Abstract
In 2014, the Associated Press (AP) began using algorithms to write articles about firms’ earnings announcements. These “robo-journalism” articles synthesize information from firms’ press releases, analyst reports, and stock performance and are widely disseminated by major news outlets a few hours after the earnings release. The articles are available for thousands of firms on a quarterly basis, many of which previously received little or no media attention. We use AP’s staggered implementation of robo-journalism to examine the effects of media synthesis and dissemination, in a setting where the articles are devoid of private information and are largely exogenous to the firm’s earnings news and disclosure choices. We find compelling evidence that automated articles increase firms’ trading volume and liquidity. The effects are most likely driven by retail traders. We find no evidence that the articles improve or impede the speed of price discovery. Our study provides novel evidence on the impact of pure synthesis and dissemination of public information in capital markets and initial insights into the implications of automated journalism for market efficiency.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
The Associated Press is a nonprofit cooperative owned by thousands of member news organizations. AP members pay a fee to support AP’s media production, and AP distributes this content over newsfeeds. AP members and other customers can then freely republish AP content in print, online, radio, and television outlets.
Panel B of Appendix 1 provides an example non-AP media article about Inventure Foods. The article is from August of 2012 because Inventure does not have any “earnings” or “revenue” media articles in RavenPack or Factiva for its October 2014 earnings announcement. The article is from the outlet RTT Financial News, which is one of the most frequent sources of non-AP media articles about our sample firms found in RavenPack and Factiva (see Section 3 for further discussion).
As discussed in Section 2, “news flashes” are headline-only, nonnatural-language snippets that are used by Dow Jones to quickly disseminate firm news to market professionals, typically within seconds of a firm’s disclosure; for example, “United Technologies 2Q EPS $1.84 > UTX,” which was issued within the same minute as the earnings announcement on July 22, 2014.
As discussed in Section 3.2, RavenPack Web Edition provides an incomplete, lower bound on the republication of automated AP articles because AP’s automated articles are not included in other common databases.
For example, The Wall Street Journal covered fewer than 1% of CRSP/Compustat firms’ earnings announcements from 2012 to 2015, and even the specialized business press outlet Dow Jones Newswire provided articles for fewer than 18% of earnings announcements.
The academic capital markets literature typically focuses on linguistic techniques to extract information from text, as a form of natural language processing. This study examines the rise and benefits of using natural language generation instead to convert disparate information into synthesized text that is more easily understood by humans. The importance of natural language generation for creating text and natural language processing for analyzing text is likely to continue growing.
A related literature examines the capital market effects of increased dissemination by firms (Blankespoor et al. 2014). Our study focuses on synthesis and dissemination of public information by intermediaries such as the media, rather than dissemination by original sources such as firms. However, their findings provide further support for the potential of media synthesis and dissemination to influence capital market outcomes.
Relatedly, Li et al. (2011) analyze the impact of dissemination using brief newswire articles that report extracted information from 10-K and 10-Q filings. These articles provide previously disclosed information with some delay, but they also contain private information in the form of fact selection and spin and are subject to selection concerns, as evidenced by only 9% of filings being covered.
In a few cases, missed automated stories can also occur if AP’s editors determine ex ante that something unusual is happening with a firm that might make an automated story less useful (e.g., during a major bankruptcy). AP instead assigns a journalist to cover these firms. We cannot identify these cases in our data, but AP informed us that these are rare and involve high-profile firms that would likely have received AP coverage before automation and thus are already excluded from our final sample of firms.
Zacks provides the data included in the AP articles. The automated articles are actually prepared by Automated Insights, a third-party company partially owned by AP during our sample period.
The median market value of treatment firms declines for treatment groups with later implementation dates, consistent with AP’s automation beginning with the largest firms. The exception is the 2015 Q4 treatment group of 23 firms, which includes several larger firms whose initiations of automation were likely delayed until firm- or industry-specific considerations were incorporated into the algorithm (e.g., 18 of 23 are in mining, electric utilities, or banking).
