Earnings releases, and the “seasons” made up out of them, have become almost a comforting ritual in the investment world. The days leading up to a company’s release often incites a good deal of more or less well-informed guesswork.
That guesswork may become significantly more informed through crowdsourcing. That is the point of a recent article.
Part of the ritual is the discretion that companies possess as to how they make their presentation. One might almost say that’s part of the fun. On February 2, 2017, Ford Motor Company reported an $800 million loss for its fourth quarter 2016. Understandably perhaps, it did so with a presentation that managed to de-emphasize the fact that this was a fourth quarter release, with a list of highlights and a commentary that each discussed the year 2016 as a whole.
But the figures for Ford’s 4Q performance were all there somewhere, and it seems unlikely that sophisticated investor were fooled, or even slowed in making what they deemed the appropriate trades or position adjustments.
For sophisticated investors, the questions that surround these ritualistic announcements are empirically subtle ones. For example, can Big Data, drawing on the financial social media databases, allow for inferences as an announcement approaches that are better than those of other preannouncement variables (such as track record, earnings volatility, and coverage)?
There has long been a suspicion in many scholarly quarters that all the tweets and posts of people ranting about how malign forces are tanking their portfolio, or on better days of people boasting about their own brilliant powers of precognition – that all of this is so much noise, best ignored. But might there not be a signal amidst the noise?
Four Conclusions on Crowd Sourcing
That question is the subject of the new article on crowdsourcing by Jim Kyung-Soo Liew, assistant professor, finance, Johns Hopkins, Carey Business School, co-authored by two Johns Hopkins masters students at the Department of Applied Mathematics and Statistics, Shenghan Guo and Tongli Zhang.
These scholars drew on a pair of prominent financial social media databases, Estimize and iSentium. Lieu et al reach four conclusions:
- Crowdsourced consensus earnings were slightly better than Wall Street’s consensus earnings through the years 2013 and 2014;
- They are slightly better because Wall Street as an industry engages in low-balling as a predictable systemic bias. Thus, while the crowd’s estimates understate actual reported earnings by 52-54%, the Wall Street consensus understates them by 65-68%;
- Tweet sentiment does contain significant information not available via traditional preannouncement variables, whether “significance” here is understood economically or statistically; and finally
- This tweet sentiment predicts post-announcement risk-adjusted excess return over a period of a few days, with the statistical significance soon waning as the market incorporates the previously overlooked information.
This is not Liew’s first article on the subject. A year ago, working then with Tamás Budavári — also of the Applied Mathematics Department at Johns Hopkins – Liew studied the time series variations of daily stock returns in connection with the direct tweet sentiments provided by StockTwits.
That article reached conclusions broadly compatible with the more recent one. Moreover, it contended that social media reaction to a stock might well deserve a place in the standard models of stock performance, alongside the Fama-French list of five factors: market risk, company size, value-versus-growth, the robustness of profitability, and the balance of aggressive-versus-conservative capital investment by the issuing company.
“If social media platforms give us access to a pool of largely independently generated sentiments, then exposing the component of idiosyncratic risk that comes due to participants’ aggregated sentiment could make markets more efficient,” concluded Liew and Budavári a year ago. “Since sentiment information appears to be transmitted and incorporated quickly and efficiently, it becomes vital to our understanding of security price behavior.”