How automated financial news is changing quarterly earnings coverage

Jun 06, 2018
Traditional news organizations are exploring the power of technology to supplement humans’ quarterly earnings coverage. What does that mean for IR teams?

Quarterly financial reporting has long been the bane of many business journalists’ existence. Trawling through earnings reports to input figures into a relatively proforma article template can be dry and monotonous – as if investor relations professionals need to be told. That’s why news organizations singled this out as an exercise that could be automated, at least in part. In doing so, they have created the welcome prospect of greater coverage, but also the fear of potential inaccuracies in such sensitive reporting.

It’s an area that has spawned several academic reports that – depending on which one you read – suggest these robo-stories can move a stock price, increase liquidity or expose retail investors to greater risk. The question remains: are the bionic newsrooms of tomorrow entirely well suited to the quarterly news releases of today? For IROs, that means keeping a close eye on developments and your company’s coverage.

The Associated Press (AP) made headlines in fall 2014 when it started publishing automated earnings stories. Four years later, financial journalism assisted by artificial intelligence (AI) has also been trialed by Dow Jones, Bloomberg, Reuters and many others.

AP’s coverage of earnings releases has expanded from an average of 300 per quarter in 2014 to 4,700 in the first quarter of 2018, according to data from AP’s partner Automated Insights. The majority of US public companies now receive some form of earnings coverage on the AP newswire, along with a growing number of Canadian companies, says a spokesperson for Automated Insights.

Automated uptick

You can spot an auto-generated story by its footnote, which reads: ‘This story was generated by Automated Insights using data from Zacks Investment Research’. AP’s model was the subject of a survey by academics at the universities of Stanford and Washington last year, which found that these robo-stories could lead to an increase in trading volume of about 38 percent.

‘What’s nice for smaller caps is that I see news organizations like CNBC, Yahoo Finance and local news outlets carrying these automated stories, whereas these outlets might not publish a press release,’ says John Viglotti, vice president of investor relations products and services at Cision/PR Newswire. ‘A lot of our smaller-cap clients are looking at investor visibility to improve liquidity.’

Lisa Gibbs, director of news partnerships at AP, says these earnings stories are more accurate now than when they were written by humans. Due to the efficiencies created by accuracy and the volume of output, AP estimates that automating these stories frees up time for three full-time journalists to work on more in-depth reporting and investigative projects throughout the year.

AP’s process for automated stories isn’t universally admired at public companies, however. ‘Automated earnings reports are a mixed blessing in the eyes of some industry experts,’ observes Neil Hershberg, senior vice president of global media at Business Wire. ‘While they certainly provide greater visibility to small and mid-sized companies that were previously excluded from editorial coverage, the template format of these reports can often result in material information being left out of stories.’

The issue of nuance

AP’s earnings stories are concerned with revenue, earnings per share, adjusted revenue and stock price performance. The data is extracted from earnings releases by humans and then input into essentially an earnings-story conveyer belt. These stories are less than 100 words in length and are intended as brief snapshots to be disseminated through a newswire. Gibbs explains that where these stories begin to go awry is when there’s a complicated company story. M&A transactions and non-Gaap figures are two of the biggest sources of friction.

Industry observers say numerous companies, including Netflix and Under Armour, have faced situations where a tech-driven news service has produced a story in which the figures are correct but there is insufficient context to explain that they are due to unusual circumstances. Representatives from both companies declined to comment for this article.

The impact of such reports can be felt particularly sharply by small caps. Telaria (formerly Tremor Video) is a small-cap US stock specializing in video monetization. In February 2016 the company had to restate part of its net and gross revenue. An AP story reported a 30 percent fall in revenue, but ‘it wasn’t an apples-to-apples comparison,’ says Andrew Posen, vice president and head of IR at Telaria. ‘The stock dropped by 15 percent. I had to spend a significant amount of time explaining to my board of directors and CEO what had happened.’ Posen adds that the AP story wasn’t incorrect – it just lacked the context of the restated revenue figures, compared with the previous quarter.

If an earnings story is inaccurate, Gibbs says issuers should contact the news desk at AP, which routinely updates the stories. She adds that the AP team will consult Zacks Investment Research and pay attention to what the Street is saying through consensus models to determine whether a story needs correcting. She did not comment specifically on Telaria.

A small problem

This throws up a potential point of vulnerability for small-cap companies looking to correct stories, as Posen illustrates: ‘We have two analysts covering us and the gap between their consensus numbers is relatively wide – so the numbers aren’t as relevant as the numbers we would guide to.’