Our research design uses AP’s staggered implementation to evaluate within-year-quarter changes in market trends for firms that have begun receiving automation articles. The control firms in each quarter include not only the 34.4% of nontreatment firms never covered by automated articles during our sample period but also those treatment firms not yet covered by automated articles. Accordingly, Section IA3 of the internet appendix analyzes pre-treatment parallel trends in dependent variables not only between treatment and nontreatment firms but also between each group of staggered treatment firms. All trends are similar except for the 2015 Q4 treatment firms that comprise 1% of our sample. As discussed in the internet appendix, excluding the 2015 Q4 treatment firms produces qualitatively unchanged results for all tests discussed herein.
Dow Jones distributes articles from The Wall Street Journal, Barron’s, and Dow Jones’s own reporters. However, although The Wall Street Journal Online occasionally publishes AP’s automated articles, these articles are not contained in news archives such as RavenPack or Factiva. It is unclear exactly why AP articles published by WSJ or other non-AP outlets are excluded from RavenPack and Factiva. The most likely explanation seems to be that these databases contain only content created by those outlets’ own reporters and not republished content.
Further, even if there are general trends in media coverage over time, our empirical analyses include year-quarter fixed effects to control for overall time trends. Thus, to confound our results, any general changes in non-AP media coverage over time would have to vary systematically between our five groups of treatment firms and one group of nontreatment firms in a way that mimics the staggered implementation of AP articles.
Our manual online searches also find that many media web sites remove automated AP articles after a period of weeks or months, meaning it is not possible to retroactively measure republications based on manual web searches. Thus the lower bound from RavenPack’s Web Edition is our best, but incomplete, source of data on AP article republications.
Consistent with the work of Roberts and Whited (2013) and Bertrand et al. (2004), we use the term “difference-in-differences” simply to describe a model that compares trends in a dependent variable between different groups of treatment and nontreatment firms. We do not use the term “difference-in-differences” to imply that the treatment is random. The identifying assumption underlying our model is that, in the absence of automated articles, the temporal trend in Abn_Vol would have been the same between groups of staggered treatment firms as well as nontreatment firms. Section IA3 of the internet appendix investigates the validity of the “parallel trends” assumption for Abn_Vol and all other dependent variables.
We note that electronic trading in recent years has reduced firms’ bid-ask spreads (Harris 2015), which means that depth is likely now more sensitive than spreads to changes in a firm’s trading environment.
An alternative approach to investigating the speed of price discovery at earnings announcements would be to examine whether automated articles are associated with differences in earnings response coefficients and post-earnings announcement drift. We use IPT because it has precedent in the media literature (e.g., Twedt 2016) and because IPT has the advantage that it does not condition on a specific signal (e.g., analyst-based earnings surprise).
Our IPT_Adj formula includes the simplifying assumption that daily returns accrue at the beginning of each day. Using a more complex assumption that returns accrue evenly during the day creates an adjusted IPT measure that is 99.3% correlated with our simplified version and produces qualitatively unchanged results for all of our tests. Section IA4 of the internet appendix provides formulas and links to code to calculate both versions of adjusted IPT.
Volume from dark pools is included in TRF (Kwan et al. 2015). Thus, to exclude institutional trades in the “D” TAQ trades, we rely on the tendency of brokers to route retail trades off-exchange for a small price improvement and exclude trades that do not receive these small price improvements. Specifically, institutional trades usually occur at the penny or half-penny, so we exclude trades transacted at the round penny or around the half-penny (0.4, 0.5, or 0.6 cents to allow for negotiation around the mid-quote by some dark pools) (Boehmer et al. 2016).
Marketable limit orders are limit orders to buy (sell) shares at a price greater than (less than) or equal to the National Best Offer (National Best Bid). Essentially, these are limit orders that can be transacted immediately. Many brokers will route nonmarketable limit orders to public exchanges (Battalio et al. 2016). Thus these retail trades would not be in TAQ with exchange code “D.” Boehmer et al. (2016) note that about half of retail trades are limit orders. We don’t have a sense of how many limit orders are marketable versus nonmarketable, as the SEC’s Rule 606 does not require brokers to report separate routing information for marketable and nonmarketable limit orders (Battalio et al. 2016). In our setting, retail investors who trade in response to an AP article in one quarter potentially set nonmarketable limit orders. Thus excluding nonmarketable limit orders likely underestimates the increased trading volume attributable to AP articles.
