The Sentiment Feature of 10-K MD&As and the Financial Misstatement Prediction: A Comparison of Deep Learning and "Bag of Words" Approaches

• In Event: SET Learning

Abstract

The objective of this study is to provide insight into two research questions: (1) Does the sentiment feature of 10-K MD&As extracted by deep learning approach provide essential information for financial misstatement prediction? (2) How effective the deep learning approach performs as compared to “bag of words” approach in terms of prediction accuracy? We analyzed 30,239 MD&As of 10-K filings for fiscal years from 2006 to 2015 using the two approaches separately and obtained two sets of sentiment scores, Sentiment_TM and Sentiment_DL. Utilizing CHAID (CHI-square Adjusted Interaction Detection) algorithm, we established two classification models and compared their predictive performance. Additional 35 misstatement predictors examined by previous literature were added to the classification models. The two models have the identical structure with the exception of the sentiment measure. The results show that both model 1 and model 2 perform better than previous prediction models for the financial misstatement. The sentiment feature extracted by Deep Learning approach performs no worse than that obtained by “bag of words” approach. We conclude that Deep Learning based textual analysis is a promising technique for audit analytics.

Authors

• Ting Sun, Rutgers University-Newark

• Yue Liu, Rutgers University - Newark

• Miklos A Vasarhelyi, Rutgers University - Newark