Researchers comb the Web for new stock market predictors
It’s an inexact science, experts say, but we’ll be seeing more of it
By: SUSAN SMITH
Date: April 12, 2016
Investors these days are more hungry than ever for information. Powerful computers chew through data using sophisticated algorithms in search of patterns that will provide a jump on the market, early indicators of economic trouble or the likelihood that a company is poised for rapid growth.
Mining what is called Big Data has led to a slew of new indicators, experts say.
“Imagine you’re interested in trading ahead of a firm’s earnings announcements,” says Ambrus Kecskes, associate professor of finance at York University’s Schulich School of Business. “You might harness rumours circulating on the Internet to try to predict what those earnings will be before they are announced.”
Other novel technological approaches include using satellites to “get a sense of where economic activity is happening, where trucks are moving, or reading heat maps of cities to see how much energy is being used,” Dr. Kecskes notes.
Providers of sentiment analysis or opinion mining – the measure of positive and negative statements about a company or product expressed on blogs and social media – have multiplied and become more sophisticated since Google Trends was launched a decade ago. Why just study Twitter or Facebook? Why not mine references from employees who work for a certain company or live near a certain plant or are known to be interested in certain products?
“You can try to determine whether development of a new product or service is taking longer than expected or will come out on time,” Dr. Kecskes says. “These are not people in a position to be leaking insider information. These are things people chat about casually, and if you know what you’re looking for and can tie it together with a firm’s economic fundamentals, maybe you can filter the data, plug it into your model and get a better sense of the earnings estimates than you would by simply relying on more publicly available information.”
This sort of analysis will become more prevalent over time, says Lee Davidson, head of quantitative research at Morningstar Inc., a Chicago-based global investment research and management firm.
“If you think about markets as information-processing machines, then any information that is pertinent, relevant or important for efficiently pricing companies should be predictive about future price movement,” he says.
Morningstar is not trying to predict the stock market in this manner but has developed a prototype text-mining algorithm to help it identify similarities in language used by companies in publicly released documents. The goal is to use the information for insight into structuring portfolios.
The key to successful predictive models, Mr. Davidson says, is relevant data and sound analysis. “When you’re looking at this kind of unstructured data, you need to be very careful about how you’re interpreting it to make sure there’s a reasonable economic hypothesis, a reasonable causal mechanism.”
That’s because if you crunch data hard enough, you’ll find all sorts of spurious relationships that are irrelevant to the task of predicting economic activity or the movement of a stock.
Adam Saunders, assistant professor in the management information systems division of the University of British Columbia’s Sauder School of Business, says he believes in the power of mining the Internet for ways to predict stock movement, but has similar reservations.
“I think there is a tremendous variation in terms of how well you can do it,” he says.
It’s hard to get information on what works, he adds, because those with winning formulas tend not to share them. “My hunch is that there are people who are doing this but that there are a lot fewer people who are going to want to talk about it. … If everyone starts doing the same thing, a lot of the gains will disappear.”
In his own academic research, Mr. Saunders has looked at the frequency of words such as “data driven,” “data warehouse” or “data centre” in the public documents of certain companies and found a correlation between their use and a company’s increased profitability down the road.
“We found that when firms had more intensive discussion of technology-related investment and practices, they did not necessarily show a significant increase in profits immediately. But we found that the increase in profits got stronger the further out we went.”
Most of this is way over the heads of individual investors, who tend not to buy satellite time or write algorithms to help them see correlations in Big Data.
“Hedge funds have been using this kind of information for years and building it into their models,” Dr. Kecskes says. “But for the average investor, it’s way beyond useful.”
|Do your investments measure up? How to assess them||How to invest a sudden cash windfall|