Twitter Can Predict Stock Market

Home > News & Articles > Twitter Can Predict Stock Market

Twitter Can Predict Stock Market

October 20th, 2010 ipadoholic Leave a comment Go to comments

The romantic drum coaster prisoner on Twitter can envision the ups and downs of the batch market, a new investigate finds. Measuring how composed the Twitterverse is on a since day can foretell the citation of changes to the Dow Jones Industrial Average 3 days after that with an correctness of 86.7 percent.

“We were flattering amazed that this obviously worked,” mentioned computational amicable scientist Johan Bollen of Indiana University-Bloomington. The new results be present in a paper on the arXiv.org preprint server .

Bollen and grad tyro Huina Mao stumbled on this computational clear round roughly by accident. Earlier studies had found that blogs may be used to guess open mood , and that tweets about cinema can envision box office sales . An open source mood-tracking apparatus called OpenFinder sorts tweets in to certain and disastrous bins formed on emotionally charged words.

But Bollen longed for to erect a more nuanced romantic barometer. He used a typical psychology apparatus called the Profile of Mood States , a rapid petition that is used often in curative research or sports medicine.

The initial petition asks people to rate how keenly their feelings tie in 72 not similar adjectives, inclusive “friendly,” “peeved,” “active,” “on edge” and “panicky,” and uses the responses to portion mood along 6 dimensions: calmness, alertness, sureness, vitality, affability and happiness.

Bollen and colleagues checked a outrageous Google database to see what other difference are ordinarily used in conjunction with the initial 72 adjectives, and updated the difference to their lexicon. Then the researchers took 9.8 million tweets from 2.7 million tweeters between February and December 2008, choosen the tweets that indicated a admission of guilt of tension (tweets that enclosed the difference “I feel” or “I’m feeling,” for instance), and ran the assessment on the whole information set.

“We’re using Twitter similar to a psychiatric patient,” Bollen said. “This allows us to portion the mood of the open over these 6 not similar mood states.”

As a reason check, the researchers looked at the open mood on a few easily-predictable days, similar to Election Day 2008 and Thanksgiving. The results were as expected: Twitter was worried the day before the election, and ample calmer, happier and kinder on Election Day itself, even though all returned to normal by Nov. 5. On Thanksgiving, Twitter’s “Happy” score spiked.

Then, only to see what would happen, Mao compared the national mood to the Dow Jones Industrial Average. She found that one emotion, calmness, lined up surprisingly good with the rises and falls of the batch marketplace – but 3 or 4 days in advance.

“I sank in to my chair. That’s a flattering large result,” Bollen said. “It was one of the ‘Eureka!’ moments.”

But this startling interdependence mentioned nothing about either Twitter could be used to discuss it the future. To assessment that idea, the researchers lerned a machine-learning algorithm to envision either the batch marketplace would go up or down, first using only the Dow Jones Industrial Average from the past 3 days, then inclusive romantic data.

The algorithm did flattering good using batch marketplace information alone, presaging the figure of the batch marketplace with 73.3 percent accuracy. But it did even improved when the romantic information was added, reaching up to 86.7 percent accuracy.

“Including this mood information leads to aloft accuracy,” Bollen said. He stressed that their algorithm is rarely simplified, and not the most appropriate batch marketplace predictor any person could advance up with. But “we’re supposed on the basement of what you found, if you have a few type of super-duper algorithm and you increase our time series, its correctness will go up, as well.”

The fact that Twitter mood could envision the batch market’s movements even in the center of 2008 is moreover significant, Bollen added.

“This was probably one of the most tough durations to predict,” he said. “We had a presidential election, you had what looked to be financial Armageddon, you had the beginning of what has been the deepest and paramount retrogression since the 1930s… If our algorithm was able to envision Dow Jones Industrial Average in that period, you figured that may settle a few type of descend baseline. It could do a lot improved in other durations of time.”

But because does it work? “The partial answer is, you do not know,” Bollen said. It’s in accord with to pretence that people’s moods will have a few outcome on their investments, he says, but more research is indispensable to figure out precisely how.

“It’s a flattering engaging result,” commented P.C. scientist Sitaram Asur of HP Labs. But even even though the interdependence is there, Asur is demure to think that the moods prisoner on Twitter can result in the batch marketplace to change. Not everybody on Twitter plays the batch market, he notes, or even lives in the United States. And he would similar to to see the algorithm used on tweets from a wider camber of time.

“If it is true, if you can obviously find this interdependence to be consistent, that will be a really critical result,” he said. “But correct now, I would be prudent about adage how critical this is.”

Bollen agrees that the result has a few shortcomings. “We must be spread this,” he said. The next step, he said, is to “put a few of our allowance where our mouths are, and try to do this in actual time.”

Image: flickr/ Perpetualtourist2000