Paper of the year: Peters, Shevchenko, Hassani and Chapelle

Operational Risk Awards 2017: Authors provide rigorous proof of SMA’s pitfalls and suggest ways to improve existing approach

Bertrand Hassani and Ariane Chapelle

Paper of the Year: Should the advanced measurement approach be replaced with the standardized measurement approach for operational risk? By Gareth Peters, Pavel Shevchenko, Bertrand Hassani and Ariane Chapelle

For much of the past year, debate in the operational risk industry has centred around the Basel Committee on Banking Supervision’s proposals for a new standardised measurement approach (SMA) for operational risk capital, unveiled in March 2016. The SMA would replace the existing advanced measurement approach (AMA), which allows banks to build their own models, and its two simpler alternatives.

While the previous three-pillared approach to calculating operational risk capital was far from perfect, the SMA proposal is widely considered deeply flawed. Risk managers and researchers have shown that the new approach would produce capital requirements that are disproportionate to the underlying risk and overreact to tail losses and differences in the size of banks, leading them to plead for the retention of the AMA.  

This year’s winning paper, which was published in the Journal of Operational Risk in September 2016, set out to study the pitfalls of the SMA and to determine whether it could feasibly replace the AMA. It was not the only time the topic was covered in the publication, but a team of experienced authors presented a detailed stress test of the SMA together with a set of proposed remedies, which put the paper well ahead of its peers.

The paper is the work of four authors: Gareth Peters, a lecturer in statistics at University College London (UCL); Pavel Shevchenko, at the time a researcher at Australia’s Commonwealth Scientific and Industrial Research Organisation, and now a finance professor at Macquarie University; Bertrand Hassani, global head of research and innovation at Santander; and Ariane Chapelle, a risk consultant and honorary reader at UCL’s computer science department.

The team used simulated loss histories for banks of different sizes to show how, under the SMA, operational risk capital levels could double from year to year – or differ by a factor of two from bank to bank – despite there being no change in the underlying risk profile.

They pointed out that SMA capital figures would be backward looking and unresponsive to improvements in operational risk management, as the only inputs are business size and internal loss history. Further, the approach’s lack of granularity when compared with the AMA, which recommends splitting loss data into 56 categories, means it lacks risk sensitivity. And though it provides a capital incentive for large banks to break themselves up, which theoretically improves stability by addressing the ‘too big to fail’ issue, the resulting smaller banks could be significantly undercapitalised, which would increase systemic risk.

“[SMA] is proven to be problematic if it is adopted,” says Peters. “It doesn’t resolve issues associated with standardisation in operational risk modelling practice, and it adds new challenges and issues due to its problematic specification. It fails to utilise operational risk data in most data classes and is too simplistic in its specification to be meaningful.”

The SMA has faced a barrage of criticism since it was published for consultation, with practitioners and some regulators calling for it to be dropped. It remains to be seen whether the Basel Committee will bow to industry pressure, but Chapelle believes this paper has provided “more rigorous proofs that others maybe do not [and] more substantiated criticism and more detailed recommendations”.

“The feedback we have received from banks and regulators unofficially… is that they believe SMA will require substantial changes if it goes ahead,” adds Peters.

So, what’s next for the SMA? Chapelle believes regulators are unlikely to implement the approach in its current form, noting that many of its most prominent supporters – Daniel Tarullo for one – have since left the Bank for International Settlements and the US Federal Reserve. But the authors believe calls to discard op risk modelling, or even op risk capital itself, are misguided. They advocate an improved AMA as an alternative to the SMA.

The AMA was criticised for being too subjective, with comparisons between banks nearly impossible due to the dependence on idiosyncratic internal models, assumptions and data collection procedures. “There is a general agreement that AMA has failed – not in its intention but in its implementation – by allowing too much leeway and therefore too much bad practice or reverse engineering,” says Chapelle.

Instead, the authors suggest compelling banks to use specific models – generalised additive models for location, shape and scale – for severity and frequency in each of the 56 Basel categories, and specifying a set of business environment indicators and control factors to include in each one. Operational risk modelling has advanced a great deal in the 15 years since the AMA was first proposed, Chapelle argues, making a prescriptive approach much more practicable.

Both academics and practitioners have welcomed the paper’s proposal, according to the authors. Academics favour the move towards more standardised models, while practitioners like the opportunity to preserve the granularity of the AMA, as they feared that without a clear regulatory requirement to keep collecting loss data at a detailed level, their budget to do so would be at risk.

“There was a lot of criticism of the failure of AMA banks to incorporate key risk indicators,” Shevchenko adds. The authors’ proposal would provide a new incentive to include them – another advantage in the eyes of operational risk practitioners.