Skip to main content

Big Data, Big Money

Can a simple Google Trends algorithm beat Wall Street?


Now that we know a single hacked Twitter account can erase $136 billion from the American stock market in seconds, maybe it’s time to re-evaluate this whole “let the machines do the trading” strategy.

Or not. A remarkable study, published this week in Nature Scientific Reports, details how a simple Google Trends algorithm makes a better day trader than most of the suits on Wall Street.

Tobias Preis, of the University of Warwick, led a trio of researchers in designing the trading strategy. It started with a simple idea: Investors—whether skittish or bullish—make financial decisions by first gathering information. And these days, most information gathering happens on the Web. “By analyzing changes in Google query volumes for search terms related to finance, we find patterns that may be interpreted as ‘early warning signs’ of stock market moves,” Preis writes in the paper.

Had the team invested in the market in 2004 they would have realized a 326 percent profit.

The team identified 98 such “early warning” search terms, including words like “debt,” stocks,” “portfolio,” and “inflation.” Using Google Trends, which offers data as far back as 2004, the researchers were able to calculate whether relative traffic for those terms—i.e. the number of Americans searching for “debt” vs. total searches—rose or fell from week to week.

Then they crafted a simple strategy: if “debt” traffic was on the decline—meaning it wasn’t weighing heavily on Americans’ minds—they’d buy long positions in the market, hedging that investors were feeling bullish and the Dow was headed up. But if “debt” traffic was on the rise—a sign of Americans’ anxiety—they’d buy “short” positions, hedging that investors were feeling bearish and the Dow was on its way down. (“Shorting” involves borrowing equities from a broker, selling them at a high price, and then buying them back when the price falls, turning the investor a tidy profit.)

Preis and his colleagues then ran a mock trading game using their Google Trends strategy and historical search data. Had the team invested in the market in 2004 and, for seven years, made weekly bets—those long and short positions—based only on the popularity of “debt,” they would have realized a 326 percent profit. (To put that in perspective, a seven-year “buy and hold” strategy would have yielded a 16 percent return.)

Not every term was as profitably predictive as “debt”—trading based on the relative popularity of “environment,” “garden,” and “kitchen” actually lost money—but the strategy proves that big data, used correctly, might mean big money.

As for taking market advice from your Twitter feed, well, sometimes it’s best not to listen to the crowd.