AP maintains a hand-selected list of exceptionally high-profile firms that always receive human-written earnings announcement articles. In our “available sample,” 210 firms (related to 3166 firm-quarters) receive 100% human AP coverage before automation. As shown in Panel C of Table 1, these 210 firms have a median market value of $26.5 billion, which is several times larger than the median market value of firms receiving occasional human AP coverage. Eighty-seven percent of these 210 firms begin receiving automated articles in the fourth quarter of 2014, and the remaining 13% were likely intentionally delayed for firm-specific reasons. We exclude these exceptionally high-profile firms from our analyses because they are not amenable to a difference-in-differences design exploiting staggered implementation and because they are highly dissimilar from our other treatment firms.
References
Barber, B. M., & Odean, T. (2008). All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors. Review of Financial Studies, 21(2), 785–818.
Battalio, R., Corwin, S. A., & Jennings, R. (2016). Can brokers have it all? On the relation between make-take fees and limit order execution quality. The Journal of Finance, 71(5), 2193–2238.
Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-differences estimates? Quarterly Journal of Economics, 119(1), 249–275.
Blankespoor, E. (2016). The impact of information processing costs of firm disclosure choice: Evidence from the XBRL mandate. Working paper.
Blankespoor, E., Miller, G. S., & White, H. D. (2014). The role of dissemination in market liquidity: Evidence from firms’ use of twitter. The Accounting Review, 89(1), 79–112.
Bloomfield, R. J. (2002). The “incomplete revelation hypothesis” and financial reporting. Accounting Horizons, 16(3), 233–243.
Boehmer, E., Jones, C. M., & Zhang, X. (2016). Tracking retail investor activity. Working paper.
Bushee, B. J., Core, J. E., Guay, W., & Hamm, S. J. (2010). The role of the business press as an information intermediary. Journal of Accounting Research, 48(1), 1–19.
Bushman, R. M., Smith, A. J., & Wittenberg-Moerman, R. (2010). Price discovery and dissemination of private information by loan syndicate participants. Journal of Accounting Research, 48(5), 921–972.
Butler, M., Kraft, A., & Weiss, I. S. (2007). The effect of reporting frequency on the timeliness of earnings: The cases of voluntary and mandatory interim reports. Journal of Accounting and Economics, 43(2), 181–217.
Cohen, L., & Lou, D. (2012). Complicated firms. Journal of Financial Economics, 104(2), 383–400.
Dai, L., Parwada, J., & Zhang, B. (2015). The governance effect of the media’s news dissemination role: Evidence from insider trading. Journal of Accounting Research, 53(2), 331–366.
De Chaisemartin, C., & d’Haultfoeuille, X. (2015). Fuzzy differences-in-differences. Working paper.
deHaan, E., Shevlin, T., & Thornock, J. (2015). Market (in) attention and the strategic scheduling and timing of earnings announcements. Journal of Accounting and Economics, 60(1), 36–55.
deHaan, E., Madsen, J., & Piotroski, J. D. (2017). Do weather-induced moods affect the processing of earnings news?. Journal of Accounting Research, 55(3), 509–550.
Diamond, D. W., & Verrecchia, R. E. (1991). Disclosure, liquidity, and the cost of capital. The Journal of Finance, 46(4), 1325–1359.
Dougal, C., Engelberg, J., Garcia, D., & Parsons, C. A. (2012). Journalists and the stock market. Review of Financial Studies, 25(3), 639–679.
Drake, M. S., Guest, N. M., & Twedt, B. J. (2014). The media and mispricing: The role of the business press in the pricing of accounting information. The Accounting Review, 89(5), 1673–1701.
Drake, M. S., Thornock, J. R., & Twedt, B. J. (2017). The internet as an information intermediary. Review of Accounting Studies, 25(2), 543–576.
