IU informatics professor explores economic implication of Twitter

A tweet about a stressful day, delicious lunch or relaxing afternoon might not have much to do with the stock market.

However, research by Johan Bollen, an associate professor in the School of Informatics and Computing at IU, supports speculation that these components could be related.

Bollen conducted an empirical study of more than 10 million Twitter posts — tweets — during 10 months in 2008.

His findings revealed a startling correlation between aggregate mood expressed on Twitter and the Dow Jones Industrial Average.

The predictions are nearly 90 percent accurate up to a week in advance of the Dow’s close.

“Really, it’s no surprise,” Bollen said of his findings. “You’ve got 140 million people on Twitter right now. That’s become a good segment of society.”

Bollen met resistance in the financial field when he and his fellow researchers attempted to publish their study. He said he thinks many people in the financial sector think of social networks as “playthings.”

However, Bollen said he believes these networks provide more than entertainment.

“If you look at the numbers — the sheer numbers — you have 500 million people on Facebook, at least 140 to 150 million people on Twitter,” he said. “These are environments that have become larger than most industrialized nations.”

Some experts believe the volume of people on Twitter doesn’t provide credibility to the study. Robert Klemkosky, a stock market expert on the Chicago Board of Trade and former chairman of the finance department at the Kelley School of Business, said Twitter users aren’t the ones effecting a change in the New York Stock Exchange.

“The fact is that trading on the NYSE is not done by people who Twitter,” said Klemkosky, who currently works in South Korea as the founding dean of a graduate business program there. “Half the trading activity on the NYSE is done by computers, called high-frequency trading, and another 40 percent is done by institutional investors, such as mutual funds, pension funds and exchange-traded funds.”

This analysis casts doubt on the results of this study, but Bollen is quick to defend his claims.
“Let me put it this way — I find it very difficult to understand why the mood states of people that are not directly invested, why that still wouldn’t have any effect on the markets,” he said. “The markets are driven by a whole bunch of things.”

Other research provides support for Bollen’s argument. A study by Werner Antweiler, a business professor at the University of British Columbia, and Murray Frank, a finance professor at the University of Minnesota, found a correlation between discussion on stock market chat rooms and individual stocks.

The study, titled “Is All That Talk Just Noise? The Information Content of Internet Stock Message Boards,” was a finalist for a 2004 national award. The professors’ research discovered that the correlation between volume of message board posts and return on the stock is stronger than many other tools of comparison, such as trading volume.

Scott Smart, the director of the Corporate Finance Academy at Kelley, said Twitter is essentially “the new chat room,” and that this finding promotes evidence of a possible correlation between Twitter and the Dow Jones.

Further individual research by Paul Tetlock, a professor of finance at the Columbia University, and Joseph Engelberg, a finance professor at the University of North Carolina, suggest similar correlations, Smart said.

Tetlock’s work shows a relationship between the words media outlets use when describing a company’s outlook and the performance of its stock. Engelberg’s research indicated a connection between the attitudes of the public toward the economy and the performance of stocks.

“Their studies seem to suggest you can predict which way a stock is going to go based on the direction of the language, basically,” Smart said.

Bollen said the greatest indicator of the market’s performance was associated to collective calmness versus collective anxiousness. He stressed his assumptions about causation were speculative and the relationship he found was purely a correlation.
However, Bollen provided reasons as to why these feelings have the greatest effect.

“Investment strategies are largely driven by the expectation of profit,” he said. “If we’re nervous, in essence that means that we’re a little more anxious and also perhaps a little more anxious about our investments. That may make us more reluctant to invest.”

A separate issue regarding Bollen’s study involves the long-term influence of Twitter on the market. Bollen’s predictions were accurate up to a week in advance, but such mood studies might not be as accurate in the long run.

“Whether you’re looking at this for a few months, or you’re looking at it for a few years, you could say, ‘What does the communication on Twitter say about the market next week, and what does it say about the market next year?’” Smart said. “That is the long-term effect. And that, I think, is almost certainly going to be zero.”

Although the effect might be negligible, Bollen argued that even a slight advantage based on the algorithm he provides could prove highly beneficial to investors.

“Let’s just say it turns out that we do this for very long periods of time, and instead of 86.7 percent accuracy, you only have 55,” Bollen said. “Well, that’s still a five percent edge. If I had a billion dollars, I would be very interested in a 5 percent improvement.”

Bollen emphasized he did not think his findings were “the trade mechanism.”
His goal, Bollen said, was to show that information about public mood contains valuable socioeconomic information. Despite claims of the prevalence of young people on Twitter and the idea that social networks are purely for entertainment, Bollen said the influence of people’s moods on these forums cannot be ignored.

“I think that that’s also an old story,” he said of the concerns about Twitter. “I think if you look at the mere scale, it’s increasingly becoming a good reflection of our society overall.”

Bollen said despite the current lack of causation, there is no reason to assume there is nothing to learn from his work.

“Sometimes, a correlation is just a correlation, but it’s a really good starting point,” he said. “We do that all the time.”