Study: Twitter A Predictor Of Stock Performance */?> Study: Twitter A Predictor Of Stock Performance

Posted by · March 22, 2012 3:49 pm

Image representing Twitter as depicted in Crun...With more than six years worth of data compiled on its servers, Twitter is a treasure trove of information. And while some of that information is utterly useless, a recently published study from the University of California, Riverside, shows that some of it is also priceless. Buried within its 340 million daily tweets are patterns that will predict quite accurately the trends of different stock prices.

According to the UCR team, which was led by professor Vagelis Hristidis, there is a powerful correlation between the amount that users talk about a given company and the value of that company’s stock.

Although predicting the rise and fall of stock prices was not the team’s original goal, says Hristidis, “we found there was some potential to do that.”

In order to demonstrate the connection, his team focused on the number of specific conversations taking place regarding a company in a given day. For example, a big company like Google might be the subject of several different conversations. One conversation might be about the Google’s alleged plan to make a Nexus-branded tablet, while another could be focused on the ongoing antitrust concerns over Google’s use of its web search business.

Regardless of the tone of those conversations, the study found, the raw number of these conversations had a positive correlation with the volume of trading for that company’s stock. That is, the more of these conversations—dubbed “connected components” by the team—the greater the amount of trading for that company’s stock.

A separate, yet weaker correlation also existed with regard to the actual price of that stock. More of these connected components tended to yield a higher price. However, this correlation was not nearly as strong as that of connected components and trade volume.

To test these findings, Hristidis’s team developed a Twitter-based trading model and, over a four-month period of time, outperformed the Dow Jones Industrial Average by two whole percentage points, with their model suffering a 2.2-percent loss and the Dow losing 4.2 percent. However, the study warns, the same conclusion should not be assumed if the Dow had been on the rise during that trial period.

And according to Hristidis, the reason for the correlation is even more elusive than the reliability of their trial results.

“To be honest, I’m not sure why we found this correlation,” he ponders, showing no delusions of grandeur about his findings. Furthermore, he is under the logical impression that the tone of these conversations may actually have some bearing on the strength of their newly discovered correlation.

That being said, the professor does see a potential role for the team’s findings as “just one more signal” among the many that compose a trader’s overall trading strategy.

To view the study’s findings published in their entirety, one can download them in a PDF version here.