Crowd trouble: the FRTB’s war on basis risk

Flexibility is the foundation of the market-making business. A dealer’s ability to hedge single-stock positions with indexes, Eonia with Euribor, or a nine-and-a-half-year-swap with a 10-year swap, is how it manages risk while making money. But under new trading book capital rules, this flexibility comes with an increased cost – and there are growing concerns about the second-order effects.

The Fundamental review of the trading book (FRTB) punishes with extra capital anything that does not offset perfectly. This could affect the price of less liquid trades and push end-users towards more standardised products, some fear; the result would be less diversity, more crowding, and potentially a more fragile market.

“The FRTB’s standardised approach is basically central planning of risk pricing, and it will produce Gosplan-like results,” says Craig Pirrong, professor of finance at the University of Houston, referring to the state economic planning unit of the Soviet Union.

It’s not just the regulator-set standardised capital formula, though: banks argue tighter constraints on internal modelling will produce similar effects.

“The revised standard rules are intended to be calculated and executed uniformly across the industry, and you’ve also got a much more constrained internal modelling framework where the expectation is that models will become more similar in their treatment of risk. That has the potential to increase herd behaviour,” says Ed Duncan, a London-based director in the risk function at Barclays.

These worries are now starting to get more attention – not just among banks, but also academics, corporate hedgers, and even some regulators. According to two consultants, a number of European banks have been encouraged by their supervisors to stick with the internal models approach (IMA) on the grounds that it will allow more diversity. 

Risk.net spoke to 10 dealers for this article; all agreed, to varying degrees, that the FRTB was a recipe for more homogenous trading. That view holds for both capital methods: the standardised sensitivities-based approach, and the IMA.

“Both approaches encourage banks to trade only standardised instruments and converge on the same types of risk and underlyings. We’ve raised this point with the European Central Bank and other supervisors, as well as Basel’s Trading Book Group,” says Arie Boleslawski, deputy head of trading at Societe Generale Corporate & Investment Banking in Paris.

A source at the European Commission disputes this, saying they do not think the FRTB will increase systemic risk. The Basel Committee did not respond for comment by press time.

If [the FRTB is] not softened, the volatility of non-vanilla products would be likely to increase materially. Banks may possibly look to offload this into shadow banks
Mark Penney, HSBC

The potential ramifications for the industry if these behaviours arise are stark. At times of market stress, liquidity drains from exotic products and often gaps even in more vanilla products – an effect that will be accentuated by the FRTB, some fear, stoking procyclicality. This may lead dealers to pre-emptively cut off certain hedging relationships with end-users and drive business into less well-regulated corners of the financial system.

“Liquidity, or its absence, certainly becomes more likely to be self-reinforcing,” says Mark Penney, head of capital management at HSBC in London. “If [the FRTB is] not softened, the volatility of non-vanilla products would be likely to increase materially. Banks may possibly look to offload this into shadow banks,” he adds, meaning buy-siders looking for bespoke trades would have to go off the beaten track for willing counterparties.

On the flip side, concentration in certain products will lead to banks’ portfolios becoming more closely correlated and liable to move in unison under market stress. Max Verheijen, managing director at pension fund asset manager Cardano in Rotterdam, says banks may push buy-side firms into more cleared and standardised trades. The result? “The system gets more fragile. If it collapses now, you will have collateral damage,” he warns.

Liquidity bias

The SBA and IMA implementations incentivise uniform trading in slightly different ways: the former by the calibration of the standardised formula used to generate the market risk charge, and the latter by the constraints loaded on to dealers’ use of their own models.

“Ultimately the impact of the FRTB on the market, whether banks use the standardised or internal models approach, is to reduce liquidity. The less liquid an instrument is to start with, the less likely banks will want to keep trading it, which is where the herd mentality kicks in. Sure, it will be viable to maintain a US Treasury trading book – but what about sub-investment grade, non-G7 currency eurobonds?” says Moorad Choudhry, professor at Kent Business School.

