The equity research industry has been buffeted by a number of recent trends, including the rise of passive investing as well as new regulations aimed at combating conflicts of interest at investment banks.
As a dwindling number of equity researchers are forced to cover a growing body of regulatory filings, computers may be able to pick up the slack, and offer investors an edge in understanding often overlooked textual information included in quarterly and annual reports, Vincent Deluard, global market strategist for INTL FCStone wrote in a Thursday note to clients.
Deluard collaborated with a former colleague at a quantitative research firm to develop and deploy a tool that can “train machines to read the text of [quarterly and annual reports] rather than their numbers,” he wrote.
The numerical information contained in financial reports are rigorously studied by analysts and their algorithms, but the textual content of regulatory filings aren’t so closely studied, Deluard said.
“Analysts often skip footnotes, treat the section about ‘risk factors’ as needless legalese and view the Management’s Discussion and Analysis of Financial Condition and Results of Operations as boilerplate talking points, but these paragraphs are carefully reviewed by the most senior corporate executives, lawyers, accountants, and auditors,” he said.
“The most crucial information is often hidden in rarely-read text, rather than in the heavily-scrutinized financial statements numbers: for example, it was a footnote that revealed the complex fraud hidden behind Enron’s record-beating numbers,” he said.
Deluard’s initial experiments with such sentiment analysis suggested that the count of words used in quarterly filings could help predict events like a recession or changes in the value of currencies.
For example, he studied instances of the words “slowdown” and “recession,” to see if there was a correlation between the use of those words and an economic downturn, and found that the use of such terms started to increase two years before the 2001 and 2008 recessions. “The low occurrence of these terms in recent filings is a good omen,” for the U.S. economy in the near term, he concluded.
Stocks fell sharply in late 2018, knocked down in part by fears of an economic slowdown, but have bounced back strongly in January. The S&P 500
is on track to end January with a nearly 8% monthly rise.
The practice of using machine-learning enabled sentiment analysis is in its infancy, and there are many hurdles for it to overcome before it becomes a mainstream tool for investors, like the fact that language in financial statements carries different valences than in everyday speech. (For example, the words ‘depreciation’ and ‘liability’ are inarguably negative’ terms in everyday use, but aren’t necessarily so in a financial context.
Deluard points to the recent struggles of high-profile quantitative investors as reason to believe that cutting-edge investors will be forced to find an investing edge to linguistic rather than strictly numerical analysis.
“I am convinced that training machines to read the text of financial statements is a new frontier for quantitative strategies,” he wrote.
Providing critical information for the U.S. trading day. Subscribe to MarketWatch’s free Need to Know newsletter. Sign up here.