By JACK HOUGH
In addition to serving> as gatekeeper to the store of human knowledge, Google's (GOOG)
It turns out it can do that, too--although that's not a feature the company likes to tout. New research shows Google's public search data can be used to beat the stock market by as much as 10 percentage points per year.
Investors have limited attention, so they tend to focus on only a few stocks within a universe of thousands. To know which stocks they're paying attention to is to know which ones they'll buy, one theory goes. Measuring attention isn't easy, however. Terrance Odean at Berkeley is among researchers who've studied the matter using a variety of factors meant to either reflect or affect investor attention: recent extreme returns, abnormal trading volume, news stories, advertising budgets and so on. All have shown some predictive power, but none is an especially clean gauge of attention. News stories might not attract many readers. High trading volume might attract increased attention but could just as easily be the result of it.
A trio of professors -- Zhi Da and Pengjie Gao at Notre Dame and Joseph Engelberg of UNC at Chapel Hill -- has come up with a more direct way to measure investor attention. It relies on something most investors and indeed much of the developed world ends up staring into at some point during the day: Google's search box. Determine which stocks surfers are searching for information on, they theorized, and you'll know which ones they're more likely to buy, and thus, which are more likely to rise. After all, institutional investors are more likely to use Bloomberg terminals than Google for stock information, says Da, so Google searches likely reflect the attention of ordinary retail investors, and they are far more likely to look for stocks to buy than for ones to bet against.
There were some logistical hurdles. A search for "Amazon" could reflect interest in the retailer's shares or a desire to save the rainforest, says Da, so it's best to use ticker symbols. There again, "BABY" could be a query on shares of Natus Medical (BABY),
The results are scheduled for publication in the Journal of Finance. Stocks that saw a surge in search volume (one standard deviation, in statistics parlance) beat the market by two-tenths of a percentage point over the following week. That means an investor who could consistently identify such stocks and trade on a large number of them could beat the market by 10 percentage points a year, minus transaction costs.
Some might already be doing exactly that. Da says several investment firms have had him visit and explain the study results, and at least one indicated it has used the strategy profitably. He declined to say which one and said he received no compensation for his talks.
Ordinary investors won't trade frequently enough to take advantage of the study results, save perhaps for one finding on initial public offerings. Heavily searched ones outperformed less-searched ones by seven percentage points on their first day of trading, the study showed. Investors who have an opportunity to buy into an IPO at the subscription price, then, might want to have a look at Google Trends first.
As for the broader study results, the share price surge seen by heavily searched stocks during early weeks tends to reverse within a year, so firms seeking to trade on Google data had better be nimble.
They may want to check with their lawyers first, too.
According to Da, after Google learned that the researchers downloaded Excel files detailing search trends on 3,000 tickers a day, the search firm began limiting such downloads to 10 tickers per day per user. Da says at least one investment firm told him it worked around the limit by dynamically changing its Internet protocol address, used to identify users. If there's such a restriction now, I was unable to find it while downloading ticker data. A Google spokeswoman declined to comment.
It's easy to see why Google would want to use caution in how and whether it guards such information. Restricted ticker search data might be considered "non-public" and the Notre Dame study confirms the information is "material" to stock performance. Together, those two terms define "inside information" -- not the sort of thing Google would want to become a source of.