The SBA was designed as a more risk-sensitive update to Basel 2.5’s standardised approach. Its purpose is to act as a credible fallback for – and potentially a floor to – the internal models approach, with dealers using their own pricing models to assess the sensitivities of instruments in their trading book to myriad prescribed risk factors (see box: The sensitivities-based approach).

Dealers say the structure of the SBA is biased against banks running basis risk on behalf of clients, and hence will discourage the warehousing of these positions.

The less liquid an instrument is to start with, the less likely banks will want to keep trading it, which is where the herd mentality kicks in. Sure, it will be viable to maintain a US Treasury trading book – but what about sub-investment grade, non-G7 currency eurobonds?
Moorad Choudhry, professor at Kent Business School.

Specifically, idiosyncratic risks hedged by vanilla instruments are vulnerable. A combination of high risk weights for specific factors and the mechanics of the correlation formulas are to blame.

Take the example of the equity risk charge. SBA delta and curvature equity risk factors are spot prices, and vega factors the implied volatilities of options that reference these prices.

Charges apply to net sensitivities, so long and short positions on specific equity names are offset – but the SBA denies this benefit where a long position in single stocks is hedged by a short position on an index, or vice versa.

Why? Because the rules demand dealers break down index positions into their individual underlyings and calculate a separate notional position for each of the constituents, equal to the market value of the index multiplied by the percentage of the index that the constituent represents. This formula means the long position in each single stock in the above example will only partially be offset by the short position in the relevant index constituents.

This will seriously hamper popular equity strategies such as dispersion trades, says SG’s Boleslawski, in which investors play off the price of an equity index against its constituents.

“Say you are short a basket of single-name variance swaps including automotive stocks and are long variance swap on an index. If you look historically at these positions, in terms of correlation their mark-to-market moves inversely, and our P&L reflects that. Under the SBA, you will shock the automotive stocks by 30%, with a strong impact on single-name variance swaps, but when you shock the hedging index swap, after looking through, the shock to the automotive constituents may only result in a risk weight of, say, 2%. On top of this the correlation formula that recognises hedging and diversification within buckets assumes low correlation. This is very, very penalising for positions where you have single names hedged by indexes,” he says.

Different stresses

A dealer must submit their portfolios for each risk class to three different stresses assuming high, medium and low correlations. The scenario that generates the largest capital requirement is selected as the ultimate risk class charge. The compounded effect of the equity risk weights and correlation formula is what may wreck these basis positions.

A similar problem will bedevil rates desks. The delta risk factors for this asset class correspond to specific vertex points on a risk-free yield curve for each currency: for example, for the euro this could be the Eonia swap curve. Instruments with delta sensitivity to interest rates in the same currency – priced with a curve other than Eonia – are captured through prescribed correlation formulas

The problem is the combined effect of the correlation formulas with the overarching correlation scenarios will punish any basis positions a dealer has on its books. For instance if a trader is long Eonia at the one-year point and short three-month Euribor at the one-year point, they would consider themselves economically hedged – but the SBA applies a 99.9% correlation factor for such a position. This results in a capital charge equivalent to 4.5% of what it would be had the long position been entirely unhedged. The charge is also punitive for positions referencing separate points on separate curves, and separate vertices on the same curve.

The 4.5% charge sounds manageable – but the ultimate charge for each risk class is determined by the correlation scenarios.

Basis mismatches within a portfolio are magnified by the low correlation stress, almost guaranteeing it will produce the highest charge – even where economic correlations within a risk class are high.

In the example of a five-year euro fixed-for-floating rate swap referencing six-month Libor, €100,000 DV01, this would require €10.6 million in capital under the SBA. Imperfectly hedged with another five-year euro swap of the same maturity, but referencing one month instead of six month Libor, would cut only €3 million from the unhedged charge assuming the low correlation scenario applies – far exceeding the risk implied by the P&L volatility of the position.

