Remembering Lehman: CCPs hardwire collapse into models

Five years on from the collapse of Lehman Brothers, the chaos that followed is now being erased from some value-at-risk models – and clearing houses do not agree on how to prop up their margin requirements. By Tom Osborn

Photo: David Shankbone/Wikimedia

With their cardboard boxes and dazed expressions, the employees of Lehman Brothers slunk across US television screens again during September, marking five years since the investment bank’s collapse. Countless retrospectives made it a difficult anniversary to forget – but for any value-at-risk model based on five years of historic data, the turmoil following the September 15, 2008 bankruptcy filing is now being erased, day by day.

Banks and central counterparties (CCPs) are free to choose a so-called lookback period for their VAR models – which calculate trading book capital and initial margin requirements, respectively – and while banks often settle on a period of between one and three years, some still have a five-year lookback. The industry’s regulators have responded by hardwiring the Lehman collapse into other capital metrics (see box, Stressed VAR: Basel 2.5’s forget-me-not). CCPs currently have more freedom – LCH.Clearnet switched to a 10-year lookback earlier this year, but CME Group currently uses a five-year period, and Eurex uses three years of data. 

On the face of it, that could produce some big differences. “For the most part, initial margin requirements are calculated using 10-day VAR – in other words, it’s directly affected by how much rates can move during a 10-day period. In September 2008, they moved a lot – a 40-basis-point shift was not unheard of. As we’ve lost a couple of those 10-day periods from five-year historical lookbacks, there can be a significant impact – no question,” says David Kelly, director of financial engineering at Calypso Technology, which provides risk technology to CCPs including CME Group and Eurex.  

“You’ll get crazy effects with a five-year model right now, as the back end of it rolls through the crisis, up to the end of January 2014,” says Christopher Finger, executive director for applied research at analytics firm MSCI in Geneva. As each turbulent day falls out of the model’s 1,250-day memory, the scenarios it generates become progressively less violent. According to analysis run for this article by software vendor SunGard, margin requirements could drop by as much as 80% for shorter-dated swaps when using a lookback period from which the Lehman collapse and its aftermath is removed (see box, Life without Lehman).

Looking for ways to avoid this effect has dominated the risk agenda for some CCPs this year, says Finger, with some clearing houses choosing to emphasise certain periods in a shortened time series – known as filtering – while others have extended their lookbacks to retain the post-Lehman stress. “We’ve seen people shorten their windows and choose to smooth things out using filtering, or take the opposite tack and move to a 10-year window to keep Lehman in,” he says. 

You’ll get crazy effects with a five-year model right now, as the back end of it rolls through the crisis

In Europe, CCP regulation is also responding. Technical standards written by the European Securities and Markets Authority (Esma) and approved by the European Commission in December 2012 offer clearing houses three ways of mitigating pro-cyclicality in their margin models. The first is a buffer that can be run down in times of stress; the second is a form of filtering that would give more weight to stressed periods during the chosen lookback. The third option is a floor, in which a CCP must ensure margin requirements “are not lower than those that would be calculated using volatility estimated over a 10-year historical lookback period”. 

That last option has some advantages, says MSCI’s Finger: “When we’ve run the numbers on any CCP backtests, results are still dominated by whether you are including 2008 and the beginning of 2009. Moving to a 10-year lookback doesn’t eliminate that problem – it just postpones it. But if the lookback is longer, days dropping out have less weight relative to the total window, so it will, depending on the weight a CCP’s volatility filtering gives it, have a smaller effect.” 

LCH.Clearnet became the first major over-the-counter derivatives clearing house to double its lookback period to 10 years, or 2,500 trading days, for its SwapClear service – a decision motivated by a desire to stabilise future margin levels, says Dennis McLaughlin, the CCP’s group chief risk officer. “When Lehman leaves the time series, we could see a fairly large drop in initial margin. That would mean we’d be prone to pro-cyclicality concerns – in essence, a sudden increase in margins if another seismic event were to occur. We don’t want that, which means we need to switch to a longer lookback. For SwapClear, we’ve already decided to do this. That recalibrates the initial margin by the appropriate amount. We’ve done the calculations, and if we keep Lehman in the time series, the margins would be virtually unchanged from the level they are today.”

