InformationWeek
Defining The Business Value Of Technology
Supercomputing: Einsteins Of Info Systems
Scientists are using quantitative analysis to revolutionize Wall Street. Can they do the same for information management?By Laton McCartney
Issue date: Feb. 26, 1996
D eep in the recesses of the Los Alamos National Laboratory in New Mexico, where the first atomic bombs were designed, a team of scientists and engineers wields a formidable high-tech arsenal of supercomputers, neural networks, and visualization tools. Until recently, Los Alamos dedicated its massiv e computing power solely to top-secret government projects. While that work continues, Los Alamos' brainpower is now tapped-for a fee-by business. DuPont, General Motors, Xerox, and other companies each pay Los Alamos as much as $100,000 annually for ideas and research. Wall Street has a nickname for the scientists at Los Alamos and others like them who stray from their labs and classrooms to work in business: "quants," short for "quantitative analysts," referring to work many have been hired to do. Whether math whizzes, artificial intelligence experts, or specialists in other disciplines, a growing number of quants are taking jobs in the "real" world-mostly in financial services. "Business has become a gray-matter game," notes Bill Stone, president of Securities Software and Consulting, a Hartford, Conn., investment technology firm. And quants are using high-tech systems to play that game.
Morgan Stanley, the investment house, has a team of in-house qu ants who use visualization tools to manage risk in derivatives trading. At Koeneman Capital Management, a Singapore firm, quants use AI, supercomputers, and visualization for everyday trading. Swiss Bank Corp. is the exclusive backer of Prediction Co., a startup founded by two quants who use chaos theory, nonlinear regressions, and other forms of rocket science to predict the movement of financial markets.
But as quants and their technological tools join the business mainstream, they sorely test the resourcefulness-and resources-of corporate information systems managers. "This can cause the CIO headaches," says Dan Schutzer, VP of advanced technology and research at Citibank, another company that employs quants.
Quants in business aren't entirely new. Big investment banks and brokerage houses first began recruiting quants more than 20 years ago to build mathematical models and apply their number-crunching and quantitative analytical abilities to the investment arena. Then, in the 1980s, scientists began migrating to Wall Street in significant numbers.
"Wall Street was gearing up for major expansion and began to go after these people aggressively, particularly after it became apparent that their quantitative skills were transferable to the stock market," says Jay Berger, president of Pathways Executive Search Inc. in New York, which counts several Wall Street firms among its clients.
Steeped In The Classics
Many of the newcomers were Russian and Eastern European physicists and mathematicians. Some took jobs researching complex financial instruments, such as derivatives or financial futures. Others signed on as risk managers and software developers. "The Russians are steeped in classical mathematics and physics, and they're able to apply them to commercial applications," notes Stone of Securities Software, which employs several
Russian scientists to develop software for insurers and securities firms.
For example, one of Se curities Software's latest software offerings is a calculation engine that lets risk managers and chief financial officers project future interest rate shifts. It was designed by a Russian physicist who used principles in plasma physics to show the gradual diffusion of interest rates over time.
But the fall of communism has meant that Russian émigrés, previously granted political asylum almost as soon as they applied, now must wait in line with immigrants from around the world to gain U.S. visas. So Wall Street began recruiting some of America's best minds from the scientific and mathematical communities. It lured them with whopping salaries and bonuses, the dollars to develop expensive quant tools like neural nets, and the adrenaline rush of working in a frenzied environment.
Bradford Paskewitz, 37, and Hezzie Lamdan, 35, are among the quants who've benefited from Wall Street's increased need for high-tech tools. Although both have worked in the financial community for several years, Israeli-born Lamdan earned a doctorate in image and pattern recognition at New York University, while Paskewitz studied engineering and computer science at Princeton and the University of Pennsylvania.
During the late '80s, Lamdan developed a fault-diagnostic system for Citicorp. At the same time Paskewitz managed financial futures and derivative portfolios for Banque IndoSuez. They met in 1990 when both won consulting contracts to build models at Lehman Brothers, then owned by American Express. "They wanted to apply advanced technology methods to forecasting markets," Paskewitz explains. "It was a big job with the opportunity to control significant resources."
Working within Lehman's technical forecasting group, Paskewitz and Lamdan built a sophisticated neural network. Every day-and before long, several times a day-the system crunched enormous quantities of data on the futures market. "We were able to show how the market behaved and make predictions," Paskewitz says. "If they proved wrong, we'd adjust the models slightly with a learning algorithm."
