MOLINE, Ill. - John Deere & Co. Inc., a company whose main business is as down to earth as can be, is a pioneer in the cutting-edge use of artificial intelligence technology for investing.
Since December 1992, the company's $3.5 billion pension fund has been internally managing a large-capitalization stock portfolio using neural network technology.
The effort is the brainchild of James Hall, who as an engineer in Deere's technology center returned to school to earn his doctorate in neural networks in 1990.
Three years ago, "we were applying neural networks to different intelligent machines and factory control. One day I gave a routine presentation to the CFO on neural networks, and a week later I was working in the pension fund," said Mr. Hall, who is manager of active equity investments.
Pierre Leroy, who then was vice president-finance and now is CFO, and other top managers at the earth-moving equipment manufacturer, are very supportive of the use of artificial intelligence techniques like neural networks.
Artificial intelligence-based computerized systems - incorporating neural networks, expert systems and/or genetic algorithms - can analyze a virtually unlimited number of events in a matter of minutes. A neural network is a mathematically based software model that can recognize patterns and statistical aberrations in data. It can process new information and adapt it to new input, thus learning from its mistakes.
Mr. Hall's first project was to investigate whether Deere could use "the same technology in financial things."
The pension fund has a $500 million internal, actively managed equity portfolio that holds both large- and small-capitalization stocks in both growth and value styles.
"We wanted to supplement (the equity portfolio) with another style. Can we use some of these (neural network) technologies to change styles, rotate from style to style?"
After three months of research, the answer appeared to be yes, and in the subsequent nine months through the end of 1992, Mr. Hall and two members of the investment staff developed a neural network investment system.
"All of the pieces were already in place," he said. Deere had the investment professionals, and he had the neural network expertise.
By December 1992, the $100 million high-tech, large-cap stock portfolio was "on line," he said.
Deere builds one neural network for the entire large-cap market and seeks a particular style or theme or a combination of factors to identify stocks that outperform. The model, which is built each week, continually adapts to changes in the market.
In August 1993, the pension fund also hired LBS Capital Management Inc., Clearwater, Fla., to run a $100 million portfolio of midcap stocks.
The portfolio is now shy of $110 million.
LBS builds individual neural networks for thousands of stocks. In a sense, LBS is bottom-up and Deere is top-down.
"We're doing a similar thing in two very different ways," Mr. Hall said.
Mr. Hall declined to provide performance numbers for the in-house portfolio, especially because the portfolio has no two-year track record.
He did say the portfolio tends to have high volatility from month to month and very high turnover - 400% to 500% a year. For that reason, it needs to return far more than the other Deere equities portfolios just to beat its benchmark: the Standard & Poor's 500 Stock Index plus 300 basis points.
If the neural network approach doesn't result in outperformance over a period of at least three years, "we won't do it," Mr. Hall said.
Besides transaction costs, the only costs Deere incurs in the effort are the salaries of the staff, none of whom was newly hired. The software is run on personal computers.
"We're doing ongoing research to extend (neural networks) to other things. We're working on engineering applications, other financial applications. I personally am working on other applications for investing," he said.
Mr. Hall said several other pension funds are looking at the strategy but none is using it as yet.
"I think Darwinism takes time. I don't think it will be popular in five years," Mr. Hall said.
"Two to three years ago there was a big flurry of activity in neural networks. Some people felt this will be the Holy Grail. 'We'll make money hand over fist.' They tried it and it didn't work."
Neural networks are "just tools. They're more sophisticated but not magic. ... There was a lack of understanding of what the tool was and excessive expectations. That will take a lot of years to overcome."
Still, interest is certainly growing. Mr. Hall has been invited to speak at a number of conferences on artificial intelligence.
"I think it's a good technology that deserves to be talked about. As we go out and talk about it, we've had a sharing of ideas. It's mutually beneficial."