There’s no way I can write an opening paragraph that explains exactly what this post is about while simultaneously providing a “wow” factor nearly as well as the title already does. Twitter can predict the stock market. Huh. That’s f*&@ing cool.
Johan Bollen and Huina Mao from Indiana University’s School of Informatics teamed up with Xiao-Jun Zeng from the University of Manchester to prove 140 characters can do much more than launch a talk show. Though the results can’t predict which stocks will rise and fall, or exactly where the numbers will lie at the ringing bell, being able to accurately predict whether the Dow Jones Industrial Average will rise or fall is a pretty neat trick. Especially since the stock market is supposed to be unpredictable.
On a day-to-day basis, the most accepted theories on the actions of the economy state that it is based on news. And because the news is unpredictable, the stock market should be as well. However, numerous studies show that the stock market does not follow a “random walk,” and can be predicted to some degree. Exactly what factors influence the economy is up for debate, but this team of researchers theorized that since people make decisions based in part on emotions, the general public mood should be a pretty good indicator on what will happen to the market.
To test this hypothesis, they turned to Twitter. Sure, there are much more accurate ways to gauge the public mood, but Gallup polls and the like take a long time to complete and they cost a lot of money. Reading tweets is free, quick and easy.
Though an average tweet doesn’t contain a ton of information, being limited to 140 characters, many of them do give an indication of how the person is feeling. Numerous platforms have been created that search for specific words in a tweet that express these feelings.
For example, OpinionFinder searches for telltale words like “I feel” and “I am” that precede emotional declarations. Then, it determines if the general gist of the post is positive and negative. By comparing the ratio of the two tells whether the general Twitter-using public is feeling up or down that day.
Another much more robust program is Google-Profile of Mood States, which uses 72 terms from an often-used psychometric questionnaire called the Profile of Mood States in order to gauge six specific types of moods: calm, alert, sure, vital, kind and happy. For this study, the group expanded these 72 terms to a dictionary of 964 to better identify tweets that would help the study.
The group then sifted through nearly 10 millions tweets from 2.7 million users over a 10 month span. Okay, so they had a computer do the sifting, but the results came out well. Just to make sure they were on the right path, the scientists decided to test their twitter mood test to two emotional days: the November Election and Thanksgiving. As expected, the latter produced a one-day spike in happiness while the former caused a three-day swing from a drop in calmness the day before the election to a reversal after the vote in conjunction with increases of vital, happy and kind. The last three indicate a public that is energized, happy and friendly on Election Day, all of which return to normal afterwards.
Now knowing that their Twitter methods could accurately gauge the public mood, the team started feeding the data into already existing predictive models for the stock market. Though most of the measured moods did not help their accuracy, two of them did. Shifts in public calmness and happiness as rated by the Google program clearly matched shifts in the Dow Jones three to four days later. In fact, with the help of this information, predictive models could guess whether the market would rise or fall 87.6 percent of the time.
The researchers haven’t put their money where their mouth is, though. Nowhere have they tried to invest money based on their methods.
Someone buy me stock in Indiana University’s School of Informatics STAT!