Technology For A Better Bottom Line

Quants use visualization, genetic algorithms, and neural networks to analyze business data

---

By Laton McCartney
Issue date: Feb. 26, 1996

With their mensa-level IQs and advanced degrees, quants bring one overriding ability to the world of business. "They make numbers tell a story," says Jay Berger, president of Pathways Executive Search.

More specifically, quants search out those trends, patterns, correlations, and anomalies in market data that can be used to make a profit or minimize risk.

One big target is what's called a market inefficiency. For instance, quant computers might spot a U.S. Treasury note that's selling for $999 in Japan but $998 in London. By trading in huge volume, quants can capitalize on this small price discrepancy.

Another example: "By analyzing historical data you may find a pattern indicating that when IBM stock goes up, Unisys' price follows suit," notes Dan Schutzer, a VP of advanced technology at Citibank. "If you're right 52% of the time, you can make a lot of money."

Quants typically rely on mathematical models that incorporate reams of historical data as well as so-called high-frequency data-information that moves across financial trading screens. These models also include other inputs such as interest rate moves.

Additionally, quants use many of the technologies and tools they brought with them from the scientific and engineering communities. Among the most common: visualization. Initially developed by NASA to track meteorological conditions, visualization systems convert complex data into pictures that range from simple charts to 3-D environments. Financial institutions use visualization as fiscal weather maps. These systems help pinpoint risks, anomalies, and market opportunities and are especially useful with complex instruments such as derivatives.

At some investment firms, including Morgan Stanley, quants have developed proprietary visualization systems-but off-the-shelf alternatives are available from vendors such as NeoVision in Pittsburgh.

genetic algorithms. Digital Darwinism applies here. Genetic algorithms evolve through reproduction, random changes, and competition until the fittest or best algorithm emerges to provide the optimal solution for a given problem. Wall Street uses them to develop search and optimization strategies. There are also so-called genetic programs that evolve through the same process and ultimately produce a "best of breed" program.

neural netwo rks. These are based on theoretical work done in the early 1940s, when scientists proved that a sufficiently large number of neurons linked to a network could solve computational problems.

Modeled after the biological nervous system, neural nets emulate the workings of the human brain. Rather than being programmed, the process elements learn from the kinds and patterns of the input they receive. Quants use them to assimilate and analyze huge sets of historical data and to make correlations.

Comments on this story?

InformationWeek http://techweb.cmp.com/iw

---