The two quants also used genetic algorithms to develop forecasting models. Working with a visualization expert brought in by American Express, they also experimented with scientific visualization, technology developed by NASA for scientists examining weather conditions from satellite data on a 3-D grid. "We used different visual presentations of the futures market to identify patterns and manage risk," Paskewitz recalls.
After Lehman was sold to Smith Barney in 1993, Paskewitz and Lamdan continued developing and enhancing the ideas they'd explored at Lehman and formed P&L Financial Inc., a money-management firm. Today, P&L manages a $10 million fund, trading in futures and interbank currencies. Its tools include models that emulate various trading philosophies. "We have a simulated trader who follows market trends, another who is contrarian, and another who
is a fundamentalist and anticipates market moves," says Paskewitz.
Using real-time market data feeds, neural nets, and genetic algorithms that provide pattern recognition, P&L's simulated trading system automatically flags trading opportunities throughout the day. For example, its computers can detect a pattern that, based on the historical data in the system, has preceded market upturns 70% of the time, Paskewitz explains. "With this information in hand," he says, "we can trade accordingly."
Running To Stay In Place
So how much money has P&L's rocket science earned? Paskewitz won't say, but it's clear he and Lamdan haven't finished experimenting. "Trading in the financial markets is analogous to an evolutionary process," he says. "The more successful traders flourish and the losing traders quickly disappear. You have to continually improve and evolve strategies. It's the old 'Alice In Wonderland' syndrome-you keep running just to stay in place."
Another quant, Norman Packard, espouses a similar philosophy. Packard, a physicist, is one of several scientists who, on behalf of the investment firms funding their research, seek to refute the long-held Efficient Market Hypothesis. It holds that the financial markets are unpredictable since they assimilate information efficiently. Packard and his colleagues suspect that's bunk.
Packard and business partner Doyne Farmer were boyhood buddies who grew up together in Silver City, N.M. Both ended up studying physics in grad school at the University of California at Santa Cruz. There they participated in a groundbreaking study. They analyzed structure in seemingly random data generated by turbulent fluid measurement. Their findings, which reverberated through the scientific world, became known as chaos theory and were later chronicled in James Gleick's best-selling book, Chaos (Viking Penguin, 1987).
"We had debunked the theories in physics that no structure existed in certain kinds of seemingly random chemical or physical data," says Packard. "We set out to debunk similar economic theories saying no such structure existed in financial markets."
By the end of the 1980s the two physicists had identified a structure they concluded might be tradable, meaning investors had slightly better than a coin-toss chance of making money from their predictions. Armed with their findings, Packard and Farmer quit their jobs in 1991 to form Prediction Co. in Santa Fe, N.M. "We had developed a toolbox we could apply to any kind of data," says Packard. "We thought by applying it to the financial markets, we'd find new ways of making money."
Prediction is now one of several research firms run by quants. GK Capital in Houston and Olsen & Associates in Zurich, Switzerland, are among Packard's competitors. All use quant tools such as chaos theory, nonlinear regressions, neural nets, so-called state-space search techniques, and statist ical software to predict market trends.
"Think of all the variables that impact the yen-dollar exchange-long- and short-term interest rates in both countries, the performance of the U.S. and Nikkei stock markets," Packard explains. "With this kind of information, combined with price-stream data and auxiliary inputs, we build predictive models. These enable us to look for what are called state spaces, in which different variables interact predictably within a particular time frame."
Analytical Strength
Based on the predictive structures found by these models, Packard and Farmer make directional trades, dealing primarily in the futures market, which provides liquidity and ample historical data to support the firm's extensive modeling. The firm's efforts are backed exclusively by Swiss Bank. "We've always been strong on the analytical side, so for us this is a good fit," notes Craig Heimark, a Swiss Bank managing director and global head of information technology.
Packard is quick to note that even quant magic has its limits. "These tools don't deliver anything like miraculous results," he notes. "Even if they produced a performance that was comparable to the best hedge fund, the securities industry still would take a wait-and-see attitude because we haven't developed a long-term track record yet."
Models For Survival
Yet Packard argues that within a few years predictive technology-intelligent systems fed by reams and reams of high-frequency data-will be widely used for a variety of financial applications. He thinks two good candidates are hedging and portfolio holding strategies.
"In five years you'll see more companies developing predictive models as survival tools and devoting almost as much of their resources to financial and predictive modeling as they do to their IT infrastructure today," he asserts.
