题目: Informativeness of Text in Analyst Reports: A Naïve Bayes Machine Learning Approach
演讲者：Allen Huang, Assistant Professor, The Hong Kong University of Science and Technology
时间: 2010年11月5日（星期五）3:00—4:30PM
地点: 嘉庚二501
参加者: 对会计研究有兴趣的广大师生
主持人： 沈哲老师
论文简介：
This study is the first to use the naïve Bayes machine learning approach to extract opinions from the text in analyst reports. The automatic approach enables us to process a large sample of 389,096 analyst reports issued for the S&P 500 firms during the period 1995 – 2008. We document that the classification accuracy using the naïve Bayes approach is substantially higher than using General Inquirer and Linguistic Inquiry and Word Count, two dictionary-based content-analysis methods. With the textual opinion measure, we show that the text in analyst reports is not provided merely to support any one of the quantitative forecasts issued contemporaneously, but reflects the favorableness conveyed by all the quantitative signals including levels and revisions of recommendation, earnings forecast and target price. We find significant market reactions to textual opinions, both unconditionally and conditional on all the quantitative forecasts released contemporaneously, consistent with the text in analyst reports providing additional information about firm values. We also find that investors attach twice as much weight to negative textual opinions as to positive textual opinions. In addition, they react even more strongly to negative textual opinions when reports contain sell recommendation or recommendation downgrades, and when the opinions are from forward-looking statements. Lastly, we find that after the enactment of Reg-FD, textual opinions become less informative, suggesting that the information content from the text partially relies on analysts’ private communication with managers.
论文作者简介：
Dr. Huang is assistant professor of accounting in the Hong Kong University of Science & Technology. He got BA from Pekin University and PHD from Duke University.
下载： 论文2010.pdf