Study suggests investors should skip Wall Street analysis, mine Twitter for trades

Investors might want to put aside all that Wall Street research and instead turn their eyes to Twitter.

A new study by academics at Johns Hopkins University finds that crowdsourced company earnings estimates and sentiment data generated by tweets may be both more accurate than Wall Street's offerings and help generate trading profits.

The study examined crowdsourced earnings estimates gathered by analysis firm Estimize, which allows users to make estimates and then crunches the data, as well as tweet company sentiment data produced by iSentium, a firm specializing in such work.

"It appears that tweet sentiment has power to predict post-earnings risk-adjusted excess returns," Jim Kyung-Soo Liew, Shenghan Guo and Tongli Zhang of Johns Hopkins wrote in the paper.

"There appears to be some indication that a well-formulated strategy incorporating both data sets could be an interesting avenue of future research and may lead to annualized excess gross returns in the 10 to 20 percent range."

This conclusion comes, inevitably, with many caveats, but first let's look at the findings.

The study, based on earnings announcements from November 2011 to December 2014, used iSentium linguistic analysis data to create a positive or negative tweet sentiment value for companies in the run-up to their earnings release.

For example, shorting stocks with negative tweet sentiment that record a negative earnings surprise generates a bit more than half a percent of risk-adjusted excess returns within five trading days.

That equates to an annual excess return of about 26 percent, which is huge.

Risk-adjusted excess returns are what a stock generates compared to a portfolio of similar stocks.

Other strategies, such as going "long" on stocks with positive sentiment and positive earnings surprises also generate excess returns, though on lower scales.

To be sure, these results are gross of fees. As well, some of these stocks may not be large capitalization, meaning that they could be hard to borrow for a "short" sale or that a trade could not be done in very large size.

The Estimize earnings estimates are purely crowdsourced, but appear to be better than those Wall Street generates at great expense.

"Over the whole period of our data we find that Estimize's consensus is 56.4 percent more accurate than Wall Street's consensus — similar to earlier findings," the authors write.

Fewer estimates in the Estimize data "lowball" earnings and more overestimate, something which few Wall Street firms do. This makes the crowd sourced estimate more accurate.

That the crowd can do a better job estimating earnings is not surprising. The financial services industry has a well-documented history of underestimating earnings going into announcements. This may be so that companies can "beat" expectations and please their constituencies.

Some of this is about agency problems, the misaligned incentives of the financial services industry. In other words, equity analysts are likely not telling you what they really think because they have other interests to advance and protect.

"Not being overly influenced by the need to gain access and favor from company executives appears to be a reasonable explanation supported by the data," the authors write.

"Additionally, building relationships over time with conference calls, equity research events and social gatherings may have led many Wall Street analysts to lose their objectivity when forecasting company's earnings."

It remains to be seen if the results of the study can be replicated and the extent to which the strategies it suggests will work in real life. That said, it is a great example of a larger trend: Twitter, and other means of open-sourced information, are going to drive down the margins of middlemen in and outside of the financial industry.