I, Robot Journalist
AI-generated reporting of corporate earnings in the financial press increases trading volume and liquidity
There’s a new name in financial journalism. And it isn’t human.
The Associated Press (AP), responding to cost pressures across the news industry, has begun publishing computer-generated articles as an inexpensive means of broader reporting on quarterly earnings of private corporations.
This “robo-journalism” has certainly expanded the coverage of such information in popular channels frequented by investors.
But does it actually make a difference in the financial markets?
A new study by Ed deHaan, Elizabeth Blankespoor and Christina Zhu demonstrates that these automated earnings articles increase firms’ trading volume and market liquidity, likely due to the response of “retail” traders.
“We wanted to learn whether a story about an earnings announcement, written by a robot, is of interest to anyone. Do investors read and respond to them?” asks deHaan, an assistant professor of accounting at the University of Washington Foster School of Business. “It turns out, they do. And this intentionally simple application of automation actually strengthens capital markets.”
Just the facts
The AP began publishing computer-generated financial articles in October 2014. For each, an algorithm synthesizes a variety of public information—company press releases, analyst reports, stock performance—and uses natural language processing to communicate the pertinent information about a firm’s quarterly earnings in straightforward prose.
The most objective of reportage, these articles convey unembellished facts, without analysis, interpretation or spin. “An algorithm simply draws data from a pipe and wraps some words around it,” says deHaan.
This elementary automation has allowed the AP to expand its coverage of corporate earnings announcements dramatically, from 400 to 4,000 per quarter by the end of 2015. Thousands of firms are receiving coverage for the first time.
And their coverage travels. These automated “wire” stories appear across the spectrum of popular financial media outlets, including Yahoo Finance, CNBC, Huffington Post and Investor’s Business Daily.
Volume and depth
Learning of the AP’s experiment in automation piqued the interest of deHaan and collaborators Blankespoor, an associate professor of accounting at Foster, and Zhu, a graduate student at the Stanford Graduate School of Business. The rapid deployment of robo-journalism would allow them to cleanly measure the impact of this new market intermediary by observing firms in the eras before and after their earnings releases received news coverage via the AP’s robo program.
Specifically, they examined the market effects of quarterly earnings announcements issued by 2,268 mid-sized public companies—with a median market cap of $250 million—from early 2012 to late 2015.
The findings were clear. A significant increase in abnormal trading volume—11 percent, on average—followed earnings announcements covered by an AP-published automated article.
DeHaan explains that this spike in trading activity appears to be due to the response of everyday “retail” investors, who lack the resources, access and sophistication of institutional investors.
The study also indicates that automated earnings stories increase market depth, a key measure of liquidity and a key to strong markets. Depth refers to the number of firm shares that you can trade at a given price. The authors believe that this increase in depth is the product of big investors feeling assured that the beneficial increase in trading volume is not occurring because someone holds some proprietary information that they have missed. It’s simply retail investors reacting to the news.
And their news is not really “news” at all. In the time it takes to produce and disseminate automated articles on earnings announcements to the general public, the institutional investors have already acted. As a result, this robo-reporting of earnings has no effect on price discovery—the market’s ability to reach the fair value of a stock at any given moment.
“Smart money has this information within seconds of its release,” says deHaan. “So by the time that these articles appear on the Internet, the big price movements have already occurred.”
Good for the markets, but what about journalists?
As living, breathing journalists face an existential crisis in this age of automation, their computerized replacements are already offering a measurable benefit to the capital markets.
And there is little doubt that the use of robo-journalism is going to expand in scope and sophistication. “These algorithms can get much more sophisticated,” deHaan says. “They could begin processing unstructured data to deliver investigations, interpretations, even recommendations. This is coming.”
And enhanced machine learning will likely threaten the livelihood of a range of intermediaries in the financial markets, from journalists to analysts to advisors.
But deHaan cautions against writing off humans just yet.
“An algorithm is only as good as its source data,” he says. “This is easy to manipulate. And right now, humans still have an edge in adapting quickly to new trends and filtering out spurious correlations from real ones.”
He calls this study a first step in understanding the bigger picture of automation in the financial markets. “So far, we find this technology is efficiency enhancing. It’s giving media coverage to thousands of companies that have never been covered before, and it’s increasing volume and liquidity in the markets,” he says. “But the technology is nascent and the potential for errors and negative outcomes is high. We need to proceed with caution.”
“Capital Market Effects of Media Synthesis and Dissemination: Evidence from Robo-Journalism” is published online in the October 2017 Review of Accounting Studies.