CME Group, which is competing for a share of the interest rate market currently dominated by SwapClear, may take the same path. “There’s a consensus we need to keep the events of five years ago in the mix,” says Sasha Rozenberg, managing director of risk management at the exchange. “Esma’s pro-cyclicality rules effectively gave us a choice of three methods of adjusting our initial margin calculations, and we chose what we felt was the most appropriate. We’ve made our internal decision, but it still needs to be approved by our interest rate swap risk committee. For CME Clearing Europe, that proposed change formed part of our reauthorisation application to Esma.”

In a neat coincidence, the deadline for these applications – required by the European Market Infrastructure Regulation – was September 15. 

The length of a lookback is not just about the size of initial margin numbers, but also their stability. Over a longer period, the impact of gaining a day and losing a day matters less; losing one day from a 100-day period means a 1% change in the dataset every day. For a 2,500-day lookback, it is 0.04%. The resulting change in margin numbers depends on the weight applied by a CCP’s volatility filter, but the general principle is the same for all of them, says MSCI’s Finger. 

Not everyone is convinced a longer time series is the answer, however. Sinan Dikmen, head of risk methodology and new products at Eurex Clearing, argues that shifting to a 10-year lookback just results in insensitive margin requirements and postpones the inevitable period when the crisis starts to fade from the model’s memory. Instead, the CCP has a three-year lookback, but uses a stressed period as a floor when calculating margin requirements. This currently includes the post-Lehman period, he says. The bourse is also able to make on-the-fly modifications to its margin models to adjust for correlation breaks, taking account of rate moves dislocating from historical trends. That, says Dikmen, gives Eurex greater control over the margin it can demand from clients in all market conditions. 

“Whatever the length of your lookback period, you will always get stressed episodes dropping out. You can extend the lookback, but the information then grows less relevant. For that reason, we adopted a more advanced methodology in which a shorter, three-year rolling lookback using dynamic volatility filtering is floored with permanent stress-period scenarios, and Lehman is one of them. Over a long period, there may be more stressful events – but only then would Lehman be kicked out. Not only is expanding your lookback period not as effective, but it also creates additional operational burdens. What are you going to do in five years’ time? Increase the lookback to 20 years?” he says.

Other changes CCPs have made to their initial margin models have been dictated by members rather than regulators. In December, Risk reported that LCH.Clearnet was recalibrating SwapClear’s margin model to measure interest rate moves on an absolute, rather than relative, basis in response to member concerns that the model’s responsiveness was being deadened by the ongoing low-rate environment (Risk December 2012, page 8). Following the initial recalibration, the clearer embarked on a series of other changes – of which switching to the 10-year lookback was one. In tandem with that switch, it moved from a VAR to a conditional VAR model, also known as expected shortfall, which its proponents see as a better measure of tail risk because it takes the average of VAR numbers generated above a given confidence level.

The changes were an attempt to find a margin model that is sensitive to risk, without being too pro-cyclical, participants say. “Contrary to VAR, expected shortfall looks at all the scenarios above a desired percentile to determine the average tail loss. As such, it mitigates the cliff effect observed when an extreme scenario drops out,” says Alex Bon, head of credit risk management products at Paris-based risk technology vendor Murex, a supplier to LCH.Clearnet.

The model has already had its first test during a period of heightened volatility, when fears that the Federal Reserve would wind down its quantitative easing programme triggered a bond market sell-off in May and June. Margins increased, but the model did not overreact, says a clearing specialist at one member firm. This is exactly what the CCP was looking for, adds Dan Maguire, head of SwapClear US and global head of product management. “We recalibrated our initial margin model earlier this year to ensure its performance in all rate environments. May and June provided a stiff test for the revised model, but we’re delighted it maintained a steady and appropriate level of initial margin rates for members and their clients.” 

The changes had less to do with dealers’ systemic risk concerns than with rising awareness of the costs involved in client clearing, says one senior credit risk expert. “The shift from relative to absolute returns at LCH.Clearnet was driven by member lobbying. Generally, the big clearing banks don’t like default fund contributions to be high. They much prefer that initial margin covers as big a portion of the risk as possible, because they can pass that along to their clients more easily. So generally, that’s what they lobby for,” he says. In response, LCH.Clearnet points out that relatively few SwapClear members currently have live client clearing offerings.

Given the variety of approaches and the impact they can have, some are asking whether CCP margin modelling will become as regulated as capital modelling for banks. Last year, Edwin Schooling-Latter, head of the payments and infrastructure division at the Bank of England, ruffled feathers when he intimated that CCPs had long escaped closer scrutiny of their margin models on the grounds that such information was sensitive intellectual property. This secrecy would not be acceptable in the era of mandatory clearing, he said. 

