Pete Peterson of The San Francisco Chronicle, one of the more perceptive writers on business and data mining issues today, tells a pointed story of two giant hedge funds, Long-Term Capital Management and D.E. Shaw and Co. who tried to use data mining to find a foolproof method of playing the numbers — and failed.
Most hedge funds are in the business of protecting their investors from losses by making counterbalancing, or “hedging” transactions. This is a proven strategy, and makes consistent if not spectacular returns.
Yet these two hedge funds decided to try to build up profits by betting on securities prices. This is a departure for hedge funds, whose instincts run to protecting existing money, not going out on a limb for eye-popping profits. They used data mining to spot statistically prevailing ratios between the prices of different classes of securities, since data mining can analyze a relatively constant ratio between the prices of U.S. Treasury bonds and the prices of other bonds. The formula Long-Term and Shaw used showed that whenever one goes higher or lower than the average ratio between them the deviated ratio will eventually tend to return to it historic average.
Focusing exclusively on the one ratio, for a while Long-Term and Shaw racked up impressive gains. Sure enough, the ratio, or spread, tended not to change as overall security prices moved up or down together. Whenever the ratio deviated they bet that sooner or later it would return to its historic average. Long-Term Capital even had two Nobel Prize winners working with them.
But the danger no one realized was that the one ratio couldn’t handle other factors that changed the average between the bond prices. Everyone was so hypnotized by the one correlation data mining had showed them that they lost track of the underlying assumptions. So when the Asian crisis hit in the summer of 1998 and investors moved massive amounts of money from foreign stocks and ordinary bonds into U.S. Treasuries, the assumptions underlying their golden ratio sank into the swamp and the two hedge funds ran into the ditch.
As a result, Long-Term Capital had to stitch together a $3.6 billion bailout. D.E. Shaw’s losses in the BankAmerica debt portfolio forced the bank to take a $372 million write-off.
The point of this sad little morality play is that while data mining can provide you with usable business information, it’s not a one-shot cure-all. Companies that do best with data mining are ones that learn to use it well before they use it at all.
