Basel 2.5 meets the Sarkozy trade: new rules could hit bond demand

Regulators want banks to hold more capital against government bond positions, but the regime is being changed at a time when the industry is the main source of demand for big eurozone issuers such as Italy and Spain. In addition, banks fear modelling difficulties could make the numbers meaningless. By Laurie Carver

It's known as the Sarkozy trade – using the cheap three-year loans doled out by the European Central Bank (ECB) in December and February to buy government bonds, particularly those issued by a bank’s own sovereign. According to ECB data released at the end of April, Italian banks added €67.5 billion to their holdings of eurozone government debt in the four months from the start of December to the end of March, while Spanish banks added €77.4 billion – exactly what France’s former president, Nicolas Sarkozy, had hoped to see. “This means that each state can turn to its banks, which will have liquidity at their disposal,” he said of the ECB’s first long-term refinancing operation.

It has had an obvious effect. Yields on 10-year Italian bonds fell from 7.26% on November 25 to 4.68% on March 8; Spanish yields fell from 6.70% to 4.91% over a similar period, although both have since rebounded. But the Sarkozy trade could have a cost, too.

On January 1, the European Union (EU) introduced new rules on trading book capital, known as Basel 2.5. This package of capital add-ons includes the incremental risk charge (IRC), which is designed to capture default and ratings migration risk in trading book assets – and requires banks to model the risk of even high-rated government bonds for the first time, according to clarifications provided by the Basel Committee on Banking Supervision and the European Banking Authority (EBA) (see box). That has obvious implications.

“For countries such as Spain, Portugal, or Italy, domestic banks are the only people buying their government’s debt, and they rely on a zero risk weighting. With a non-zero default probability they might have to come up with more capital to buy the bonds. And raising capital is going to be pretty tough for them right now because there is a risk aversion to periphery financials. So you end up with the risk of a slide to default – with the prudent capital requirement exacerbating the troubles of the sovereign and the banks,” says Michael Hampden-Turner, a credit strategist at Citi in London.

That isn’t happening yet, says Juan Blasco, head of credit markets at BBVA in Madrid – but it will if national regulators follow the path laid out by the Basel Committee. “If the IRC is implemented across Europe for sovereigns, it would definitely have an impact. The size of the resulting capital charge would depend on the rating of the asset we’re talking about – for AAA it will be pretty minimal, but for anything below A it will be in double-digit basis points,” he says.

You end up with the risk of a slide to default – with the prudent capital requirement exacerbating the troubles of the sovereign and the banks

It could take a while for the impact to filter through. For one thing, national regulators have to approve IRC models for use with government bond portfolios – a process that has taken the best part of 18 months in Switzerland, where Basel 2.5 was adopted a year earlier than the EU. While waiting for approval, Swiss regulators required Credit Suisse and UBS to apply a capital placeholder, according to one of the banks. But when it arrives, the IRC hit could be substantial – one European banker says the charge for a single bond position could be up to 10% of the mark-to-market value, even for investment-grade countries, and might equal the assumed loss given default for issuers rated below BBB.

In addition, some regulators already appear to be granting more flexibility than the Basel Committee and the EBA would like. A spokesman at the Banco de España says Spain’s lenders are allowed to treat Spanish bonds as free of default risk for the purposes of the IRC, though not of migration risk.

In Italy, Rita Gnutti, head of market and counterparty risk internal models at Banca Intesa San Paolo, says any sovereign rated BBB or above qualifies as default risk-free in the bank’s model – currently, that includes both Italy and Spain. The Italian regulator, the Banca d’Italia, declines to say whether the model has been approved.

If banks are being given some flexibility on the IRC by regulators in the peripheral eurozone countries – modelling ratings migration risk but not default risk, for example – it might look like prudential regulation is being sidelined for political reasons. But that is not the whole story – regulators elsewhere are also deviating from the path laid out by the Basel Committee and the EBA.

