The Cutting Edge: Computing / Technology / Innovation : Darwinian Software Being Tested as Survival Aid in Investment Jungle
Stocks, bonds or Treasury bills? Charles Darwin’s theory of evolution--of all things--might just have some answers to this perennial investment question.
Investment managers are experimenting with computer software that runs investment possibilities through a simulation of natural selection, complete with mutation, crossover and that all-time favorite, survival of the fittest.
The software uses a set of mathematical techniques known as genetic algorithms. Long employed in computer science and the development of artificial intelligence for military applications, they have only recently begun to trickle into the world of finance.
“It’s bizarre-sounding stuff,” admits Richard Bauer Jr., a professor of business administration at St. Mary’s University in San Antonio, whose recently published “Genetic Algorithms and Investment Strategies” is the first book on the subject. “But the power of being able to search huge numbers of trading possibilities in an efficient manner is mind-boggling.”
The interest in genetic algorithms comes as Wall Street professionals are relying more and more on computers’ processing power to develop market-beating investment strategies and new types of financial products. Many of these high-tech money management techniques carry big risks--just ask Procter & Gamble, Bankers Trust and others that have lost hundreds of millions of dollars on computer-generated “derivatives” instruments. But there may be big rewards too.
First Quadrant Corp., a Pasadena-based investment firm, was one of the first to apply the computerized Darwinism, using the software to develop investment models for the approximately $5 billion in funds it manages.
David Leinweber, the firm’s director of research, says genetic algorithms have made his clients an additional $25 million since he started incorporating them into his investment modeling strategy over the last year--an incremental return of 0.5%. He estimates the algorithms have increased the predictive power of his model by 50%.
“The genetic algorithm starts with your ideas and expands upon them and combines them in ways that produce investment forecasting models which are better than what we started with,” he says.
A former Rand Corp. scientist, Leinweber became interested in applying the algorithm software to finance when he was working at an artificial intelligence company in the early 1980s: “People started showing up from Salomon Bros. wanting to know what all this AI (artificial intelligence) stuff was about. They bought a ton of our software,” he recalls.
A typical use of the genetic algorithm might involve developing rules for building an investment portfolio. A series of different rules--such as one designating all stocks with a price-earnings ratio of less than 15, or one picking stocks with an earnings growth of more than 40% over a given period--are fed into the computer and played against each other in a first round of “natural selection,” based on historical data.
Like successful genes in nature, the rules that do well in the first round gain greater power and influence in the second round. Then there is the mutation process, in which some rules are randomly zapped to diversify the population.
“Crossover,” the biological process in which a pair of chromosomes come together and then break apart, each taking part of the other, is also replicated in the program--a process that aims to create new and better combinations.
Eventually, the population converges, leaving one rule that--if all has gone well--will be the single best rule of the millions of computer-generated possibilities. Darwinian selection might take millions of years, but the genetic algorithms can be played out in a few hours on a powerful personal computer.
Fast PCs and low-cost computer workstations are now being used to try out a wide variety of esoteric investment strategies based on computer science concepts such as neural networks, chaos theory and expert systems.
The genetic algorithms have yet to be widely tested as an investment tool. But as to why and whether they work, Bauer takes the cosmic view:
“That’s almost a fundamental question of life there,” he muses. “To say why does the whole process of evolution seem to work as well as it does--I don’t know. It’s really just a mathematical technique for finding solutions to large, complex problems. And whether it’s an investment problem or an engineering problem or species evolution doesn’t really seem to matter.”