Web surfing patterns as people read financial news can be used to make accurate predictions of stock movements from a few minutes to up to a couple of hours in advance, a new study suggests. With further development, the technique may be useful to financial authorities as they monitor markets and seek to fend off emerging crises.
A team of physicists led by Gabriele Ranco of the IMT Institute for Advanced Studies in Lucca, Italy, suspected that better stock predictions might be made by looking at more than the positive or negative sentiment expressed in an article about a particular company. Another key indicator might be how many people actually click on links to those articles, a sign of the social influence of the article and how much its readers are paying attention.
The team used data collected over a yearlong period in 2012 and 2013 from Yahoo! Finance, an online portal for financial news and data. Looking at the 100 U.S. companies with the most frequent mentions in news articles, the researchers calculated a measure of sentiment for each article. They then weighted that measure to reflect readers’ behavior: Articles counted for more if more readers clicked on links to them.
The result was a moment-by-moment signal for each company showing how sentiment and interest fluctuated during the day, and how strongly. The researchers then compared these signals with actual market fluctuations of prices, volume and volatility for the 100 different stocks. The signals offer a significantly improved predictive capacity for stock movements, even for movements only a few minutes later, the researchers conclude January 25 in PLOS ONE.
Rosario Mantegna, an expert in mathematical finance at the University of Palermo in Italy, suggests that this method could prove useful for financial authorities worried about the potential for explosive financial events —a bank run triggered by a surge of investor fear, for example. “Monitoring in this way could provide some useful indicators,” he says.
An important aspect of the new work is its ability to monitor web activity on timescales as short as a minute, says study coauthor Guido Caldarelli, a physicist also at the IMT Institute for Advanced Studies. In principle, he says, this kind of analysis could be carried out in real time if financial authorities had access to the data.
More generally, says Caldarelli, the study illustrates the potential of “big data” to provide new means to detect broad social patterns of belief —a problematic task in areas ranging from public health to political polling. “Questionnaires are slow and people don’t always give their real views,” he says. “An advantage of web data is that people tend to be more sincere when they’re browsing.”
Editor’s note: This article was updated February 2, 2016, to correct the time range used in the study, to clarify reader behavior and to correct the primary academic affiliation of the coauthor.