Accurate sell-side research coverage is a perennial challenge for small caps, even without the projected disruption that Mifid II will prompt. From April 2018 onwards, AP changed the headlines of the automated stories to call them ‘earnings snapshots’ rather than reflecting a positive or negative trend from results. This serves as a form of disclaimer that these stories aren’t a source of in-depth financial journalism, Gibbs says. ‘Anyone expecting to truly understand a company’s business and financial picture would be doing far more due diligence than just consulting a 100-word story from AP,’ she adds.

In Telaria’s case, however, Posen says both retail investors and hedge funds have traded based on misleading – but not, importantly, inaccurate – earnings stories. Recently he knew there was a figure in Telaria’s earnings release that might make a splash, so he prepared his management team and proactively called around to some of his retail investors to make sure they understood the story.

Gibbs says AP would consider turning off automated stories for a specific company if there was a good case for it and a consistent pattern of stories that required correction. Indeed, she says AP has decided to selectively turn off automated stories when a certain industry has faced accounting challenges that could result in misleading stories. This was most recently the case during the first earnings season after the US tax reform was passed, when many companies took large one-time hits to overseas revenue.

Partly in response to the Stanford and Washington universities study that linked AP’s stories to an increase in liquidity, Nicholas Guest, a PhD student at the Massachusetts Institute of Technology, examined the market effects of earnings stories published by the Wall Street Journal. He looked specifically at stories covering the S&P 500 and his research finds that stories with more original reporting and less content reproduced from an earnings release increase trading volume and improve price discovery.

Guest notes that WSJ articles will likely have a broader audience, one that includes potential retail investors, than newswire coverage from the likes of AP and Reuters, which tend to feed into investor terminals such as Bloomberg.

‘As most readers of earnings coverage are also investors, or at least potential investors, any value relevant information in an article will likely be reflected in their subsequent trading and ultimately in market prices,’ Guest points out in his research paper. ‘Absent a press article, some or all of these investors would have to gather and process the information themselves.’ The impact on trading and price discovery would therefore be delayed.

Essentially, automated stories can lead to greater liquidity, but stories with added input from journalists can also yield price discovery. Although Guest’s study was concerned with the WSJ ’s output, it’s worth noting that all other publications mentioned in this article as using some form of automated earnings reporting – such as AP, Bloomberg and Reuters – still operate full newsrooms that contribute analysis and reporting to earnings stories.

What machines do best

At Reuters, Reginald Chua, COO of editorial, is experimenting with new ways of deploying machine learning and AI to enhance financial journalists’ reporting. Both of these tools are focused on using technology solutions to scan data for trends and leads that could be used by reporters – creating what Chua calls the ‘cybernetic newsroom’.

The most recent tool – Lynx Insight – scans market data and looks for trends that might be of interest to journalists. Once a trend has been identified, it is run through a natural language generator to produce a sentence, which is then sent to the journalist covering that specific company.

‘It will send something to a journalist writing about IBM with the basic facts, like what the market closed at,’ Chua says. ‘But then it will say things like, The market closed up for five days in a row, the first time it has done so since January. Then it can go on and say the number of analysts who rated the company a buy and how that has changed in the last six months.’

This tool is being used by the Reuters stock buzz and markets teams and will soon be used by the companies team to monitor pre and post-earnings activity. In a recent blog post about Lynx Insight, Chua speculated about the potential future uses of this technology for individual investors, such as personalization.

What this kind of personalization could do to impact liquidity or trading activity isn’t known, but both academic studies referenced in this article make a case for financial disclosures having some effect on stock performance. Both Chua and Gibbs say they aren’t ready to let this kind of automation loose on the world without being filtered by a human just yet. But there’s also a sense that this brave new world of business journalism isn’t too far away.

Technology such as this poses a potential information asymmetry between what a financial journalist knows and what the IR and communications teams know about their own company performance. As AI tools improve to bring investor relations professionals up to par with the sophisticated technology used by Wall Street, it’s also worth keeping an eye on similar advances from newsrooms.

 

How AP’s automated stories work

‘I have had queries from companies and have had to correct the perception that robots were scraping earnings releases,’ says Lisa Gibbs, director of news partnerships at the Associated Press (AP).

  • Step 1: In 2014, AP journalists, including the standards editor, spent months agreeing on article templates that could be used by an automated service and were consistent with AP’s style guide. These templates are still used today.
  • Step 2: Once a public company releases its earnings report, humans at Zacks Investment Research review the releases and input data into a structured file.
  • Step 3: That structured file is passed to Automated Insights, a firm that uses natural language generation to combine AP’s templates and the data from Zacks Investment Research to produce a brief story that is published to AP’s newswire.

 

This article originally appeared in the Summer 2018 issue of IR Magazine.

Sign up to get stories direct to your inbox
logo-black logo-black
Loading