"A compendium of gossip is still gossip," Samuel Johnson declared, but what if you can be early in the gossip chain, buy on it, and sell by the time it dawns on the brutish masses? Then it's prophecy. That seems to be roughly the gist of the highly sophisticated mathematical data mining behind the latest financial trend, scanning Web chatter for hints of changing sentiment that will move markets, according to a largely enthusiastic report in the New York Times. Some of the biggest media names are joining in:
[T]he development, years in the making, is part of the technological revolution that is reshaping Wall Street. In a business where information is the most valuable commodity, traders with the smartest, fastest computers can outfox and outmaneuver rivals.
"It is an arms race," said Roger Ehrenberg, managing partner at IA Ventures, an investment firm specializing in young companies, speaking of some of the new technologies that help traders identify events first and interpret them.
Many of the robo-readers look beyond the numbers and try to analyze market sentiment, that intuitive feeling investors have about the markets. Like the latest economic figures, news and social media buzz -- "unstructured data," as it is known -- can shift the mood from exuberance to despondency.
Tech-savvy traders have been scraping data out of new reports, press releases and corporate Web sites for years. But new, linguistics-based software goes well beyond that. News agencies like Bloomberg, Dow Jones and Thomson Reuters have adopted the idea, offering services that supposedly help their Wall Street customers sift through news automatically.
Some of these programs hardly seem like rocket science. Working with academics at Columbia University and the University of Notre Dame, Dow Jones compiled a dictionary of about 3,700 words that can signal changes in sentiment. Feel-good words include obvious ones like "ingenuity," "strength" and "winner." Feel-bad ones include "litigious," "colludes" and "risk."
The software typically identifies the subject of a story and then examines the actual words. The programs are written to recognize the meaning of words and phrases in context, like distinguishing between "terribly," "good" and "terribly good."
Wait a minute. Didn't we get into our present mess partly because everyone from bankers to home buyers discerned and followed waves of enthusiasm until the bottom fell out? Where did all the prophecies of people like Benoit Mandelbrot and Nassim Nicholas Taleb go? (More thoughts on Mandelbrot here.) The sources seem to think that's because our algorithms were so crude; that this time is different. The trouble with the old short-term thinking evidently was that it wasn't short enough.
Economists and psychologists have for over a decade been analyzing information cascades, in which people's observations of each other's judgments may accelerate trends for worse as well as better. These systems might turn cascades into torrents. Think, too, of the ethical quandary of journalists working for the financial news agencies offering the services, knowing that any turn of phrase may nudge somebody's machine into a decision. And to make things even livelier, speculators will be able to program banks of computers to generate and broadcast verbiage that will feed the analysis machines and move markets, as some operators have already taken advantage of the quirks of search engine algorithms.
The piece brought to mind another Times article on, of all subjects, Vladimir Putin and tiger conservation, in which an expert on human-animal interactions, Dr. Stephen Kellert,
said that humans often like to think of themselves as big cats, even though they really are more akin in their social habits to sheep.
"We are both a herd animal and predator, but the herd tendency runs deep," he said. "But we like to think we are like tigers: independent, self-sufficient and predatory."