Basic Research Can Bring a Company Profit in Unexpected Ways

As manager of active equity investments at John Deere & Co., Jim Hall handles more than $100 million in pension funds for the farm equipment giant. And he’s pretty good at it, thanks largely to a “neural network” computer program he wrote that searches for buying opportunities among some 1,200 stocks.

What’s remarkable about Hall’s story, though, is that stock picking wasn’t even on his mind when he created the software. Rather, he was helping to develop a driverless grain harvester that could teach itself not only how to navigate a field, but also how to get the most grain from every acre.

Hall’s move from tractor developer to equity analyst speaks bushels about the nature of basic research. For such research hardly ever results in products for the companies that fund it--that’s not what it’s for. Yet the imaginative use of basic research can nonetheless produce plenty of profits.

There are two kinds of research: basic research and research and development. The purpose of research and development is to invent a product for sale. Thomas Edison invented the first commercially successful light bulb, but he did not invent the underlying science that made the light bulb possible.

Basic research is something else--ostensibly the search for knowledge for its own sake. Basic research provides the scientific knowledge upon which R&D; is later built. If a product ever results from basic research, it usually does so 15 to 20 years down the road, following a period of research and development.

The companies that can afford to do basic research (and can’t afford not to) are ones that dominate their markets. They have both the greatest resources to spare for this type of activity and the most to lose if, by choosing not to do it, they lose their technical advantage over competitors. It’s cheap insurance, since failing to do basic research guarantees that the next major advance will be owned by someone else.

Since their true product is insurance, not knowledge, basic researchers in industry often find their work is at the mercy of the marketplace and their captains-of-industry bosses. In the business world, just because something can be built does not at all guarantee that it will be built, which explains why RCA first invented and then dropped the liquid crystal display.

RCA made this mid-1960s decision because LCDs might have threatened its then-profitable business of building cathode ray picture tubes. Thirty years later, of course, RCA is no longer a factor in the television market, and LCD displays--nearly all made in Japan--are everywhere.

This explains why researchers at Xerox Corp. invented in the 1970s lots of computer technology that Xerox never used. Computer workstations, networks, and graphical user interfaces were all invented by Xerox just in case the world traded in its paper for computer screens. And since the world is still hooked on paper, the only result of this research that Xerox bothered to exploit was the laser printer.

*

Now we understand why Deere, which makes lots of tractors and combines, has for years been pursuing research on how to cope in a theoretical world in which nobody wants to be a farm worker: They invented a combine that drives itself.

The harvester, which has never left its secret test field outside Moline, Ill., uses Hall’s neural nets--a type of software distinguished by its ability to teach itself--to trundle about and harvest grain with a remarkable degree of efficiency. It’s very advanced technology--but Deere hasn’t needed it yet, because there are still people willing to drive farm machinery.

So Hall and his computer program traveled cross-town, where they are now analyzing stocks--and easily satisfying Deere’s goal of outperforming the S&P; 500 by at least 3%.

Hall uses the same program to help pick stocks for his personal portfolio. “It kicks out some interesting stocks from time to time,” he said. “I hold the shares for as little as a week or up to eight weeks, depending on what the program says, and because I don’t have the safety constraints we place on the pension fund, I have been able to achieve results that are an order of magnitude better--at least 30% above the S&P; 500. With more effort, I think it would be possible to reliably beat the S&P; by 50%.”

Sounds great, but this is coming from a fellow who has a Ph.D in the application of neural networks to complex systems. Hall’s software, which took more than a year to develop and tune, is not for sale. “There is no reason why Deere would sell the software,” he says. “If someone develops software that adapts to market changes and makes reliable returns, they’d use it themselves and never sell it.”

Like most basic research, then, Deere’s neural net will never be a product for sale. It’s too busy harvesting investment profits.