Predicting Future Earnings Change Using Numeric and Textual Information in Financial Reports

摘要

The main propose of this study is to build a more powerful earning prediction model by incorporating risk information disclosed in the textual portion of financial reports. We adopt the single-index model developed by Weiss, Naik and Tsai as a foundation. However, other than the traditionally used numeric financial information, our model adds textual information about risk sentiment contained in financial reports. We believe such a model can reduce specification errors resulting from pre-assuming linear relationship, thus can predict future earnings more accurately. The empirical results show that the modified model does significantly improve the accuracy of earning prediction.