Are day traders bias free?—evidence from internet stock message boards

Abstract

Internet Stock Message Boards, Retail Investor Sentiment, Sentiment Bias, Day Traders, Text Classifiers, G11, G12,

Stock Market Prediction without Sentiment Analysis: Using a Web-Traffic Based Classifier and User-Level Analysis

“…There is evidence over a strong positively-biased sentiment populating financial on-line communities (Zhang et al [8]), confirmed by our survey as well. This allows us to presume that traffic could be a proxy for at least positive sentiment.…”

“…Other features are post length, rankas provided by the blog editor software, number of posts, comments, size of loyal and outliers. These features are then correlated to the market dynamics training a support vector machine with the following results: 78% accuracy in predicting the magnitude of the movement and 87% in the accuracy of the movement after one week (weekly) Similar works in the area are the ones by Agarwall et al [6] on the general problem of identifying influential bloggers in a community and the work by U. Zhang [8], that studied the correlation between past-performance of an user and its reputation. The authors provides insight on what constitutes a reputable and respected user, and concludes how reputation derive from a more complex synthesis of various behavioral factors besides its textual contributions, implicitly confirming the validity of nontextual features.…”

Internet big data and capital markets: a literature review

“…Simultaneously, most literature has evaluated the information contents of forum posts using textual analysis to construct the content class indicators. The main types include indexes based on numbers of opposers and supporters and the degree of divergence in opinions (Antweiler and Frank 2004;Das et al 2005;Nan 2015;Sabherwal et al 2011;Zhang and Swanson 2010). In a study on information exchange over the Internet and herding behavior in the stock market, Zheng et al (2015) selected the natural logarithm for the total number of daily posts as the proxy variable for investors' communication.…”

“…Zhang and Swanson (2010) collected data from posts on message boards in Yahoo! Finance that corresponded to 30 U.S. listed companies.…”

“…Through simple comparison, Antweiler and Frank (2004) found that the accuracy of algorithm classification is actually higher than that of manual classification. Other than the Naive Bayes algorithm, some scholars have also used the maximum entropy model for text classification (Zhang and Swanson 2010).…”