“This means a bank hedging a client exposure – for example interest rate risk to a specific curve, using a different more liquid benchmark curve such as Libor – would have to calculate net sensitivities for each curve without the benefit of offsetting or netting, regardless of how closely they are correlated,” explains Thomas Ehmer, London-based senior manager at consultancy Baringa Partners.

Mind the gap

Regulators are alive to the industry’s gripes. In a paper published by the European Banking Authority in November last year, it noted that the low scenario assumes basis positions are correlated at a shade under 75%, resulting in “highly inflated charges for highly correlated positions”, and explained that the industry was seeking “clarification” on whether basis positions should be exempt. No such clarification was forthcoming in Basel’s long-awaited FAQ paper published on January 26, however.

Faced with this threat to warehousing basis risk, banks have few escape routes. They could seek to avoid such positions, or pass the cost of running them on to clients. Either option would be bad news for end-users.

The Association of Corporate Treasurers, representing a wealth of derivatives users, is concerned the FRTB “will affect real economy enterprises which may find the cost of hedging becoming so expensive as to become prohibitive”, says Steve Baseby, associate policy and technical director in London.

Dealers will of course focus on businesses where they are not exposed to idiosyncratic risk. This means a concentration on standardised trades referencing liquid risk factor
Arie Boleslawski, Societe Generale Corporate & Investment Banking

Alternatively, dealers could encourage clients to migrate to standardised products and wear the basis themselves. However, not only would this throw complex risks on to institutions less capable of dealing with them, it could build up systemic risk within the banking system.

“Dealers will of course focus on businesses where they are not exposed to idiosyncratic risk. This means a concentration on standardised trades referencing liquid risk factors,” says Boleslawski. “For equities, this means increased activity around indexes. Yet if this happens at a time of market stress, everyone will be holding the same position. Right now positions in single stocks differ from one dealer to another, depending in particular on its client base, but if everyone comes to indexes we will always be trading the Euro Stoxx 50.” 

The dangers of regulatory-induced crowding among banks were illustrated in a Bank of England working paper released in January. The BoE found that Basel II, which, like the FRTB, included both a standardised and internal models approach, concentrated high loan-to-value mortgage risk in lenders using the standardised approach – those by nature less likely to have sophisticated risk management tools at their disposal.

Other dealers say end-users will simply shake up their businesses to accommodate this herding. “Corporates will start restructuring their own transactions so their hedging needs become standardised. Certain contracts will change and the market can accommodate this. It’ll be inconvenient, but not all negative,” says a risk model head at a North American bank.

As an example of where banks’ behaviour may change, one risk management consultant cites the inflation swaps market.

“A dealer may use an interest rate as a proxy hedge to inflation because he is not able to effectively hedge inflation risk directly for that country. If a corporate takes out an inflation swap and the facing dealer hedges with an interest rate swap, that becomes prohibitively expensive and the corporate will be encouraged to buy the interest rate as a hedge rather than inflation,” he explains.

Yet crowding around certain benchmarks comes with both costs and benefits – and the size of the latter cannot be discounted.

“With Libor, because of the agglomeration of liquidity around this benchmark, we have deep and liquid markets connecting a wide range of related instruments. But there is indeed a cost: it is now difficult for an even better benchmark to emerge, given that liquidity will strongly remain with Libor until some regulatory change or another bad event of manipulation. Moreover, Libor is not as sound a benchmark as it should be, given the paucity of underlying term unsecured bank borrowing transactions. We should not rely so heavily on a benchmark unless it is extremely robust,” says Darrell Duffie, professor of finance at Stanford University.

Own model bias

The SBA is painfully conservative by design – in part, to drive banks to use internal models. An industry survey reported in June last year that the standardised approach would increase capital requirements by 240%. Internal models, in contrast, would generate a 150% increase.

Policymakers charged with implementing the rules say they do not want firms taking on these sorts of increased costs, suggesting a preference for the less capital-intensive IMA.

“After we have materially strengthened the quality and quantity of bank capital in previous reforms of the Capital Requirements Regulation, a further significant increase of capital requirements across banks and types of risk should be avoided,” says a European Commission official.