Easley, D., & O’Hara, M. (1992). Time and the process of security price adjustment. The Journal of Finance, 47(2), 577–605.
Engelberg, J. E., & Parsons, C. A. (2011). The causal impact of media in financial markets. The Journal of Finance, 66(1), 67–97.
Factset. (2017). Narrative Science and FactSet Unveil Alliance for Investment Management Firms. http://www.factset.com/news/narrative-science-and-factset-unveil-alliance-for-investment-management-firms. Accessed 22 Feb 2017.
Fang, L., & Peress, J. (2009). Media coverage and the cross-section of stock returns. The Journal of Finance, 64(5), 2023–2052.
Fang, L. H., Peress, J., & Zheng, L. (2014). Does media coverage of stocks affect mutual funds’ trading and performance? Review of Financial Studies, 27(12), 3441–3466.
Finley, K. (2015). In the Future, Robots will write news that’s all about you. Wired. https://www.wired.com/2015/03/future-news-robots-writing-audiences-one/. Accessed 22 Feb 2017.
Grossman, S. J., & Stiglitz, J. E. (1980). On the impossibility of informationally efficient markets. The American Economic Review, 70(3), 393–408.
Gurun, U. G., & Butler, A. W. (2012). Don't believe the hype: Local media slant, local advertising, and firm value. The Journal of Finance, 67(2), 561–598.
Harris, L. (2015). Trading and electronic markets: What investment professionals need to know. Research Foundation Publications, 2015(4), 1–79.
Hirshleifer, D., & Teoh, S. H. (2003). Limited attention, information disclosure, and financial reporting. Journal of Accounting and Economics, 36, 337–386.
Hirshleifer, D., Lim, S. S., & Teoh, S. H. (2009). Driven to distraction: Extraneous events and underreaction to earnings news. The Journal of Finance, 64(5), 2289–2325.
Holden, C. W., & Jacobsen, S. (2014). Liquidity measurement problems in fast, competitive markets: Expensive and cheap solutions. The Journal of Finance, 69(4), 1747–1785.
Huberman, G., & Regev, T. (2001). Contagious speculation and a cure for cancer: A nonevent that made stock prices soar. The Journal of Finance, 56(1), 387–396.
Kwan, A., Masulis, R., & McInish, T. H. (2015). Trading rules, competition for order flow and market fragmentation. Journal of Financial Economics, 115(2), 330–348.
Lecompte, C. (2015). Automation in the newsroom. Nieman Reports. http://niemanreports.org/articles/automation-in-the-newsroom/. Accessed 22 Feb 2017.
Lee, C. M., Mucklow, B., & Ready, M. J. (1993). Spreads, depths, and the impact of earnings information: An intraday analysis. Review of Financial Studies, 6(2), 345–374.
Li, E. X., Ramesh, K., & Shen, M. (2011). The role of newswires in screening and disseminating value-relevant information in periodic SEC reports. The Accounting Review, 86(2), 669–701.
Mathieson, S. A. (2015). Natural language generation progresses from robo-journalism to finance. ComputerWeekly.com . http://www.computerweekly.com/feature/Natural-language-generation-progresses-from-robo-journalism-to-finance. Accessed 1 Aug 2016.
Merton, R. C. (1987). A simple model of capital market equilibrium with incomplete information. The Journal of Finance, 42(3), 483–510.
Miller, G. S. (2006). The press as a watchdog for accounting fraud. Journal of Accounting Research, 44(5), 1001–1033.
Miller, G. S., & Skinner, D. J. (2015). The evolving disclosure landscape: How changes in technology, the media, and capital markets are affecting disclosure. Journal of Accounting Research, 53(2), 221–239.
Odean, T. (1998). Volume, volatility, price, and profit when all traders are above average. The Journal of Finance, 53, 1887–1934.
Peress, J. (2014). The media and the diffusion of information in financial markets: Evidence from newspaper strikes. The Journal of Finance, 69(5), 2007–2043.
Roberts, M. R., & Whited, T. M. (2013). Endogeneity in empirical corporate finance. Handbook of the Economics of Finance, 2, 493–572.
