Yahoo! for Amazon: Sentiment Extraction from Small Talk on the Web

We develop a methodology for extracting small investor sentiment from stock message boards. Five distinct classifier algorithms coupled by a voting scheme are found to perform well against human and statistical benchmarks. Time series and cross-sectional aggregation of message information improves the quality of the resultant sentiment index. Empirical applications evidence a relationship with stock returns – visually, using phase-lag analysis, pattern recognition and statistical methods. Sentiment has an idiosyncratic component, and aggregation of sentiment across stocks tracks index returns more strongly than with individual stocks. Preliminary evidence suggests that market activity influences small investor sentiment. Thus, the algorithms developed in this paper may be used to assess the impact on investor opinion of management announcements, press releases, third-party news, and regulatory changes. ∗We owe a special debt to the creative environments at UC Berkeley’s Computer Science Division and Haas School, where this work was begun. Thanks to David Levine for many comments and for the title. We are grateful to Vikas Agarwal, Chris Brooks, Yuk-Shee Chan, David Gibson, Geoffrey Friesen, David Leinweber, Asis Martinez-Jerez, Priya Raghubir, Sridhar Rajagopalan, Ajit Ranade, Mark Rubinstein, Peter Tufano, Raman Uppal, Shiv Vaithyanathan, Robert Wilensky and seminar participants at Northwestern University, UC Berkeley-EECS, London Business School, University of Wisconsin, Madison, the Multinational Finance Conference, Italy, the Asia Pacific Finance Association Meetings, Bangkok, and the European Finance Association Meetings, Barcelona, for helpful discussions and insights. Danny Tom and Jason Waddle were instrumental in delivering insights into this paper through joint work on alternative techniques via support vector machines. The first author gratefully acknowledges support from the Price Waterhouse Cooper’s Risk Institute, the Dean Witter Foundation, and a Research Grant from Santa Clara University. Please address all correspondence to Professor Sanjiv Das, Breetwor Fellow & Associate Professor, Santa Clara University, Leavey School of Business, Dept of Finance, 208 Kenna Hall, Santa Clara, CA 95053-0388. Email: [email protected]. Mike Chen is in the Computer Science Division, UC Berkeley, [email protected].