Understanding the value of financial advice on social media

Abstract

This thesis includes three separate studies which examine different aspects of the value of financial advice on social media. In the first study, we examine the effect of the long or short positions of non-professional analysts (hereafter, NPAs) writing on the social media outlet Seeking Alpha (hereafter SA) on the direction of investor trading and subsequent stock returns. We find that NPA positions contribute directly to short-window (less than one week) order imbalances after the article’s publication. We also find that purchasing stocks with the most favorable sentiment and short selling those with the least favorable sentiment, together with daily portfolio rebalancing, yield a positive abnormal gross return that is statistically, but not economically, significant. There is no evidence that the information on Seeking Alpha can be used to generate economically significant abnormal returns. In the second study, we study the effect of providing financial incentives to NPAs on SA on the quality of stock recommendations. We find that NPAs are more likely to join the premium partnership program on SA and receive monetary payments if they have joined SA for a longer time and contributed more articles. We show that financial incentives reduce the quality of free stock recommendations. NPAs react to financial incentives and put their best work out where it generates the most income. The quality of NPAs’ long (short) stock position recommendations in fee-based articles after joining the premium partnership program is worse (better) than the quality of their long (short) stock position recommendations in free articles prior to joining the premium partnership program. This study contributes to the literature on the role of social media in financial markets, the role of sell-side analysts in financial markets, and the understanding of the role of financial incentives in influencing the quality of user-generated content provided by NPAs on social media. In the third study, we examine the performance of textual analysis methods on data collected from financial microblogging websites HotCopper (hereafter, HC) and StockTwits (hereafter, ST), which have been frequently used for sentiment analysis and stock market return predictions. We show that machine-learning classifiers have better accuracy than the Loughran & McDonald (2011) dictionary when classifying short text from HC. When conducting sentiment analysis on short text from social media and examining the effect of social media sentiment on stock market abnormal returns, researchers should try to use a financial social media like ST rather than a more informal social media like Twitter.