In recent months, so-called meme stocks such as GameStop have confounded traditional Wall Street investors. They’ve demonstrated that social media chatter by itself can boost a stock’s price, regardless of the company’s financial performance.
In new research, Ashish Agarwal of Texas McCombs takes a deeper look at how internet discussions affect stock prices. Their effect, he finds, depends on the speed with which information flows through cyberspace about a company and its stock.
The associate professor of information, risk, and operations management has created a tool to measure information flows, and through them, capture the overall intensity of attention that a stock receives online. Adding that tool to existing models, he discovers, makes them more accurate at forecasting returns.
“There’s been anecdotal evidence that what people are talking about on social media moves stocks,” he says.
Agarwal’s point is not just theoretical. He builds a trading strategy around his tool, and he uses it to outperform the market.
Yardstick for Attention
Agarwal has long been interested in how investor attention affects stock prices. In earlier research, he measured internet searches for multiple companies at once. The more that companies were searched for together, the more their stocks tended to rise and fall together.
But internet searches were a limited way to measure the impact of investor attention. Agarwal wanted a more comprehensive yardstick, one that could look at several forms of attention and compare one stock against another.
“We wanted a holistic measure of information flows due to internet attention for an individual stock,” he says.
He took an existing tool, called Eigenvector Centrality, which mapped the connections among nodes in a network. He adapted it to map the online flow of information about public companies. The new tool scoured the internet for both traditional media, such as news stories and analyst reports, and social media, like posts and comments in investment forums.
To gauge the overall impact of that attention, the tool broke stock mentions into two types. It gave some weight to direct attention, paid to an individual stock. But it also included indirect attention, when an investor looks for information for the stock along with other stocks.
For each stock, the result was a single score, which Agarwal dubbed Eigen Attention Centrality, or EAC. A higher EAC meant there was more information flow about the stock in the market.
Forecasting the Future
Creating an attention yardstick was only half the battle. The bigger question was whether the EAC could make stock market models more accurate.
“Can we use it to predict what is going to happen in the future?” Agarwal asks.
He tested his new metric on data from the Chinese stock market, between June 2015 and July 2016. He worked with former McCombs colleagues Prabhudev Konana, now at the University of Maryland, and Alvin Chung Man Leung, Ph.D. ’14, now at the City University of Hong Kong. Other collaborators were Wuyue Shangguan of Xiamen University and Xi Chen of Zhejiang University.
For each stock, they calculated its EAC on a weekly basis. They compared its attention scores with its abnormal returns: how much it beat or trailed peer companies. The results:
- When EAC was added to a model, the model became up to 32% more accurate in predicting returns.
- Social media predicted returns more accurately than traditional media. “Even if the original information comes from an analyst’s report, it spreads more effectively when people talk about it on social media,” Agarwal says.
Meme stocks turned out to be the exception rather than the rule. Instead of raising returns for most stocks, higher levels of information flow reduced them. As EAC went up 1%, abnormal returns went down 4.4%.
That’s because the faster information spreads, the faster it’s incorporated into a stock’s price, Agarwal says. Quick traders can’t make profits from information most investors already have.
Better Crystal Ball, Clearer Payoff
The researchers devised a trading strategy to see whether low EACs would produce higher returns. They tested it on data from the Chinese stock market in the following year, 2017.
Their faux portfolio, they found, generated profits of up to 0.33% a week. Added up over the course of a year, Agarwal calls that “a pretty healthy gain.”
Real-world investors could apply similar strategies, he says, although he doesn’t plan to commercialize a tool himself. Big companies could incorporate EACs into automated trading algorithms, while small investors might buy data from infomediaries such as Bloomberg.
But the uses of EACs could extend beyond securities, Agarwal believes. In future research, he hopes to explore whether these scores can predict other trends, from product demand to real estate prices.
“There is some correlation between property prices and people looking at homes,” he says. “Conceptually, we can apply a similar idea and see whether social media can predict real estate’s rise or fall.”
“Developing a Composite Measure to Represent Information Flows in Networks: Evidence from a Stock Market” is forthcoming, online in advance in Information Systems Research.
Story by Steve Brooks