No Country for Old Risk Models

Congress and taxpayers agree on one thing: Financial firms cannot continue to make large, risky bets and expect a government bailout. Survival in 2010 will hinge on more-effective risk management processes and technology.

Why It's Important: If another failure on the scale of Lehman occurs, bailouts may not be so readily available, and Wall Street firms will have to be able to inform regulators about their exposures across asset classes and geographies. Nassim Taleb, author of "The Black Swan," testified this fall before Congress: "There are many variations of Value at Risk [the commonly used method for projecting the losses an institution could suffer on a given day] out there, and for me, they are all equally defective." Even methods meant to improve the standard VaR, such as "expected shortfall" or "conditional VaR," are insufficient, he said, because past losses do not predict future losses. "Stress testing is also suspicious because of the subjective nature of a 'reasonable stress' number -- we tend to underestimate the magnitude of outliers."

Adds Irene Aldridge, managing partner and quantitative portfolio manager at Able Alpha Trading, a proprietary trading and investment consulting company, "The major problem with VaR is it underestimates the tail risk." VaR has been extended in many ways to reflect liquidity risk and take into account operational risk and basic stop losses, but if there's an extreme change in price, VaR is thrown off, she says. VaR needs to be supplemented with other risk management efforts, such as considering worst-case scenarios, experts agree.

Where the Industry is Now: Many firms are jumping right back into high-risk products -- for instance, by selling securitized insurance policies (referred to as "collateralized death obligations") to investors and hedge funds. But at the same time capital markets firms are installing more risk mitigation technology and starting to rethink risk modeling tools, including VaR.

As the cost of high-speed computing has dropped, risk modeling has become almost commoditized and fewer humans are needed to run them, says Aldridge. "You can price complex derivatives in a matter of microseconds, whereas 10 years ago it took several days," she explains. Nonetheless, Aldridge adds, because of the expense, "On Wall Street we've seen significant resistance to applying technology toward minimizing risk."

Focus in 2010: Wall Street needs to put more-sophisticated models in place next year, argues John Lietchy, associate professor of marketing and statistics, Pennsylvania State University. "Firms can benefit from more-sophisticated time series models of correlation structures and dependence structures between market places," he contends. "One major firm does its Value at Risk calculations using a four-year rolling window -- they just go back four years and every day they throw one day off the end of the pile and add one onto the front. An event can roll through for four years, and then suddenly it's gone." Further, VaR models often don't account for the extreme correlations that occur during a market crisis or a disruption, Lietchy adds.

But Paul Wilmott, a researcher, consultant and lecturer in quantitative finance, cautions against too much model sophistication. "The more sophisticated your tool, the greater potential for pretending there's no risk," he says. More important is to have risk managers question existing models and apply real-world knowledge to them, Wilmott asserts.

Able Alpha Trading' Aldridge says she expects to see risk management advances in the areas of portfolio management, high-frequency trading and the risk ratings of loans.

Industry Leaders: Aldridge says Citi has been doing a "phenomenal job" of making good use of risk management technology and that much of the credit should go to Shakil Ahmed, the firm's cohead of electronic trading. "They have a dark pool that's state-of-the-art and they identify people who are engaging in predatory behavior; if they're using the pool to prey on other market participants, Citi will impose fees on those people," Aldridge says. At the same time, Citi has dramatically cut technology costs by using the R programming language to create risk models.

Technology Providers: The R Project for Statistical Computing, The MathWorks (usually referred to as Matlabs), Numerix and Quantifi all provide modeling solutions that can be applied to risk assessments; RiskMetrics Group and SunGard are among the providers of risk management applications.

Price Tag: Varies tremendously. Companies are starting to use free, open source tools, and many perform calculations on an Excel spreadsheet. But the high-performance hardware, programmer time and quant time that is needed to effectively address risk is expensive.