Money Minded: How to Psychoanalyze the Stock Market

The tendency to throw good money after bad is another repeated theme. In a 2004 paper, "Holding on to the Losers: Finnish Evidence," Mirjam Lehenkari and Jukka Perttunen, researchers at the University of Oulu in Finland, looked at the transactions of all individual investors on the national stock market. Not only do investors hang on to stocks long after they should cut their losses, but the research showed that the tendency continues even in a bear market, where there's no relief in sight. As the ship turns over, the crew grips the railing more tightly, confident their luck will change, rather than bailing out.

Same Time Next Week

Although these findings help predict the behavior of individuals, they don't answer a money manager's questions: Where the heck is the market going, when do I get in, and when do I get out? That's what Peterson aims to address.

"In traditional behavioral finance, you find your advantages in overreaction and underreaction," Peterson explains. "You're looking for deviation from an equilibrium point based on the price at that time. But those guys are just looking at history—markets change dynamically." Instead, Peterson realized, he needed a way of tracking the mood on Wall Street as it moved.

The software Peterson developed with his unnamed partner ("he's an established Web 2.0 guy," he hints) offers a way of anticipating shifts in the market. The software churns through the daily maelstrom of blog posts and earnings reports through which the public volunteers its expectations. It scans the language they use—millions of words each day—and weights each relevant word based on its historical relationship to swings in the market. The software then makes predictions about likely reversals of fortune.

Several scholars—notably Paul Tetlock, an assistant professor of finance at the University of Texas, and Feng Li, an assistant professor of accounting at the University of Michigan—have demonstrated that tracking certain positive or negative terms ("risky," for example) in corporate reports and newspaper columns, and then tailoring stock selections to the frequency of these terms' appearance, results in a portfolio that significantly outperforms the market. Peterson hopes to reproduce results like these in the real world, searching for a wider range of signals of market sentiment in a broader variety of sources.

"We've been doing simulations since 2006," Peterson says. He began by finding roughly two dozen stocks about which there was enough available discussion, and from sources—analysts, reporters, bloggers—that had track records to monitor. The number of potential stocks was also, Peterson knew, limited by liquidity: There had to be enough active trading in them to offer a nimble way of getting in and getting out.

Then, from the list of potential stocks, he chose two that the software had identified as inspiring high expectations in investors, and two that inspired low expectations. Using an algorithm that predicts changes in value, he placed simulated bets against and for them, respectively—known as going short and going long. In effect, he was swimming against what his software revealed to be the tide of public opinion.

Peterson won't reveal which stocks he picked, but he offers a few others that the simulation analyzed. "Right when it started running, Yahoo was the most positive of all stocks" in the way it was written about in the press, "and I remember there was a very rosy article about them, one of many," he says. And then "they reported earnings, investors were extremely disappointed, and the stock dropped 20 percent. The model predicted it—basically, that Yahoo couldn't meet the crowd's expectations. Cisco, meanwhile, was something people were very pessimistic about—it was all over the media—and then its earnings met expectations and the stock jumped 15 percent. We identify opportunities where people are thinking the wrong way."