Crowds on Wall Street: Extracting Value from Social Investing Platforms

For decades, the world of financial advisors has been dom- inated by large investment banks such as Goldman Sachs. In recent years, user-contributed investment services suc h as SeekingAlpha and StockTwits have grown to millions of users. In this paper, we seek to understand the quality and impact of content on social investment platforms, by empiri- cally analyzing complete datasets of SeekingAlpha articles (9 years) and StockTwits messages (4 years). We develop sen- timent analysis tools and correlate contributed content to the historical performance of relevant stocks. While SeekingAl- pha articles and StockTwits messages provide minimal corre- lation to stock performance in aggregate, a subset of authors contribute more valuable (predictive) content. We show that these authors can be identified via both empirical methods or by user interactions, and investments using their analysis sig- nificantly outperform broader markets. Finally, we conduct a user survey that sheds light on users views of SeekingAlpha content and stock manipulation. In this paper, we seek to understand the quality and impact of opinions and analysis shared on social investment platforms. We target the two primary yet quite different social invest- ment platforms, SeekingAlpha and StockTwits, and analyze the potential for investment returns following their recom - mendations versus the market baseline, the S&P 500 stock market index. We seek to understand how expertise of con- tributors can affect the quality and utility of contributed con- tent, using SeekingAlpha as an "expert" model (all content is contributed by less than 0.27% of users) and StockTwits as a "peer" model (any user can contribute). Our work makes four key contributions. First, we gather longitudinal datasets from both platforms since their inception (9 years of data for SeekingAlpha, 4 years for StockTwits). We develop sentiment analyzers on each dataset, using a mixture of keyword processing and machine learning classifiers. Validation shows our methods achieve high accuracy in extracting sentiments towards in- dividual stocks (85.5% for SeekingAlpha, 76.2% for Stock- Twits). Second, we analyze correlation between content sentiment from both services with stock returns at different time scal es. We show that content from both SeekingAlpha and Stock- Twits provide minimal forward correlation with stock perfor- mance. While the average article provides little value, we find that a subset of "top authors" in SeekingAlpha contribut e content that shows significantly higher correlation with fu ture stock performance. Third, we evaluate the hypothetical performance of simple investment strategies following top authors from both plat- forms. We show that investment strategies based on stock sentiment from top SeekingAlpha authors perform exception- ally well and significantly outperform broader markets. In contrast, strategies relying on StockTwits generally unde rper- form relative to broader markets. In addition, we show that we can identify top authors without historical stock market data, using only user interactions with their articles as a g uide. Fourth, we conduct a large scale survey of SeekingAlpha users and contributors to understand their usage, reliance , and trust in the SeekingAlpha service. Results show that despite seeing potentially intentionally misleading or manipulat ive articles, most users still rely heavily on the site content f or investment advice. Most consider SeekingAlpha unique, and