Maybe so. But meanwhile, technology managers struggle mightily to keep up w ith their quants. The demands quants place on corporate IT resources can prove a major challenge. Although most quants work in the business units, corporate IS often has to support them. "That can create tensions," says Schutzer of Citibank.
What kind of tensions? Consider that all quants have at least one trait in common: They burn up enough data to tax the most robust databases. "Often in building and testing your models, you run through enormous quantities of data," explains Packard of Prediction.
Part of the problem is that quants and their employers often need different things from their data, explains Craig Heimark, CIO and a managing director of Swiss Bank, which works with Prediction and financial organizations to focus primarily on the accuracy and accessibility of current data.
Quants, on the other hand, place serious demands on the company's historical data. "This impacts your data architecture, how data is stored and accessed, and the patterns in wh ich it is organized," Heimark adds.
"IS has to have available all the data needed to support analysis and be able to distribute it efficiently in a highly competitive environment," says Citibank's Schutzer. "That usually means bringing in data warehousing and parallel processing so that a variety of databases can be accessed quickly."
Also, some IS groups have to add substantial hardware capacity to support the compute-intensive calculation engines the quants drive. "The hardware requirements for modeling often stretch the IS budget and make it difficult to control product life cycle," Schutzer observes.
Yet such projects can provide technology managers with a chance to shine. "These decision-support tools are high profile because they're used by CFOs and risk managers to calculate the amount of risk on investments, and that's of major interest to the CEO," says Stone of Securities Software. "If the CIO can provide the support neede d to develop them, and can reconcile the gap between scientific modeling and his company's existing IT architecture, he can be a real hero."
Some quants have become heroes, too. Citibank VP Schutzer holds a doctorate in electronic engineering from Syracuse University and is a former technical director of Naval Intelligence. He's now working with quants at Los Alamos Labs to develop a security system that can immediately flag anomalies in the bank's credit-card transactions.
New Fields To Conquer
Now that quants have infiltrated the financial-services industry, they're beginning to take their computer modeling and analytical techniques into other fields. Los Alamos, for one, is working with manufacturing, technology, and energy companies.
The most visible example is the lab's $52 million High Performance Parallel Processor project, notes Bruce Wienke, a former nuclear-particle physicist and director of Los Alamos' modern user facility, the Computational Testbed for Industry. "Our goal is to make complex multidimensional simulation models accessible to U.S. industrial firms," he adds.
The project is being managed by a consortium that includes some 70 full- and part-time scientists, 15 corporate partners, and computer vendors Thinking Machines Corp. and Cray Research. The vendors are supplying the processing muscle in the form of a 1,024-node Thinking Machine CM-5 and two networked Cray T3D massively parallel processors. This hardware, combined with Los Alamos' high-speed networks, software infrastructure, and simulation software, will deliver to business the computing power once reserved for building bombs.
A semiconductor maker, for example, could use the Los Alamos setup to do an atom-by-atom simulation of chip materials, with the aim of reducing cycle times and developing more agile, adaptive manufacturing techniques.
Already, oil companies use the simulators for seismic analysis, studying basin management, and remediating chemi cal spills, Wienke says.
Cooperative Research
In what Los Alamos calls Cooperative Research and Development Agreements, its scientists are developing high-performance commercial computing applications for many major financial-services organizations and corporations. To date, Los Alamos has signed 1,000 cooperative agreements with business groups.
Los Alamos' old Bradbury Science Museum, which once displayed mock-ups of the weapons that were dropped on Hiroshima and Nagasaki, has been converted into the Computational Testbed.
Here, for fees ranging from $5,000 to $100,000 annually, representatives from DuPont, EDS, GM, Texaco, Xerox, and other companies work with Los Alamos scientists, use the labs' workstations, offices, and educational facilities, and access the massive high-performance computing power housed in the Advanced Computing Labs and Central Computing Facility.
These computing capabilities are so advanced that few of the world's largest corporations c ould ever afford to duplicate them.
Next, as part of the government's National Information Infrastructure, the computing capabilities of Los Alamos, Livermore, and Sandia-all former Defense Department labs-will be accessible via computer networks as the Virtual Lab Test Bed.
At this rate, some of the quants' most closely guarded technical secrets may soon be available to every entrepreneur with a credit card.
See sidebar "Technology For A Better Bottom Line"
InformationWeek http://techweb.cmp.com/iw
Featured Microsite
