Textual classification of SEC comment letters
Subject
Accounting
Publishing details
Social Sciences Research Network
Publication Year
2018
Abstract
The purpose of this study is to identify important SEC comment letters and examine mechanisms by which they affect firm value. A signal of importance based on abnormal returns at the time of disclosure is effective at predicting lower future operating performance and more frequent future write-downs. A signal based on EDGAR downloads of the comment letters predicts lower future earnings, a higher rate of future restatements, and a higher rate of future write-downs. I create Naive Bayes text classification signals trained to predict future restatements, write-downs, and internal control weaknesses. The resulting text-based signals are effective at predicting future restatement and write-down announcements, and have the added benefit of not needing a market response to predict importance. Topic analysis of the comment letters provides insights into the relative volume of comments among different topics, and finds that different topics are related to each of the signals of comment letter importance.
Keywords
SEC comment letters; Text classification; Topic modeling; Financial performance; Restatements; Write-downs; Internal control weaknesses; Investor attention; Naive Bayes; Latent Dirichlet Allocation
Series
Social Sciences Research Network
Available on ECCH
No