Rogers, J. L., Skinner, D. J., & Zechman, S. L. (2016). The role of the media in disseminating insider-trading news. Review of Accounting Studies, 21, 711–739.
Tetlock, P. C. (2010). Does public financial news resolve asymmetric information? Review of Financial Studies, 23(9), 3520–3557.
Tetlock, P. C. (2011). All the news that's fit to reprint: Do investors react to stale information? Review of Financial Studies, 24(5), 1481–1512.
Twedt, B. (2016). Spreading the word: Price discovery and newswire dissemination of management earnings guidance. The Accounting Review, 91(1), 317–346.
Acknowledgements
We thank Mary Barth, Mark Bradshaw (discussant), John Campbell (discussant), Kimball Chapman, Stefan Huber, Charles Lee, Roby Lehavy (discussant), Russell Lundholm (editor), Sarah McVay, Greg Miller, Hal White, an anonymous referee, and participants at the 2016 Stanford Summer Camp, 2016 UBCOW Conference, 2016 HKUST Accounting Symposium, 2017 FARS Midyear Meeting, University of Wisconsin, and the 2017 UC Davis Accounting Conference for helpful suggestions. We are sincerely grateful to the Associated Press for its extensive support. We thank Christina Maimone for programming assistance and Shauna Bligh and Mae Bethel for research assistance. Financial support was provided by the Stanford Graduate School of Business and University of Washington Foster School of Business. All errors are our own. Additional analyses mentioned in this paper can be found in the internet appendix.
Author information
Authors and Affiliations
Corresponding author
Electronic supplementary material
ESM 1
(DOCX 281 kb)
Appendices
Appendix 1: Example Articles
Panel A: AP Automated Article.
Below is the automated AP article following Inventure Foods’ earnings announcement at approximately 8 am on Oct. 30, 2014.
Inventure Foods misses Street 3Q forecasts
Inventure Foods posts 3Q profit, results miss Street Expectations
October 30, 2014 10:40 am
PHOENIX (AP) _ Inventure Foods Inc. (SNAK) on Thursday reported net income of $3.1 million in its third quarter.
The Phoenix-based company said it had profit of 15 cents per share. Earnings, adjusted for non-recurring gains, were 11 cents per share.
The results did not meet Wall Street expectations. The average estimate of analysts surveyed by Zacks Investment Research was for earnings of 13 cents per share.
The snack maker posted revenue of $72.6 million in the period, also missing Street forecasts. Analysts expected $73.6 million, according to Zacks.
Inventure Foods shares have climbed 2% since the beginning of the year. The stock has increased 19% in the last 12 months.
_____.
This story was generated by Automated Insights using data from Zacks Investment Research. SNAK stock research report from Zacks.
___
Panel B: Article from RTT Financial News
Below is a non-automated media article from RTT Financial News following Inventure Foods’ earnings announcement at approximately 8 am on August 2, 2012.
Inventure Foods Q2 Profit Rises
8/2/2012 11:53 AM ET
(RTTNews) - Inventure Foods Inc. (SNAK:Quote) Thursday reported an increase in profit for the second quarter, driven mainly by a double-digit growth in revenues.
The Phoenix, Arizona-based company’s second-quarter net profit was $1.62 million or $0.08 per share, compared to $859 thousand or $0.05 per share last year. On average, four analysts polled by Thomson Reuters expected the company to earn $0.08 per share for the quarter. Analysts’ estimates typically exclude special items.
Total revenues for the quarter grew 10.1% to $48.02 million from $43.61 million in the prior-year quarter. Four analysts had a consensus revenue estimate of $48.49 million for the quarter.
by RTT Staff Writer
For comments and feedback: contact editorial@rttnews.com
Appendix 2: Variable Definitions
Rights and permissions
About this article
Cite this article
Blankespoor, E., deHaan, E. & Zhu, C. Capital market effects of media synthesis and dissemination: evidence from robo-journalism. Rev Account Stud 23, 1–36 (2018). https://doi.org/10.1007/s11142-017-9422-2
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11142-017-9422-2