approach and then you become Harry Potter because you have to learn wizardry.”

He pointed to the unpredictability of trading desks being able to pass one of the gateways to the IMA. The profit-and-loss attribution test consists of two statistical measures that assess the correlation between the P&L of front-office pricing and risk systems, as well as the similarity in their distributions.

Related

Why Canada is giving FRTB internal models the cold shoulder 

Regulators’ FRTB estimates based on faulty premise – industry study 

FRTB could put Indian banks at competitive disadvantage 

“You need to do a lot to clean up your P&L prediction, you need to add up your risk factors and your risk models, it’s a lot of work and it’s not really granted that banks will succeed,” Vigner added.

Desks which fail these attribution tests could be shifted onto the standardised approach with immediate effect.

“Trading desks that are eligible for models one day, could be off those models the next day because of poorly calibrated model qualification standards,” said Vigner, adding that this danger could cause banks to pause plans for adopting the IMA.

Despite the concerns, Vigner urged the industry not to revert to “the stone age of risk management” with a widespread shift back to standardised models. He said internal models are better because they “stick to the risk”.

“Whenever capital is at odds with the risk, the bank is not well managed, because traders make the wrong decisions – they look at the capital but the risk is different,” said Vigner.

NMRF monster

However, while internal models are intended to align the measurement of trading book risk more accurately with risk management decisions, some speakers at the Isda conference were sceptical whether this would play out with the FRTB. In particular, banks are alarmed about the impact of non-modellable risk factors (NMRFs) on the IMA capital requirements.

A risk factor is deemed non-modellable if the bank’s data falls below one of two thresholds: 100 verifiable price observations over a one-year period, or 24 verifiable prices over the past year, as long as no 90-day period had fewer than four prices. These NMRFs are subject to a separate calculation and attract a stressed capital charge. According to Sebastian Crapanzano, global head of fixed income business unit risk management at Morgan Stanley, the outsized impact of NMRFs on the overall capital calculation could undermine the risk sensitivity of the IMA.

“The way the proposal is set up, there are a lot of non-modellable add-ons and restrictions for diversified business models, which actually makes the model-based approach look more like the standardised approach,” said Crapanzano, who was also speaking at the Isda AGM.

A risk manager at a second Japanese bank told Risk.net the lack of available data to reduce the number of NMRFs was a significant factor in the widespread adoption of the SA.

Nomura’s Epperlein also singled out what he referred to as the “monster NMRF” impact as a concern with the IMA.

“It’s bigger than the whole capital calculation from the expected shortfall,” said Epperlein.

To avoid large surcharges, banks can artificially bolster price observations on infrequently traded risk factors via more active trading in specific instruments, or purchase data from a third-party vendor.

“You’re actually buying the probability for your risk capital [to] go down, even though your risk has not changed,” said Epperlein. “It’s a very weird charge and probably the only charge that is not in [proportion] to your risk in the usual sense.”

Editing by Samuel Wilkes