Hurricane Andrew, which ravaged South Florida in 1992, drove several local insurers into bankruptcy and threw a scare into the big national firms. Payouts of $15.5 billion evaporated a huge portion of the insurers' liquid assets. Had the payouts been much bigger, the insurers would have been forced to liquidate--at a big cost--investments such as municipal bonds, which are supposed to be held for the long term.
Whizzes from IBM, insurance consultant Frank Russell Co., and other companies think they have a way to avoid that. They want to use a technique called stochastic programming to help insurers decide how much of their investments to keep liquid for the next extremely rainy day. While insurers already try out various scenarios for the direction of the economy, the real estate market, and even the chance of a major hurricane, they have lacked the mathematical sophistication to pull together those separate forecasts. Says IBM Research staff member Alan J. King: "Stochastic programming lets you choose a portfolio that's robust against all of the possible economic scenarios that you've generated."
Such higher math is something chief financial officers haven't generally had to worry much about. But that's changing quickly. As the 21st century approaches, the most savvy CFOs are tuning in to advances in math and computer science--from neural networks to data visualization. A CFO who fails to exploit the new tools out of fear or lack of knowledge is not just missing a good opportunity. "He could be failing his fiduciary duty," argues Andrew W. Lo, director of the Laboratory for Financial Engineering at Massachusetts Institute of Technology's Sloan School of Management.
HEDGE TRICKS. Before they can even think about using rocket science, CFOs need a more complete picture of their finances and risk exposure. For instance, CFOs are gradually learning to hedge risks in the flourishing derivatives market. But precise hedging is impeded by the fact that it is difficult to value swaps. And new rules from the Financial Accounting Standards Board will soon require companies to assign an updated market value to their derivatives every quarter. Glassco Park Inc. in Surrey, B.C., sells software that lets CFOs approximate such products' market value by extrapolating from related instruments, such as Treasury bills. Its customers include such giants as Intel, Microsoft, IBM, and McDonald's.
Then comes the trickier stuff. Even before Hurricane Andrew, Allstate Insurance Co. had begun teaming up with IBM Research on stochastic programming. The idea is to randomly generate a slew of possible scenarios for things such as weather and real estate prices, using something called Monte Carlo simulation. Then, those scenarios are fed into a computer model of the business to see what results. Finally, you can methodically work backward to figure out what steps should be taken today to achieve various objectives--say, a certain level of investment return--with the least risk of falling short on other performance measures.
As it turned out at Allstate, stochastic programming produced an investment strategy not too different from what the insurer was already doing. That's lucky, because Allstate senior management isn't ready to turn on a dime on the advice of a computer model, says Thomas M. Warden, director of investment research for Allstate Research & Planning Center in Menlo Park, Calif. Still, Warden says, people are getting more comfortable with the technology: "Over the next five to 10 years, I think clearly it will catch on."
Another kind of advanced math goes into calculating corporatewide risk exposure. Uncoordinated hedging is wasteful, since many companies have natural, internal hedges in which the rising value of one asset tends to offset the declining value of another. A simple example would be a company that imports from Japan and has a big portfolio of Treasury bonds: If U.S. interest rates fall, the appreciation of its bonds might compensate for a decline in the dollar against the yen. Figuring out such complex relationships among various financial assets requires huge computations of nonlinear algebra.
Computers can only go so far, however. Neural networks, the buzz in high finance just a few years ago, seek to replicate how a human brain works. But there's a big caveat. Neural networks, it seems, often do too good a job of discovering patterns in huge masses of historical data. They discover relationships among variables that are sheer coincidence and thus have zero predictive value. The bigger the data set and the more variables, the more likely it is that the neural network will extract meaningless patterns--for example, that worldwide profits go up whenever the Uruguay factory has a strike.
ON THE CUSP. With problems such as those in mind, CFOs want to make sure that the processing power of a computer is always tempered by the insight of a human being. One way to do that is through data visualization, which uses color, form, motion, and depth to present masses of data in a comprehensible way. In a program developed by MIT's Lo, a CFO can use a computer mouse to "fly" over a 3-D landscape representing the risk, return, and liquidity of a company's assets. With practice, the CFO can begin to zero in on the choicest spot on the 3-D landscape--the one where the trade-off among risk, return, and liquidity is most beneficial. Says Lo: "The video-game generation just loves these 3-D tools."
So far, very few CFOs are cruising in 3-D cyberspace. Most still spend the bulk of their time on routine matters such as generating reports for the Securities & Exchange Commission. But that's bound to change. Says Glassco Park President Robert J. Park: "What we have in financial risk management today is like what we had in computer typesetting in 1981, before desktop publishing." The software may never be able to precisely predict hurricane paths, but it should prevent storms from blowing holes in insurers' balance sheets.