PortfolioRisk+ cracks tail risk conundrum

Credit Suisse First Boston (CSFB) made available its new proprietary portfolio credit risk management tool, PortfolioRisk+, to all its clients last month – claiming that the system is the first technology able to measure how tail risk combines in credit portfolios.

Standard credit risk models – such as those based on value-at-risk or mean variance – assume asset returns are distributed normally, so that, for example, a return of 15% is as likely as a loss of 15%. But credit returns are not symmetric – they are skewed, with fat loss tails. “When you’re looking at the credit market, you’re interested in tail risk. You need to explicitly handle non-normal distributions, otherwise you’re simply missing the risk,” says Christopher Cloke Browne, London-based vice-president, global credit strategy.

Diversification was of little concern to the credit portfolio managers a few years ago – not much happened on the investment-grade market. But with the glut of big-name credit events over the past two years, diversification is in the spotlight.

CSFB’s tool has two key elements: first, forward-looking credit risk distributions are generated for each credit in the portfolio, then these distributions are combined and optimised to produce a forward-looking information ratio. The saddle-point methodology that underpins PortfolioRisk+ allows large numbers of non-normal distributions to be added up without the need for time-consuming Monte Carlo simulations.

A technical paper that Browne co-authored and published in Risk in June 2001 (‘Taking to the saddle’, page 91) was the catalyst for the new system. Credit strategists at CSFB realised the importance of the saddle-point technique discussed in the paper, though a small obstacle remained – Browne was working for BNP Paribas at the time. But such problems are surmountable, and Browne was hired by David Goldman and Evan Kalimtgis, co-heads of the global credit strategy group at CSFB.

The saddle-point methodology allows quick construction of an accurate analytical approximation of the tail of the loss distribution for a portfolio of assets with default risk. The method allows progression from models of loss events to a full portfolio loss distribution without the need for Monte Carlo simulation. Optimisations that take a matter of hours to run using the saddle-point method would take weeks to run using simulation, according to Browne. Also, measurement of loss sensitivity – that is, how adding and removing different credits in a portfolio affects the overall risk and return profile – can only be assessed when using Monte Carlo by rerunning the entire simulation. In contrast, the saddle-point technique allows one to make a direct assessment of an individual credit’s contribution to portfolio risk.

But the portfolio calculation would be of little use without accurate forward-looking measures of default risk for individual assets. The decoupling of the market price of default risk (that is, the spread) from credit ratings a few years ago made ratings a less robust predictor of future defaults. So instead, CSFB uses a modified Merton model dubbed Credit Underlying Securities Pricing (Cusp). The extra ingredient it added to the classic contingent claims analysis model is its use of the implied volatility from options on the issuer’s equity.

It took CSFB around three months to integrate the saddle-point method with Cusp, at which point the bank offered PortfolioRisk+ to a small number of its clients. Their initial feedback was extremely positive, Browne claims. However, as the Swiss bank allowed more clients access to its innovation, and back-tested it more extensively, it noticed a problem. “Cusp doesn’t account for straight event risk – it’s looking at slow systematic credit deterioration,” says Browne.

In its early incarnation, the system optimised portfolios using a straightforward risk-reward analysis. But this tended to produce greater concentrations in portfolios than most managers were comfortable with – a pressing concern in light of recent high-profile blow-ups such as Enron. To counter this problem, CSFB imposed constraints on the percentage of a portfolio that a particular asset can comprise. “Actually, back-testing showed that best performance was achieved with unconstrained weights,” says Browne. However, the performance differential was only marginal and so constraints are now used in PortfolioRisk+.

More clients are realising that an alternative to traditional VAR analysis is needed, says Frank Cerveny, a director in fixed-income credit sales at CSFB. “Clients with large portfolios are coming on board all the time. It’s an additive process for us. As knowledge of different credits is built, our analytics can be run on bigger and bigger portfolios with greater ease,” he says.

Browne adds that the firm is extending the methodology to other types of assets. These include convertibles and collateralised debt obligations, which should be straightforward, but a robust treatment of asset-backed securities such as those backed by mortgages and credit card receivables is more tricky, he says.

Browne declines to comment on whether CSFB is planning to use the methodology for managing credit risk on its own balance sheet, but given the troubles with enterprise-wide credit risk management that some firms are currently experiencing, it would be reasonable to assume that banks will look at every weapon they have in their arsenal.