A senior modeller at one UK bank says the Financial Services Authority (FSA) is allowing a small number of sovereigns to be considered free of default, including Germany, the US, Japan and the UK. The inclusion of the last of these might raise eyebrows, but the modeller defends it: “There wouldn’t be much point in us holding capital against a UK default. In such an event, our gilt holdings in the trading book would be the least of our priorities,” he says.

This was not the committee’s intent, according to a person familiar with the development of Basel 2.5. “I think it was clear in the guidelines that no sovereign was default risk-free,” he says. “I don’t think you can say with a straight face that having a zero PD isn’t against the spirit of the regulations.”

It is understood the FSA stance is derived from the Basel Committee’s guidance, which calls for a capital charge to be applied to sovereign bonds unless the PD model output implies a charge of zero. In essence, for its list of low-risk countries, the FSA is allowing banks to assume the output would be zero rather than go to the effort of modelling it.

In Germany, that would not fly: “A zero default probability is not allowed,” says Rüdiger Gebhard, director in the banking supervision department at German regulator Bundesanstalt für Finanzdienstleistungsaufsicht (Bafin). “Every name must be able to default. Besides, a zero charge is not the same as a zero probability of default (PD).”

Even with unanimity among regulators, it would be difficult to find it in the numbers produced by IRC models, banks warn. The raw material for credit modelling – default and downgrade statistics – is completely absent for higher-rated sovereigns, and patchy for issuers further down the ratings scale. There are a variety of ways to work around the data problem, but they are likely to produce capital numbers that cannot be compared, are unstable – and even meaningless – say the quants and risk managers charged with implementing it.

“I think the worst that can happen is that it’s useless,” says Massimo Morini, Milan-based head of interest rate and credit models at Banca Intesa San Paolo. “The high degree of uncertainty in the modelling approaches means two banks can come up with two numbers that really have nothing to do with each other. At the end of it you have to wonder: how believable are those figures going to be?”

“It’s a difficult process,” says Bruno de Cleen, Utrecht-based head of credit portfolio management at Rabobank. “For sovereigns, transition data is statistically too sparse to do it properly, and for defaults it’s even worse. You can choose one of many mathematical methods, but essentially you have to use your judgement to a certain extent. It’s unlikely two banks would come up with the same method.”

Some are simply throwing up their hands in response, according to one regulator. “I think everyone was a bit surprised it would apply to sovereigns,” says Christine Lang, senior specialist in the market risk team at the Swiss banking supervisor, Eidgenössische Finanzmarktaufsicht (Finma). “I was at a conference last year, and there were presentations on the IRC. Several speakers indicated they might be paying it lip service when it came to sovereigns at the time.”

From a modelling perspective, the problem banks face is simple to explain and difficult to solve: it’s not easy to work out how likely an event is to happen in the future when it has rarely happened in the past – or never, in the case of a US default. The database belonging to Hawaii-based credit risk software vendor Kamakura contains 123 sovereign defaults, compared with 2,046 for corporates. Predictably, the vast majority of those defaults relate to issuers that have always been relatively low-rated. There are few examples of a country that has tumbled down the ratings scale to default.

Statisticians have come up with various ways to predict distributions of future variables with sparse data – and the exotic names of the resulting theories give some idea of the difficulties involved: extreme value theory, imprecise probability theory, Bayesian resampling, expert fusion, low-default portfolio theory. All posit some parametric or numerical form for the proposed future distribution, and they also rely on user inputs where there is a lack of data – in essence, judgement, or something like it.

Half science, half art
The head of credit analytics at one European bank has been looking at how to use low-default probability theory to develop sovereign PDs. “It’s a way of interpolating the default curve from the low-grade end – where you have sufficient data to construct PDs statistically – to the high-grade end, where you don’t have enough default data. The method basically gives a way of doing this by combining what observations you have with simulated data to the extent that you observe at least one default with 50% probability. It’s half science, half art – a statistical technique that uses an artistic set of assumptions,” he says.