Others agree further inspection could be on the cards. “I’m not sure we’ve seen the end of regulatory scrutiny of CCP models. My impression is there has been relatively little of that to date – and there has been reluctance on the part of regulators, particularly in the US, to mandate fixed lookback periods or other modelling specifics. I find it unlikely you’d have a prescriptive model for US CCPs. What I do think we’ll see is regulators ramping up their own validation teams to check up on CCPs, with a view to tighter risk monitoring,” says MSCI’s Finger.  

Transatlantic consistency, meanwhile, may come about without official intervention if clearing houses apply the more prescriptive European regime to their business in the US or elsewhere – and that seems to be the plan. LCH.Clearnet aims to provide uniformity to members and clients across all cleared OTC markets, meaning SwapClear US will implement the same changes as the firm’s European-domiciled entities. 

CME, too, is keen on consistency. “The rules in the US are not prescriptive, so we don’t currently have to employ one of the three specified pro-cyclicality adjustments here. We use other means to manage pro-cyclicality, such as volatility floors. We have additional features that reduce the size of margin breaks when they occur. But US and European regulations may start to converge, and we’d like to be consistent in our approach,” says Rozenberg. 

 

BOX: Stressed VAR: Basel 2.5’s forget-me-not

The value-at-risk models used by banks to calculate trading book capital requirements typically use a lookback period of between one and five years, with most European institutions clustering around the one- to three-year mark, prudential regulators say – meaning the Lehman default has long since been erased from most standard risk measures. But under Basel 2.5 – the package of new risk measures that was implemented in Europe from December 31, 2011 – banks also face a so-called stressed VAR requirement, which is based on the one-year period that is most painful for each institution. In practice, that means a 12-month dataset that includes the Lehman Brothers collapse.

Giovanni Pepe, head of financial risk models assessment in the banking supervision department at Banca d’Italia, hopes that will be the case for some time. “The period of sustained volatility surrounding the Lehman default was unique for a number of reasons: we saw extreme stress in interbank lending rates, but also in equity market values and credit spreads. That would be difficult to replicate using data from another period. I have yet to see a bank argue that Lehman should not be used,” he says.

An exception might occur if a bank’s trading portfolio was dominated by a particular type of asset, says Pepe. For instance, if a regional lender’s books were largely made up of sovereign bonds, the eurozone crisis of 2011 could be considered, but that won’t be the case for multinational banks of any size, Pepe adds. 

So when will banks be allowed to forget the crisis? It may not happen until regulators unveil a brand new trading book regime, which is a work in progress, says Christopher Finger, executive director, applied research at analytics firm MSCI in Geneva (Risk April 2013, pages 26–28). “In general, national supervisors have been reluctant to be prescriptive. The posture has very much been ‘explain why you chose the period you did’. So it’s not set in stone. But think about why the stressed VAR calculation exists – principally it’s there to make the risk-weighted asset numbers bigger, but it’s also there to make them more stable. Given stability is the order of the day, I think regulators will be reluctant to see the 2008–2009 period drop out any time soon. Does that mean we have to wait for Basel IV before 2008 rolls out of the memory completely? Well, maybe so.”

For banks, it currently feels a remote possibility. “From what I’ve heard, many regulators have pushed back against any bank considering changing that stress period for some portfolios,” says a credit risk expert at one European firm.

 

BOX: Life without Lehman

It’s impossible to say exactly how initial margin requirements will react as the Lehman Brothers episode is wiped from the memory of a value-at-risk model – it depends how the model is implemented at a given clearing house, and on the specifics of an individual portfolio. But software vendor SunGard provides an illustration, showing the impact on a single position of a sudden switch from a model that includes the Lehman collapse to one that does not (see figure 1). 

The example is for a central counterparty calculating initial margin requirements for a one-month at-the-money forward rate agreement, using the expected shortfall across the six worst-case five-day holding period scenarios, and volatility-weighted absolute returns on each tenor of the US dollar curve. 

The effect is striking. “The magnitude of the change in margin as the Lehman default drops out of the lookback period depends on the portfolio. A portfolio with sensitivity primarily to the short end of the yield curve – which responded most dramatically to the default – can see a reduction of up to 80%, while reductions for longer-duration portfolios are typically more of the order of 20–30%,” says Andrew Hudson, credit risk and credit valuation adjustment product manager at SunGard.