Banks struggle with Basel 2.5

Banks are gearing up to comply with changes to the Basel Committee’s market risk framework from the beginning of 2012. But risk managers complain the rules are far from perfect and leave too much open to interpretation. Mark Pengelly reports

All eyes have been on the completion of the Basel III package of measures, including details on calibration and the transition period for new liquidity and leverage ratios and a credit value adjustment charge. But changes to trading book capital and the market risk framework – known as Basel 2.5 – are causing more immediate work for risk managers.

The rules, which were finalised in July 2009 with an adjustment in June this year, require banks to make significant changes to trading book capital calculations ahead of a scheduled implementation from December 31, 2011 – a full six years before the more controversial elements of Basel III come into force. But bankers say there are still plenty of grey areas in the market risk framework, and are calling on regulators to give more guidance on key issues of implementation.

“These proposals were created in something of a hurry in response to the sense the overall capital for the trading book was too low and needed to be fixed, rather than a precise, careful analysis of the risks and an equally careful decision as to how to capitalise them,” says one industry source familiar with the Basel negotiations.

A first draft of the package was published in January 2009, and included details of a previously proposed incremental risk charge (IRC), meant to capture trading book losses due to default and migrations, and a new stressed value-at-risk measure, designed to force banks to consider the impact of a period of stress on their trading books (Risk February 2009, page 8).

As a result of the changes, the trading book capital charge for banks using the internal model approach would consist of a general market risk charge (and a specific risk capital charge if the firm had approval to model this risk) measured using a 10-day VAR at the 99% confidence level. Those banks with approval to model specific risk would also be subject to the IRC, measured at a 99.9% confidence level over a one-year capital horizon, taking into account the liquidity horizons of individual positions or sets of positions. Added to this was the stressed VAR, which would consider a 10-day VAR at a 99% confidence interval for a one-year period of significant financial stress.

Meanwhile, resecuritisations such as collateralised debt obligations of asset-backed securities were hit with higher risk weights. More generally, securitisation exposures were excluded from IRC modelling, instead receiving a much more orthodox banking book charge based on credit ratings.

These rules created problems for banks, which argued the loose definition of securitisation meant correlation trading desks would be unfairly penalised and certain activities would become uneconomical (Risk July 2009, pages 17–20). In response, a revised proposal was released in July 2009 that would allow banks to conduct their own modelling for correlation trading (Risk August 2009, page 9). But despite this concession, many remain unhappy with the proposals.

The new framework allows banks to develop a comprehensive risk measure (CRM), covering specific and incremental risk for their correlation books, rather than use the standardised approach for securitisations. To gain regulatory approval, banks have to jump through a number of hoops: CRM models must consider the cumulative risk arising from multiple defaults, including the ordering of defaults; credit spread risk (including gamma and cross-gamma effects); the volatility of implied correlations (including the cross effect between spreads and correlations); basis risk between an index and its single-name components; basis risk between the implied correlation of an index and bespoke portfolios; recovery rate volatility; the risk of hedge slippage; and the potential costs of rebalancing these hedges.

Furthermore, banks must have sufficient market data to ensure they fully capture the risks, demonstrate their models can adequately explain historical price movements and ensure they can separate positions that have modelling approval from those that do not. They also have to apply a set of specific predetermined stresses to the portfolio on a weekly basis, reporting the results to supervisors every quarter. And any instances where such tests indicate a shortfall in the CRM must be reported in a timely manner to regulators, which could then decide to add a supplemental charge.

The standards are demanding, says Balbir Bakhshi, London-based head of the market risk group at Credit Suisse. “The bar you have to reach for your correlation approval is quite high because you need to reprice everything, including spreads, timing of default, correlations, recoveries – everything. So it’s quite onerous from an IT and calculation perspective to get a more efficient capital amount at the other end,” he says.

Making matters worse, the Basel Committee on Banking Supervision confirmed in June that any bank approved to use its own internal models for correlation books would be subject to an 8% floor based on the standardised measurement approach – a decision that has attracted significant criticism from banks.