Finma’s Lang says the arty bit of the work needs to be kept to a minimum: “I think we have been quite Calvinistic on that issue of close-to-zero PDs, really forcing them to come up with something. But however you do it you can still end up with a number in the billionths of a per cent, and at that level it’s within the margin of error,” she says.

While the EBA and the Basel Committee have been clear that PDs have to reflect the actual chances of default, they have left the door open to deriving that from market prices, by using standard hazard-rate models that convert credit default swap (CDS) spreads to default probabilities, for instance. That comes with problems of its own – CDS spreads include a risk premium reflecting liquidity and supply-and-demand factors as well as counterparty concerns, producing default probabilities that can look wildly pessimistic (see figure 1).

 

“How much is really default, how much is risk aversion, how much is liquidity premium? You can sometimes only really give an intuitive answer to these questions – it then feels like guesswork,” says Rabobank’s de Cleen.

And that is just default risk modelling. Calculating the IRC requires another two principal elements to be modelled – the change in rating of a portfolio’s assets, and the resulting effect on the valuation.

Facing deadline pressures, some banks decided to adapt existing models rather than start from scratch. “While Basel 2.5 was being implemented, there were already new things coming up like Basel III, which made it necessary for banks to allocate their resources sensibly – it made sense to use something they already had,” says one European regulator.

The obvious place to start with many of the larger banks was the internal ratings-based (IRB) model used to calculate credit risk for assets in the banking book – the IRC implementation guidelines suggest it should be “consistent” with IRB models. “Most of the banks that apply the IRC also use the IRB for the banking book, so they can build a little bit on their internal modelling. When you develop a model it can be more or less easy to adapt to changes in rules. In the cases we looked at, it was not a big problem to take the model for corporate exposures and adjust it for sovereigns,” says Finma’s Lang.

Banca Intesa San Paolo starts with sovereign transition matrices from Moody’s Investors Service, and estimates the resulting shock to the assets with a deterministic impact model referring to the spread levels the bank ties to the individual ratings (see tables A and B). One problem with this approach is that the Moody’s sovereign matrix excludes certain possibilities – an AAA-rated sovereign has a 96.8% chance of remaining in that grade over the course of a year, and a 3% chance of dropping to AA, but zero chance of falling any further than BBB. A BBB-rated sovereign has an 87.9% chance of retaining that grade a year later, only a 3.4% chance of being downgraded at all – and the matrix ignores the possibility of a three-notch downgrade. There are other, weirder results caused by the data shortage – an AAA rating has a greater chance of falling to BBB than to A, for instance.

As a result, a quant at one global bank says his employer has altered the rating agency matrix, to ensure no transition has a zero probability. The bank also prefers a stochastic model for the impact.

But transition matrices cannot be tinkered with at will. Each row – representing the probabilities of a transition from one rating to all other possibilities – has to add up to one. In addition, it is much harder to extract downgrade risk from CDS spreads than default risk. Again, modellers warn, this leads to ad hoc approaches that will differ from bank to bank.

“The probabilities in rating agency transition matrices are very often equal to zero,” says Finma’s Lang. “So you have to refine it somewhat to get an acceptable matrix. Allocating the probability mass on the different levels is more art than science.”

It’s dangerous art, too. The global bank’s quant was disturbed at how sensitive the resulting capital charge was to these inputs. “We introduced a floor for the entries in the transition matrix, because the Moody’s version wasn’t conservative enough. We started at a 1-basis-point level, but just a small change in the floor resulted in quite radical differences in the charge – up to a 50% difference,” he says. The bank eventually settled on a 0.5bp floor.

David Phillips, London-based group head of risk analytics at Royal Bank of Scotland, agrees the IRC model can produce volatile capital numbers: “It’s quite sensitive to model parameters, particularly when related to large, concentrated positions in highly rated names. While the corresponding PDs are small – on the scale of a few basis points – even a change of 1bp could cause a significant change in the IRC,” he says.

