ABSTRACT
This study investigates the relationship between the stock ratings of professional analysts and company-level sentiment. I find that investor sentiment expressed in tweets and news articles displays a positive relationship with professional analysts’ stock ratings and that the sentiment conveyed through tweets has a stronger effect than the sentiment from conventional news articles. Furthermore, the effect of sentiment derived from tweets and news articles on analysts’ ratings is stronger when the sentiment valence is the same (negative or positive) and weaker when the sentiment valence differs. The findings shed light on the link between investor sentiment and analysts’ stock ratings by demonstrating that analysts’ stock ratings are influenced by investor sentiment.
Disclosure statement
The author declares he has no conflict of interest.
Acknowledgment
I am very grateful to the editor and two anonymous reviewers for their substantial contributions to the improvement of this article.
Notes
1 From the first quarter of 2010 through the first quarter of 2019, Twitter reported an 11-fold increase in monthly active users (from 30 million to 330 million; Twitter, Citation2019a), generating over 500 million daily tweets (Twitter, Citation2019b).
2 The panel dataset was unbalanced as some firms did not generate news or Twitter media content every quarter. As noted by Wooldrige (2010), pooled OLS may be employed when selecting a different sample for each period of the panel data.
3 Following Newey and West (Citation1994), where the optimal lags L = floor (4*(N/100)2/9) and N is the firm-level sample size, and given the maximum number of quarterly observations per firm is 20, two lags were used.
4 To mitigate the effect of extreme outliers, the diluted EPS, quarterly return, book-to-market ratio, the change in EPS, and 90-day share price volatility were winsorized at the 1% and 99% level, and the debt-to-capital ratio was winsorized at the 99% level.
5 Bloomberg estimates current analyst ratings on a scale from 1–5, with 1 as the lowest ranking (sell) and 5 as the highest ranking (buy). The first quartile, mean, median, and third quartile were 3.4, 3.9, 4.0, and 4.4, respectively. In response to the positively skewed distribution, three rating groups were used: group 1, analysts with ratings from 1–3; group 2, analysts with ratings from 3–4; and group three, analysts with ratings higher than 4.
6 Using the PE ratio relative to the PE ratio of the firm’s relevant Bloomberg benchmark index, companies above the 3rd quartile (1.575) were coded to 1. Companies with negative or no earnings over the previous year were also coded to 1. All others were coded to 0.
7 The covariance matrix of the regression coefficients was corrected for heteroskedasticity and autocorrelation. All subsequent Wald tests also corrected for heteroskedasticity and autocorrelation.