How Allfirst Gambled on Monte Carlo

By Sean Gallagher  |  Posted 2002-03-12 Email Print this article Print

WEBINAR: On-demand webcast

Next-Generation Applications Require the Power and Performance of Next-Generation Workstations REGISTER >

Online exclusive: Some observers blame the bank's big loss on inadequacies in its risk-analysis software. Do Monte Carlo simulations have a place in currency trading? (Analysis)

It's a truism—or at least it should be—that the most expensive software you can buy is the wrong software. Never has that been more painfully obvious than in the case of Allfirst Financial, the U.S. subsidiary of Allied Irish Banks, where one trader managed to lose $690 million, and conceal those losses from managers over a period of more than four years.

The scandal wasn't necessarily the result of the gaps in Allfirst's currency trading systems (detailed in a story in our March issue). However, the problems may never have occurred if John Rusnak, the trader in question, had been using better risk-analysis software.

It appears that Rusnak at least was aware of the inadequacies of Allfirst's approach. In 1999, according to a report on Feb. 17 in the newsletter Derivatives Week, Rusnak had approached at least one vendor seeking new risk-management software because he felt what he was working with wasn't adequate. The purchase was alleged to have been blocked for budgetary reasons.

The company's annual reports indicate that Allfirst used "Monte Carlo simulations with 1,000 iterations." According to Peter Davies, president of risk-managment software vendor Askari, Monte Carlo simulations are generally only used for credit-risk modeling, not for currency trading.

Askari, which is owned by the financial services company State Street, provides the RiskBook risk-management software used by Allfirst's parent, AIB. Since its acquisition by State Street, Askari has focused more on the wealth-management market and doesn't have many banking customers. However, AIB remains a big client—although the RiskBook software was never installed at Allfirst.

Monte Carlo simulations are derived from a statistical description of the data. "You try to simulate possible paths that underlying assets could take in the future," says Dr. Haluk Unal, a professor at the University of Maryland's R. H. Smith School of Business and a visiting scholar at the Federal Deposit Insurance Corp. "Once you create thousands of these paths—do the Monte Carlo—then you average the value over these paths and determine the most likely outcome."

But Allfirst, according to its 10-K filing, didn't take the average. Instead, it chose the "tenth worst observation resulting in a 99% confidence that the actual losses will not exceed this amount." By the resulting calculations, the company claimed to have no more than $2 million at risk in its foreign exchange positions in 1999 and 2000.

"Monte Carlo takes a view of the data and creates artificial data that matches that statistical description," says Davies. He compares it to what happens when you compress music off a CD into an MP3 file, then re-expand the MP3 onto a new CD. "You can never recreate what was lost in the compression. It's only as good as the compression; it doesn't give you any extra detail. Every note of the original music is lost in the compression. It can be re-expanded, but there's nothing to fill in the gaps."

For this reason, Davies says, risk management is usually done using the original market data, rather than a Monte Carlo simulation. "The one exception is in credit. Credit defaults are so rare—a company, once it defaults, no longer exists, usually. So Monte Carlo creates an artificial model for risk."

Davies is puzzled by Allfirst's use of Monte Carlo for currency trading risk. "Their use of Monte Carlo would be unusual—if they thought it was giving them a more sophisticated view of the risk, they were wrong."

Unal disagrees about the suitability of Monte Carlo to currency trading. "Monte Carlo could be used in currency risk," he says. "The underlying asset [being analyzed] could be interest rates, credit risk, or currency value."

Garbage In ...
Of course, Davies and others point out that risk-management software is only as good as the data entered into it. And if Rusnak was entering fraudulent trades to counter his bad trades on the books, there's little any software could have done to detect the danger he was placing the company in.

"Even if you have a sophisticated risk-managment system, you should still be looking at trades," says Davies. "Just because you have a risk-management system doesn't mean you have risk management."

A good example of this is Barings Bank, which was essentially bankrupted by a single trader, Nick Leeson, in 1995. The company had systems in place, but Leeson hid his losses by placing them into an account that the bank apparently didn't monitor. "If you aren't looking or questioning, what you see in a system may not make a differece," says Davies.

But if RiskBook had been installed at Allfirst, he adds, it "would have provided a richer source of information, had there been an inquiring mind."

Sean Gallagher is editor of Ziff Davis Internet's enterprise verticals group. Previously, Gallagher was technology editor for Baseline, before joining Ziff Davis, he was editorial director of Fawcette Technical Publications' enterprise developer publications group, and the Labs managing editor of CMP's InformationWeek. A former naval officer and former systems integrator, Gallagher lives and works in Baltimore, Maryland.


Submit a Comment

Loading Comments...
Manage your Newsletters: Login   Register My Newsletters