ARTIFICIALLY INTELLIGENT SIGNS OF LIFE IN INVESTING UNIVERSE: THE MARKET BEATS MOST HUMANS, BUT MAYBE NOT THESE COMPUTERS

Imagine a financial planner taking a client’s money out of Fidelity Investments’ Magellan Fund and handing it over to a cybercousin of Hal, the supersmart computer from “2001: A Space Odyssey.”

Far fetched? Not to a few money managers who think a form of artificial intelligence called neural networks can be used to make better investment decisions than human managers.

Computers, after all, can almost instantaneously crunch masses of data that a human could not begin to comprehend. And neural networks are computer programs that try to mimic the intricate structure of the brain, then one-up it by recognizing complex patterns within a large array of data no human could winnow. When they find patterns that indicate investment value — perhaps two very similar securities selling at significantly different prices — they trade, unaffected by intuition, emotion or biases.

“A flesh-and-blood analyst has time to thoroughly analyze about 35 securities each month; we’re able to monitor 4,000,” says Barry Hippensteel, a portfolio manager at NeuWorld Financial Inc. in San Diego.

It’s a world that until recently has been populated by a few adventurous institutional investors, but now is beginning to reach down to a few wealthy individuals as well. NeuWorld Financial, started in 1992, already is using neural networks to manage $15 million for high-net-worth investors. Its minimum account size: $100,000.

“Going into the 21st century without some money under management with artifical intelligence,” crows Mr. Hippensteel, “is like going into a dark room without a flashlight.”

Although various forms of artificial intelligence-based investing have been around for about 10 years, the evidence of their superiority is still thin to non-existent. And it will be many more years before there will be enough evidence to decide whether neural networks, the latest form of AI investing to be tried, is the way ahead or a dead end.

Still, true believers say taking the irrationality out of decision-making has to improve results.

“Neural networks can’t act irrationally as a person might,” says Jon Quigley, a portfolio manager at NeuWorld competitor LBS Capital Management in Clearwater, Fla., which manages $100 million in institutional and individual accounts. “It (a computer) doesn’t make investment decisions based on emotion.”

At NeuWorld, Mr. Hippensteel and his partner, Tom Berghage, who developed the neural net program, buy a morass of securities information from a local data service called Ford Investment Service Inc.

Their computer, a standard Gateway PC, chews on the 116 variables it’s been designed to call out. It decides, say, that based on Micro-soft’s historical growth pattern, the stock is a buy. If the system sees a similar pattern in another security, it will rank that as a buy, too.

Interesting, but are the systems effective? The patterns they detect are identified through a series of hard-to-understand equations. (Even Mr. Hippensteel isn’t sure how it works.) And performance, like that of any money manager, is relative.

NeuWorld’s average portfolio was up 68% in the five years ending in December, vs. 141% for its benchmark, the Wilshire 5000 stock index.

Mr. Berghage blames the performance on research and development ended just last fall, and says newly revamped neural networks are making it up to investors. So far this year, NeuWorld accounts are up about 12%, he says, while the Wilshire 5000 is up 0.21%.

The same inconsistency is true of Fidelity portfolio manager Bradley Lewis. He started using neural networks to manage his Stock Selector Fund in 1991. His returns that year: 45.95%, or a good 15 percentage points higher than the S&P. But in 1998, Stock Selector returned just 14.7%, 13.9 points below the index.

Discipline Equity, which Mr. Lewis also manages, returned 36.3% in 1991, outperforming the S&P by 4.7 points. Last year, it returned 21.8% to the S&P’s 28.6%.

Of course, there are skeptics. Jim Thames, a Frank Russell Co. analyst, says of the many quantitative strategies available, neural nets may trade on the most spurious relationships.

“Using neural networks assumes that a software program can somehow see and understand data more clearly than can an investment professional,” he says. “By its very nature, it is going to lead you to relationships that aren’t causally related to one another, which is very risky.”

Correlation doesn’t necessarily mean causality. After all, October seems to have more than its share of stock market crashes, and leaves also fall from trees in October, but one doesn’t cause the other.

“Maybe over a 100-year data series, they’ll be proved right,” adds Michael Stolper, a consultant in San Diego. “But in my observations, the ability to handle more variables hasn’t led to better decision-making. The market beats up everyone.”

Separate account manager LBS Capital Management isn’t taking the markets, or the critics’ punches, lying down. It’s tempering its computerized systems with human analysis by using a knowledge-based expert system in concert with its neural network.

Knowledge-based expert systems attempt to model the decision-making processes of someone with market insights, an understanding of what drives prices and an extraordinary track record.

downloading great brains

Example: Star portfolio manager Christine Downton of the British investment house Pareto Partners Ltd., which manages more than $20 billion in institutional assets, was flown to Hughes Research Laboratories in Malibu, Calif., several years ago. She spent weeks with the wonks modeling her global bond investment approaches and insights. That information now resides in a PC at Pareto’s London offices — and manages funds worth $350 million.

LBS’s system was modeled on the thinking of its own researchers.

“We’d agree that there are people with a certain knack for picking stocks,” admits LBS portfolio manager Jon Quigley, who says LBS sometimes prefers using the outcome of its knowledge-based system. “We’d also argue that people with real expertise and a degree of success over the long term are very few and far between.”

So far, however, LBS reports its core portfolio has delivered compounded annual returns of 20% the past eight years, vs. 24.5% for its Russell 3000 benchmark.

Perhaps the biggest obstacle facing neural nets is a general lack of understanding of them.

K.K. Quah is a senior database manager at institutional money manager Advanced Investment Technology, an AI shop that was spun out of LBS Capital and is now owned almost wholly by State Street Research and Management in Boston. He says a piece of the New York City teachers pension fund that it manages uses an AI approach and in the past nine months is two percentage points ahead of the S&P 100.

But selling the technology to the pension funds to which AIT caters is not easy. Imagine, then, the difficulty in marketing such a future even to a sophisticated retail audience.

“The moment you mention neural nets to some people, their eyes glaze over,” says Mr. Quah. “It sounds like derivative stuff. There’s an emotional element to get past.”