Dealers have also been lobbying the Basel Committee to tweak the IMA to further incentivise banks to migrate to this system of calculating their regulatory capital. “Initiatives are underway [at Basel] to give some incentive for the banks to move from the SBA to the IMA,” says a source close to the discussions. January’s FAQ paper addressed some of the concerns dealers had in this regard, several say.

“The majority of banks will push to be on internal models. If for whatever reason large numbers of banks end up with the SBA, there will be a substantial capital increase across the industry which is not what the regulators wish to see,” says Dong Qu, FRTB project lead for the front office at UniCredit in London.

Those dealers pursuing the IMA may be forgiven for believing the calibration of the SBA is of no consequence to them. Yet there are two ways in which it could still influence their capital planning.

First of all, policy watchers say Basel is still debating whether IMA capital should be floored at an amount calculated by the standardised approach. If a high floor is set – at anything above 50–55% of standard rules, say dealers – the standardised approach rather than the modelled approach will have the potential to drive capital allocation. The question of floors was absent from the FAQ paper.

Second of all, a trading desk that fails the FRTB’s byzantine profit and loss attribution test will fall back on to the SBA and its more punitive capital calculation. In an example of regulatory cognitive dissonance, a well-hedged desk appears more likely to fail this test than an unhedged one. What is sensible practice under the SBA is not seen in the same light under the IMA.

At face value, the IMA affords much greater freedom to banks to generate their own capital charge. Those internal models that pass muster with regulators can be used to define additional risk factors to those specified under the SBA, allowing for a wider range of sensitivities to be modelled. In addition, banks can use empirical correlations within risk factor classes instead of Basel’s own correlation formulas.

However two aspects of the IMA regime could give rise to crowding. First of all, Basel requires IMA banks to assign their risk factors to set liquidity horizons, ranging from 10 to 120 days, to account for the length of time policymakers expect it would take firms to hedge or exit a position in stressed market conditions without causing wild price fluctuations.

The issue of NMRFs is of very high concern, because when evaluating the regulatory requirements banks may find out there are many risk factors that would be non-modellable
Rita Gnutti, Intesa Sanpaolo

These liquidity horizons adjust the expected shortfall calculations banks use to capitalise each risk position. The longer the liquidity horizon, the greater the total risk charge. These act like multipliers on dealers’ capital, says Barclays’ Duncan.

“For example if you have credit risk in the high-yield space, whether it be corporate high yield or sovereign high yield, you end up with a multiplier against the capital that you are applying today – perhaps between as much as four to six times over what a 10-day VAR may produce today. Even when you factor in a removal of the double-counting of VAR and stressed VAR and a reduction in the multiplier, the FRTB could still result in an increase in IMA capital. The scaling up of capital in the longer liquidity horizon buckets is going to make them less appealing to hold,” he says.

Second, dealers’ freedom to use their own risk factors as model inputs is subject to conditions. Those factors that cannot be evidenced by a sufficient number and frequency of verifiable quotes are disqualified. These non-modellable risk factors (NMRFs) are subject to stressed capital add-ons that exceed the charges assigned to their modellable cousins. IMA dealer participants in an industry survey say NMRFs will contribute a whopping 30% of their total market risk capital.

Given the size of this capital punch, dealers will understandably want to minimise trading positions that reference prohibited risk factors. This will concentrate activity on instruments linked to risk factors underpinned by a wealth of market data. UniCredit’s Qu says this will incentivise crowding in the same manner as the SBA.

Others agree: “The issue of NMRFs is of very high concern, because when evaluating the regulatory requirements banks may find out there are many risk factors that would be non-modellable. This risks banks having just a few liquid factors to include in the model. If this is the case, everyone will be concentrated on a few liquid risk factors, so liquidity will dry up on others,” says Rita Gnutti, head of internal model market and counterparty risk at Intesa Sanpaolo in Milan.

Dealers will also be discouraged from incorporating risk factors that lie on the border of modellability: those that qualify for IMA one day, but could fall into the NMRF bucket the next.

