Bloomberg the Company

Bloomberg Anywhere Login

Bloomberg

Connecting decision makers to a dynamic network of information, people and ideas, Bloomberg quickly and accurately delivers business and financial information, news and insight around the world.

Company

Financial Products

Enterprise Products

Media

Customer Support

  • Americas

    +1 212 318 2000

  • Europe, Middle East, & Africa

    +44 20 7330 7500

  • Asia Pacific

    +65 6212 1000

Follow Us

Industry Products

‘It’s Like the Matrix:’ Twitter May Help Read Investors’ Minds

Don't Miss Out —
Follow us on:

Oct. 21 (Bloomberg) -- The millions of messages sent daily via Twitter Inc. can help predict moves in the Dow Jones Industrial Average by signaling sentiment, researchers at Indiana University and the University of Manchester said.

By scouring tweets for key words and analyzing them using an algorithm they developed to divine the mood of Twitterers, Johan Bollen, Huina Mao and Xiao-Jun Zeng said they were able to predict the daily up and down movements of the Dow during a period in 2008 with 87 percent accuracy. Now they are testing whether their results can be applied to real-time data.

“So far we have observed that some of the parameters may change over time but that our conclusion that Twitter mood data can predict fluctuations in the stock market will stand,” said Indiana University’s Bollen in an e-mail interview. “We feel that we have merely uncovered the tip of the proverbial iceberg.”

The research suggests that Twitter analysis could provide a cheap way to improve results of other computer models that use news events and other publicly available data to anticipate how investors will act. The data may even predict to some extent the magnitude of stock market moves, Bollen said.

“It feels a little like the Matrix,” said Christian Kielland, a managing director at BTIG Hong Kong Ltd. “I definitely wouldn’t discount it. Without a doubt, sentiment and psychology are huge factors in trying to predict markets.”

Flash Crash

Automated programs that trade on information other than price have been blamed in part for the May 6 “flash crash” that wiped $862 billion off the U.S. market. A June 2 plunge in the shares of Diebold Inc. may also have been triggered when an item of year-old news was misinterpreted by the growing army of computers that search for financial information, analyze it and trade instantly without human involvement, market analysts said.

Bollen’s team outlined their findings in a paper published Oct. 14 on arXiv.org, an on-line journal. The team trawled through 9.9 million tweets using two tools, one called OpinionFinder which sorts messages into either positive or negative moods, and another from Google Inc., which classifies text into six mood dimensions: calm, alert, sure, vital, kind, and happy. While the OpinionFinder data was not very good at predicting stock market movements, one of the Google dimensions, calm, was excellent.

Mood Predicts Dow

“Changes of the public mood along these mood dimensions match shifts in the DJIA values that occur three to four days later,” the team said. “The aggregate of millions of tweets submitted to Twitter at any given time may provide an accurate representation of public mood and sentiment.”

Bollen said further research would be aimed at clarifying the relationships between the mood dimensions and to develop more formal models to predict how a change in the mood of Twitterers can influence the Dow.

“With more and more information out there and available in the public space it makes sense for people to undertake this sort of Matrix building,” said BTIG’s Kielland “It would hold up better somewhere like the U.S. because a high proportion of people there are actively involved in the market. So when all those tweets are happening it’s more likely they are coming from someone who holds some stocks or other investments.”

To contact the reporter on this story: Nick Gentle in Hong Kong at ngentle2@bloomberg.net

To contact the editor responsible for this story: Darren Boey at dboey@bloomberg.net.