“The CRM is based on bank internal modelling, but since the Basel Committee was reluctant to have a charge solely based on bank modelling, it added the 8% floor. That makes a big difference, because the floor is not based on risk-based scenarios where you stress the market parameters and look at the worst downside you have on your book – it is based on what is called the standardised regulatory charge, which is only driven by the rating of the underlying risk,” says Benjamin Jacquard, London-based co-head of global credit trading at BNP Paribas.

The rating-derived standardised method would lead to punitive treatment of unrated tranches, while offering limited capital relief for hedges, dealers argue. For businesses that rely on offsetting long and short positions, the measure is simply inappropriate, they say. “If you ever have one tranche with 100 names and a similar tranche with 99 of those names, you are running a very low risk – but the standardised method will be very punitive to you because it will assume there is no netting effect,” says Jacquard.

Even though banks approved to use internal models would only be subject to 8% of this standardised charge, it would still be higher than the capital level determined by internal models in many cases, bankers claim. “In most cases, the 8% of the standard calculation is far above the CRM calculation,” Jacquard adds.

Ironically, banks would only be subject to the 8% floor if they obtain regulatory approval for their internal models. In other words, banks would need to plough money into model development, just to qualify for a standard 8% floor that is likely to be higher than the figure calculated by the model, dealers say – in which case, the model calculation would be ignored. 

“The danger is that if you make the floor too big, then you will have spent millions building a fancy model, only to have the calculation be driven by the crude floor. Yet if you don’t build the model, you’re not allowed to use the floor,” says Eduardo Epperlein, global head of risk methodology at Nomura in London.

Indeed, the alternative is 100% of the standardised method – a capital charge that would make life extremely difficult for correlation traders. “It would be very hard to run a significant correlation trading business on those standard rules,” notes Bakhshi at Credit Suisse.

Making the situation worse, banks would still have to apply the VAR and stressed VAR calculation to correlation trading books, on top of the CRM. “Given the CRM is meant to be a comprehensive risk measure for a particular business – correlation trading – it would be better if it encompassed general market risks as well, rather than having these covered in VAR and stressed VAR,” says Andrew Abrahams, global head of quantitative research at JP Morgan in New York.

There are also uncertainties concerning the scope of the CRM. For example, the rules stipulate securitisations or nth-to-default credit derivatives that are not resecuritisations or tied to retail products must be included in the CRM. They also state products earmarked as part of a bank’s correlation trading portfolio must have reference entities with liquid markets.

Ben De Prisco, senior vice-president for research and financial engineering at Toronto-based systems vendor Algorithmics, believes this could lead to some debate over what products or names are acceptable. “The regulations specify the securitisations you choose to put into the correlation portfolio need to have a reference entity that is liquid with a two-way market. I can see some discussions going on where people claim some products have a liquid market and others may not,” he says.

There could also be some discussion over what constitutes an appropriate hedging instrument. The rules state banks can use their CRM models for correlation book hedges, providing these are not securitisation exposures or nth-to-default baskets and a liquid market exists for the product or its underlyings. “The regulations allow you to model your correlation book with appropriate hedges, which might lead to some discussions about what would constitute appropriate hedges,” notes De Prisco.

Ultimately, it is up to national supervisors to decide whether the CRM models built by dealers are appropriate. Recognising there is room for different interpretations of the rules, the Basel Committee’s trading book group has set up a subgroup to share experiences on CRM modelling and help define some common ground between regulators. But when it comes to implementation, the CRM may be just the tip of the iceberg, say dealers.

For a start, questions continue to be raised about the IRC – particularly over the coverage of the charge. “There are still quite a few questions around the IRC on scope – so what exposures or assets do you include and when can you carve things out?” asks Bakhshi at Credit Suisse.

Some risk managers portray the new stressed VAR charge as equally challenging, largely thanks to the subjective nature of the one-year stressed period. Should market participants use different stressed periods for different business lines, for instance? One European regulator told Risk it had not received any requests for this type of modelling, but would be willing to consider the idea: “We might accept a request if a bank would like to do some business-line stressed VAR modelling. Some questions would then have to be answered, like how to define the business lines and how to aggregate the different stressed VARs.”