The FRTB’s standardised approach is basically central planning of risk pricing, and it will produce Gosplan-like results
Craig Pirrong, University of Houston

A risk factor’s modellability must be assessed on a monthly basis. For a factor to avoid the NMRF sin bin, a bank must provide at least 24 observable real prices within a 12-month period, with no more than one month between two consecutive transactions. What constitutes a “real price” is an ongoing subject of debate.

“You’re not going to want to be on the borders of modellable treatment around reporting periods, particularly around quarter-end and year-end, because there is the potential for your capital measures to become volatile. Products without the requisite frequency of trading will become more expensive to maintain, and perhaps more expensive to trade. That means anything long-dated, anything a little more bespoke,” says Duncan.

For risk factors on the edge of modellability, the consideration for the dealer is whether they can make enough profit acting as liquidity providers to cover their cost of capital.

Alternatively, consultants say they could add a spread to instruments vulnerable to these factors, perhaps as a constituent to the capital valuation adjustment (KVA).

“A bank may decide they have to push the costs of the potential loss of modelled capital to the end-user,” says David Milne, national leader, quantitative advisory services for Canada at EY in Toronto.

The haves and have-nots

Any instrument that is infrequently traded will be targeted for scrutiny. Corporate bonds that trade infrequently are at risk as they may lack the requisite number and frequency of price observations to qualify as modellable.

This may prove the final straw for bit-part players in the bond market. Increasing trading velocity on these illiquid names to ensure their modellability is a non-starter for these dealers, who may already be struggling to generate sufficient return on capital to keep the business running.

“I’ll give you an example,” says Ryan Ferguson, head of credit derivatives and XVA at Scotiabank in Toronto. “Say we trade a particular bond once a week but another bank trades it once a day. They don’t need as big a bid/offer to cover their capital cost as us since they’ve got a high trading velocity where they make enough on each trade to justify that capital – whereas we’re only doing a fifth of their flow to support the same amount of capital. If you don’t have scale you will have to shut down, and that further concentrates activity among those who already have that scale.”

The same is true for over-the-counter derivatives referencing illiquid risk factors: banks face a trade-off between the cost of quoting instruments to ensure modellability and simply accepting the NMRF charge. Some may decide to exit certain markets altogether.

Dealers are working together on pooling data through various utilities in a bid to classify more risk factors as modellable – though some argue this could further the concentration of certain trading activity.

“Internal model banks could buy modellability from external vendors if they do not have the necessary data to hand. This could create another divide among IMA banks that can afford to invest in this market data, and trade across many risk factors, and further reduce trades in some illiquid market segments,” says Intesa’s Gnutti.

The sensitivities-based approach

The Fundamental review of the trading book’s standardised market risk charge is the sum of three components: the SBA, a default risk charge, and a residual risk add-on.

The SBA asks banks to map the sensitivities of all instruments in their trading books to a host of prescribed risk factors designed to capture their cumulative delta, vega and curvature exposures.

Risk factors are bucketed according to common characteristics, and these buckets are in turn assigned to one of seven risk classes: general interest rate risk, foreign exchange risk, equity risk, commodity risk, and three classes of credit spread risk (non-securitisation, securitisation, and correlation trading portfolio).

For the delta and vega components of the risk charge, the net sensitivity to each factor across instruments is multiplied by a set risk weight to define a total risk position. For example delta equity risk factors are equity spot prices, vega equity risk factors the implied volatilities of options that reference these spot prices.

Correlation formulas are applied at the risk bucket and risk class level to generate the cumulative delta and vega risk charge.

The curvature charge, meanwhile, applies to all instruments embedding optionality: equity options and interest rate swaptions being two obvious examples. Its rationale is to capture risks peculiar to instruments with convex payoffs which the delta risk component may miss out. To find the curvature charge, the risk factor is subjected to two stress scenarios involving an upward shock and downward shock, with the worst loss of the two providing the total risk position.