Sign up today and take advantage of member-only content — the kind of timely, cutting edge industry insight that only Money Management Executive can deliver.
  • Exclusive Online Only Content
  • Free Daily Email News Alerts
  • Asset Management Blogs

AI


If you're a fund manager or stockbroker in times like these, it helps to have a computer doing the thinking for you. Artificial intelligence technology, which has been used to predict markets and pick stocks for more than 15 years but still is not widespread, can insulate those managing millions from the emotional torment that can be inflicted by corporate scandals hacking at the stock market.

Humans, no matter how levelheaded, can be swayed by emotion. Machines are immune. They can't feel.

"The two biggest problems, in my opinion, in investing are the twin emotions of greed and fear. And those two emotions are not part of AI," said Shashi Mehrotra, CIO of Palm Beach Gardens, Fla.-based brokerage house The Legend Group.

Although the firm, which has $1.5 billion in assets under management, has lost a lot of money like everyone else this year, AI has helped curtail slides in portfolio performance to 8% at a time when major stock indices are free falling by 30% or more. "I'm sure if I was doing it myself, I couldn't have delivered such numbers," Mehrotra said.

He pieces together a number of AI programs to create predictive models that crunch market conditions, past and present, with company performance data. The result is a market forecast with a reasonable margin of error that no human could arrive at in the mere seconds or minutes it takes the computer. Most of the software Mehrotra uses comes from a software company called Neuroshell, based in Frederick, Md., which started offering AI solutions in 1988 for the science and business industries.

"A few years ago, we noticed so many people were buying our stuff to predict markets that we built a special product just for that," said CEO Steve Ward.

AI has been used in stock market predictions for the last two decades. In the '80s, a form called rule-based expert systems was popular. These used a set of rules to emulate an expert's decision-making process. Neural networks and genetic algorithms are now in vogue. They're described as biological because they learn like the human brain does by being repeatedly shown, rather like flash cards, company financial data and subsequent market conditions.

Computer Darwinism

More sophisticated neural networks use so-called genetic algorithms that "mate" possible solutions to problems. The latter is computer Darwinism. "We run the evolutionary process for the equivalent of eons inside the computer until we actually breed a population with some solutions in it that work very, very well," Ward said.

The technology's full potential is largely untapped, said Marc Chaikin, an independent broker. A stock broker since 1966 and a layman when it comes to technology, he's been feeding market patterns into neural networks to enhance rule-based systems for as long as the technology's been around. He says that the key is figuring out which patterns to feed the system. "You can't naively feed in information. You have to [find] patterns that reflect points of success in the past," he said.

Competitive Info

This may be why there's not more buzz about the technology after 20 years of use in the marketplace. The GIGO theory seems to apply here, too: garbage in, garbage out. Many who use AI never figure out which patterns work and are unable to get the results they seek, so they drop it.

Another possible reason for AI's relative anonymity is that those who make it work well keep their success to themselves for competitive reasons.

AI saves companies money by saving time and reducing analyst staffing needs. The software can sift through thousands of stocks and perform analyses in minutes, said Dean Kasparian, division president at AIQ Systems of Incline Village, Nev. The firm has been dealing in AI for 16 years. It was acquired by Las Vegas-based Teradata eight years ago and now gives the technology away as part of a trading system, charging $59 to $79 in royalties for a data feed from its parent. AIQ has 15,000 clients, 5,000 of which are the most active.

Advanced Investment Technology (AIT) is a Clearwater, Fla.-based asset management firm that uses neural nets and genetic algorithms for its investment strategies. The $900 billion in assets AIT manages comes from corporate pension plans, endowment foundations and high-net-worth individuals. It has used the technology since it began in 1996 and focuses on converging three types of analyses: the marketplace, sectors and individual stocks.

"The challenge is to find some relationship in the past that will, in fact, be robust enough to add value in the future," said Douglas Case, AIT's president and chief information officer.

In February of last year, AIT became part of a jointly owned subsidiary of State Street Global Advisors of Boston, called State Street Global Alliance. Although the technology is proven, it is always evolving, Case said.

"There's no reason to believe that neural nets and genetic algorithms are the be-all and end-all. It's just currently what is being applied right now."