Stress testing to determine potential losses during times of market turbulence has become part of everyday life for most central counterparties (CCPs) over the past decade. But there remains a wide variety of opinion on the best way to balance in-house risk management requirements with the expectations of regulators and customers when it comes to building the scenarios that help assess those.
One desirable property of a good stress-testing scenario is its plausibility; stress tests are plausible when they are based on scenarios that are likely to happen, since losses based on an implausible scenario would be of no use for assessing real risks. Being both plausible and stressful are acknowledged by CCPs and regulators as difficult criteria for a scenario to meet.
In this month’s first technical, Extremely (un)likely: a plausibility approach to stress testing, three senior executives from LCH – Mohamed Selmi, the head of risk methodology at LCH in Paris; Pierre Mouy, a senior risk analyst within the same team; and Quentin Archer, the head of equities business risk at LCH in London – propose a technique to produce coherent and plausible scenarios for CCP stress testing
In order to achieve this, the quants, under a general elliptical distribution, derive a set of joint moves of the risk factors – or scenarios – for a given real or hypothetical portfolio.
They then define a loss up front that represents a once-in-X-years return event – that is, the stress level – and then optimise and retain the scenario leading to that loss with the highest density under the assumed distribution – that is most plausible.
They also use a large number of scenarios to make sure they are well diversified, so that the addition of new risk factors or products into the universe of scenarios do not produce widely different losses when portfolios remain unchanged. This makes sure the scenarios are stable, and do not keep changing as new risk factors or products are added.
“If I come up with 20 scenarios that I can use, I am pretty sure that if a member is exposed to a combination of risk factors, if my scenario set is sufficiently diversified, there will be one specific scenario that will hurt them – so I’m not missing a risk,” says LCH’s Archer.
One regulatory source says this is an important property for a CCP stress-test scenario.
“The CCP needs to be able to explain what it’s doing to its members; it can’t turn around every month and say ‘we are doing a completely different set of scenarios’. It needs to have some way of figuring out a set of [scenarios] which would last for some reasonable period of time,” he says.
Though current regulatory stipulations around CCP stress testing lack uniformity, plausibility is held up as a key tenet in overarching guidance to regulators; in particular, the 2012 Principles for financial market infrastructures (PFMIs) from the Committee on Payments and Market Infrastructures (CPMI) and the International Organization of Securities Commissions (Iosco) state CCPs should be required to stress test according to scenarios of “extreme but plausible market conditions”.
The European Securities and Markets Authority (Esma) ran its first EU-wide CCP stress test in 2015, and will be completing a second test later this year. The US Commodity Futures Trading Commission carried out its first CCP stress test last year.
Discussions around greater standardisation of CCP stress testing are ongoing: CPMI and Iosco established a working group to look at CCP risk management practices and announced a formal review in 2015.
Detractors say standardisation will create model risk and prevent innovation in risk management practices, since stress-testing methods vary considerably across CCPs.
Given the systemic importance of CCPs, it is important that scenarios used to stress test them are chosen properly and do not produce unstable results or implausible losses. They should also be able to cover all relevant exposures.
These considerations are especially important when designing stress tests for CCPs at the regulatory level.
For instance, earlier this year market participants criticised the liquidity test under Esma’s EU-wide CCP stress test for being based on very unrealistic market moves
The authors of the paper show how scenario generation for CCP stress tests can be improved and made to meet certain key properties – realism being one of them.
The week on Risk.net, September 8-14, 2018Receive this by email