After drawing up the transition matrix, a revaluation model is required to estimate how the simulated ratings transitions affect the value of the portfolio. The industry standards, at least for modelling corporate migrations, are the structural models that trace their lineage back to Robert Merton’s 1974 credit model. This started with an ingenious comparison – debt defaults when a firm’s assets dip below their liabilities, which makes holding debt analogous to being short a put option. Similarly shareholders have a claim to the excess of the assets over the liabilities, making them long a call. The observation converts bond pricing into a solvable option pricing problem, for which the share price is the major input.

Popular model
Some banks are using the Merton approach to revalue sovereign bonds in line with simulated ratings transitions – the result is an IRC model in which these events are barriers triggered when the spread level passes through them. The model is popular because it is comparatively simple to implement.

But banks using this approach have a problem when applying it to sovereigns – how to come up with a substitute for the stock price that is the main input in the corporate version of the model. Banca Intesa San Paolo’s team extracts survival probabilities – effectively the flipside of risk-neutral default probabilities – from CDS spreads, finds their correlations with the constituents of a stock index, and plugs this into the model in place of the share price.

“In a way, we model the revalued sovereign portfolio as a bit like a corporate bond derived from an index of financials,” says Gnutti. “We developed a comparison analysis between survival probabilities and equity prices in the corporate world, and it showed a clear common trend. The model takes into account the potential noise related to the transmission of shock between the CDS and equity prices.”

That approach has its critics. “You shouldn’t use Merton models to model sovereign default risk. There’s no market instrument to play the role equity does for corporates. A CDS can’t stand in for equity since countries don’t default when their assets fall beneath a certain level. They default because there’s rioting on the streets,” says Jens Hilscher, senior research fellow at Kamakura and an assistant professor of finance at Brandeis University in Massachussets.

What all this means is that regulators have their work cut out for them when it comes to model approval for the IRC. Finma has the longest experience with this – Switzerland introduced Basel 2.5 for its banks at the start of 2011, and after 18 months it is understood one of the country’s big institutions now has approval to apply its model for the IRC to sovereigns, while the second is close to being signed off. It has not been easy – and Lang warns that different approaches will make it tricky to compare banks from one country to the next.

“It’s always surprising just how ingenious quants can be,” she says. “You can get very different results even within similar frameworks. As soon as you have these models you have many ways of doing it and it makes the international comaparisons challenging. All we can do is keep talking to the other regulators.”

 

BOX: Laying down the law

Since the Basel Committee for Banking Supervision first proposed rules on the capital treatment of market risks in 1993, government bonds have enjoyed a special status – no capital needed to be held against specific risk, the danger of a dramatic slump in a security’s price. Capital still had to be held against general interest rate risk, but the overall capital burden was still dramatically lower than for other securities.

Stripping sovereign debt of that status was never going to be easy – and it was not until the Basel Committee on Banking Supervision issued guidance on the new incremental risk charge (IRC) in 2010, that the plan became clear.

“Even if certain sovereign bonds are subject to a risk weight of 0% under the standardised approach, they cannot be considered as free of default and migration risk... Accordingly, they will attract a capital charge under the IRC, except where the output of the model happens not to imply a capital charge for these positions,” the document reads.

That stance was reinforced by the European Banking Authority in guidelines it published on May 16. When listing assets that have to be included in the IRC, the document particularly mentions “bonds issued by central governments (“sovereigns”), even in cases where the application of the standardised approach would result in a 0% risk charge for specific interest rate risk”.

Guidance is one thing, of course, and legislation another – so it’s worth noting that similar language also appears in the Council of the European Union version of the fourth Capital Requirements Regulation, which transposes Basel Committee standards into EU law. The text was agreed by the council on May 15.

“The internal IRC model shall cover all positions subject to an own funds requirement for specific interest rate risk, including those subject to a 0% specific risk capital charge,” the text states.