Researchers from Indiana University have devised a method for predicting changes in the Dow Jones Industrial Average through the analysis of Twitter updates. Using two mood-recording algorithms, the Google-Profile of Mood States (GPOMS) and OpinionFinder, the team analyzed 9.7 million tweets posted between March and December 2008. They found that correlations between the calmness index, one of the six “moods” measured by GPOMS, could be used to predict whether or not the Dow Jones Industrial Average went up or down between two and six days later.
According to MIT’s Technology Review, head researcher Johan Bollen and his team are announcing this news at Indiana University today. Regarding the discovery, Bollen says his team found “an accuracy of 87.6% in predicting the daily up and down changes in the closing values of the Dow Jones Industrial Average.”
As MIT notes, that’s an incredible result. Maybe too incredible. The article then questioned some of the methodology involved with this project. For example, tweets from around the world were used instead of just U.S.-based ones, which seems an odd choice given that the intention was an analysis of the U.S. stock market. However, during 2008, Twitter’s user population was largely American, so this factor alone does not entirely discount the study.
Calmness Linked to Stock Market Changes
What’s interesting about this analysis of Twitter moods is that out of the six states GPOMS measures – happiness, kindness, alertness, sureness, vitality and calmness – it’s the last one, calmness, that’s most useful in predicting stock market changes. None of the other indices, including those from OpinionFinder – a more general positive/negative sentiment indicator – reflected any stock market changes.
The researchers admit that they don’t know why or how this selection of Twitter.com user feeds was able to make predictions so accurate, and they say more research is needed.
Using Twitter to track the stock market is nothing new: StockTwits, for example, is an online community of investors where users sign in with their Twitter account to keep track of stock-related news. The service pulls in tweets tagged with a $ before a stock symbol (ex.: $AAPL). Competing service FINIF Financial Informatics, does something similar – it gathers sentiment reports in real-time from SEC filings, news headlines and Twitter. FINIF scans all recent Twitter updates that reference a stock symbol and then measures the sentiment using a custom word list to create the “sentiment score” for a given stock.
However, neither service purports to offer stock predictions on this level based on either the news or the fluctuating “moods” of the Twitter user base. In the future, perhaps, that may change, as this sentiment analysis research continues.