Exploring the Information Contents of Risk Factors in SEC Form 10-K: A Multi-Label Text Classification Application

Abstract

This study develops, implements, and evaluates a multi-label text classification algorithm that extracts textual information from the annual reports of all publicly listed USA companies. Specifically, the proposed system can automatically identify 25 types of frequently mentioned risk factors in a section called “Item 1A. Risk Factors” in SEC Form 10-K. The true positive rate on the training set with 3,153 risk factors is 80.65 percent while the false positive rate is 12.67 percent. This system is applied to extract risk factors in 10-Ks of most USA companies from 2006 to 2010. By the first-differencing panel data regression, this study shows that the extracted risk factors, the associated risk factor orderings, and the number of risk factors indeed provide additional explanation power for the target firm’s risk measures, annual stock returns, and key financial ratios.

Keywords: Text Classification, Text Mining, Multi-Label Classification, Risk Factors, Annual Reports, Financial Statement Analysis

JEL Classification: D82, G18, M41, M45, G14