Identifying and Following Expert Investors in Stock Microblogs

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1310–1319, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Identifying and Following Expert Investors in Stock Microblogs 1Roy Bar-Haim, 1Elad Dinur, 1,2Ronen Feldman, 1Moshe Fresko and 1Guy Goldstein 1Digital Trowel, Airport City, Israel 2School of Business Administration, The Hebrew University of Jerusalem, Jerusalem, Israel {roy, moshe}@digitaltrowel.com, ronen.feldman@huji.ac.il Abstract Information published in online stock invest- ment message boards, and more recently in stock microblogs, is considered highly valu- able by many investors. Previous work fo- cused on aggregation of sentiment from all users. However, in this work we show that it is beneficial to distinguish expert users from non-experts. We propose a general framework for identifying expert investors, and use it as a basis for several models that predict stock rise from stock microblogging messages (stock tweets). In particular, we present two methods that combine expert identification and per-user unsupervised learning. These methods were shown to achieve relatively high precision in predicting stock rise, and significantly outper- form our baseline. In addition, our work pro- vides an in-depth analysis of the content and potential usefulness of stock tweets. 1 Introduction Online investment message boards such as Yahoo! Finance and Raging Bull allow investors to share trading ideas, advice and opinions on public com- panies. Recently, stock microblogging services such as StockTwits (which started as a filtering service over the Twitter platform) have become very popu- lar. These forums are considered by many investors as highly valuable sources for making their trading decisions. This work aims to mine useful investment in- formation from messages published in stock mi- croblogs. We shall henceforth refer to these mes- sages as stock tweets. Ultimately, we would like to transform those tweets into buy and sell decisions. Given a set of stock-related messages, this process typically comprises two steps: 1. Classify each message as “bullish” (having a positive outlook on the stock), “bearish” (hav- ing a negative outlook on the stock), or neutral. 2. Make trading decisions based on these message classifications. Previous work on stock investment forums and microblogs usually regarded the first step (message classification) as a sentiment analysis problem, and aligned bullish with positive sentiment and bearish with negative sentiment. Messages were classified by matching positive and negative terms from sen- timent lexicons, learning from a hand-labeled set of messages, or some combination of the two (Das and Chen, 2007; Antweiler and Frank, 2004; Chua et al., 2009; Zhang and Skiena, 2010; Sprenger and Welpe, 2010). Trading decisions were made by aggregating the sentiment for a given stock over all the tweets, and picking stocks with strongest sentiment signal (buying the most bullish stocks and short-selling the most bearish ones). Sentiment aggregation reflects the opinion of the investors community as a whole, but overlooks the variability in user expertise. Clearly, not all investors are born equal, and if we could tell experts from non- experts, we would reduce the noise in these forums and obtain high-quality signals …

Content truncated to 500 words