When it comes to choosing the stressed period, banks say they are not sure what they should be looking for. The Basel Committee suggests a one-year period between 2007 and 2008 as an example of an adequate stressed period, “although other periods relevant to the current portfolio must be considered”. As a result, Abrahams at JP Morgan says it’s not obvious what supervisors want. “The Basel Committee has laid out a general concept, but the way it is specified, it is not particularly obvious whether you have to determine a stressed period that is stressful for your existing portfolio or whether it is just sufficient to calibrate using some historical period that is deemed stressful. The former is much more sensible,” he says.

Johannes Rebel, head of market risk at Nykredit in Copenhagen, agrees: “There’s a bit of interpretation as to how to actually establish the exact period for the stressed VAR. Do you just take any stressed period, or do you actually have to find the worst period of all possible periods? That could be a huge computational task.”

Another issue is how to model for new risk factors and products that have emerged subsequent to the stressed period applied by the bank. For instance, if 2008 were used as a stressed period, it was unclear how dealers would adapt it to cater for future structural changes in the market. “How do you handle risk factors or products that are relevant now but were not present in 2008? How would you handle a new currency? That is only going to get more complex as time goes by,” says Nomura’s Epperlein.

This point was acknowledged by the UK Financial Services Authority (FSA) in July, when it released a consultation paper on the new rules. “One area of concern where firms did request specific guidance was how to address instances where firms have positions in products that did not exist in the stressed historical period,” the regulator says.

The FSA also recognised banks would benefit from guidance in selecting and updating a stressed period, identifying the stressed VAR multiplier and applying stressed VAR across subsidiaries and parent entities. Providing some initial guidance on implementation, the FSA says it does not intend to define a default stressed period for all firms. Instead, it wants banks to “justify the reasonableness of their selected stressed period and its continued relevance to their portfolio”. Meanwhile, when no historical data is available for new products or risk factors, it states banks should adopt proxies in a similar vein to existing VAR models.

Notwithstanding this, bankers say the real clarification will come once supervisors begin to implement the new rules. “There is scope for a difference in interpretation across regulators. For us, the approval approach is going to be where you get the remaining clarifications,” says Bakhshi at Credit Suisse.

But of all the criticisms, the biggest is probably double-counting. Where banks would previously have used a standardised measure or a VAR-based market risk charge, they may now be faced with five separate calculations, including a standardised charge, a VAR-based charge, a stressed VAR calculation, the IRC and the CRM. Because the charges are additive, they eliminate some of the diversification benefits banks might otherwise expect, claims Bakhshi. At the same time, the stressed VAR measure will naturally resemble the VAR-based market risk charge. “The stressed VAR component has a great deal of overlap with the VAR-based charge, and particularly when you’re in a crisis, those numbers are going to look exactly the same,” Bakhshi says.

Indeed, by using a series of separate charges, market participants believe the Basel Committee could have inadvertently increased operational risk. Christopher Finger, who works in applied research at risk analytics firm MSCI in Geneva, worries the complexity of the new rules could make it harder to stop banks from exploiting loopholes. “You don’t really understand what you’re covering with these rules. There’s double-counting but there might also be risks slipping through. We add together several charges and we hope it’s a big enough number. But complex rules mean complex loopholes – harder to find, but also harder to catch once they are found,” he warns.

De Prisco at Algorithmics says the changes will force banks to use a greater number of systems for trading book capital – something likely to make implementation more cumbersome. “You now have these four or five disparate measures and what we’re seeing from banks is they don’t have a single system in place that does all of them. So suddenly, rather than using one unified system, they are trying to make do with a mosaic of systems that all have to be co-ordinated in terms of data, algorithms and reporting,” he explains.

The complexity of the charges has also led to questions about the Basel Committee’s use test, or the idea models used for regulatory capital should also be employed for day-to-day risk management. While it is clear levels of trading book capital were too low in the run-up to the recent market crisis, risk managers say they are unlikely to use the framework decided by the Basel Committee for actual risk management. Consequently, banks are in a difficult position. “You have to convince regulators you would have used the thing you never designed in the first place,” remarks one former risk manager.