An Indiana University researcher has found it may be possible to predict the ups and downs of the stock market based on what Twitter users are saying.
Informatics professor Johan Bollen’s work uses existing psychological research to group words found in almost 10 million tweets from around the globe into six categories which describe how the user is feeling. Words can describe how aware a user is or how vital a piece of information appears to be.
Those results were then graphed, Bollen says, to see if they correlated with any existing method for judging public mood. A student suggested matching up the data with the Dow Jones Industrial Average, and then presented Bollen with a graph showing a close correlation. That did not surprise Bollen until he learned that the timeframes of the two data sets had been shifted, so that what someone tweets one day shows up in the stock market four days later.
“And it turned out that we had a prediction accuracy – 87.6 percent with regards to the up-and-down movement of the Dow Jones four days out ,” Bollen says. “So it means if the algorithm says ‘up’, four days later, the Dow Jones actually goes up in 87.6 percent of cases.”
Bollen spent Tuesday fielding calls and e-mails from benefactors looking to fund his research, and he concedes one day it may be a marketable idea, noting that top stock trading companies already use computer software to try to predict market shifts and maximize profit.
Data for the study was collected from nearly 3 million Twitter users over 10 months in 2008.