Our sentiment analysis of stock microblogs shows that users tend to be much more bullish than bearish. We manually classified 2,500 tweets as either buy, hold, or sell signals. Roughly half of these messages were considered to be hold/neutral signals (49.6%). Among the remainder, buy signals were more than twice as likely (35.2%) as sell signals (15.2%). This indicates that stock microblogs appear to be more balanced in terms of bullishness than internet message boards where the ratio of buy vs. sell signals ranges from 7:1 (Dewally, 2003) to 5:1 (Antweiler & Frank, 2004).
The table below shows a few typical examples. Our analysis of the most common words per class draws a semantic profile of buy, hold and sell signals. Obviously, some features occur frequently in all classes (e.g., numbers and hyperlinks). However, beyond these universal features, the most common words reasonably reflect the linguistic bullishness of the three classes. Positive emotions, for example, are much more likely among buy signals. In addition, buy signals often contain bullish words with an origin in technical analysis (e.g., “moving average”, “resistance”, “up”, or “high”), operations (e.g., “acquire”), financials (e.g., “beat”, “earn”), or trading (e.g., “buy”, “long”, “call”). Sell signals contain many corresponding bearish words in the areas of technical analysis (e.g., “support” and “cross”), financials (e.g., “loss”) or trading (e.g., “short” and “put”). As a results of the frequent occurrence of negative adjectives (e.g., “weak”, “low”) and verbs (e.g., “decline”, “fall”), negative emotions are among the most common features in sell signals. Positive and negative emotions are much more equally balanced in hold messages, which also contain more neutral words such as product names (e.g., “ipad”, “iphone”) and make fewer references to specific price targets (i.e., dollar values).
