Analysis of year-over-year changes in Risk Factors Disclosure in 10-K filings

Risk Factor Disclosures -- Item 1A -- in 10-K forms filed with SEC is one of the important sections since it contains a company's yearly risk updates, and thus helps investors decide whether to invest in a company or not. It is crucial to read this section carefully in order to make better investment choices. Given the large number of such forms filed on a yearly basis, it is very cumbersome for humans to understand and analyze them to make informed decisions. We discuss the task of bank failure classification using textual analysis on item 1A for various banks' 10-K forms, i.e., to predict whether a bank will fail or not. We also analyze other quantitative bank performance indicators like leverage and Return On Assets (ROA), and see how well text-based methods can predict those risk indicators. In particular, to create our textual corpora, we focus on the changes in the 1A sections, retaining only those sentences that have under 30% and 40% similarity over two consecutive years (for the same bank). We implement deep learning and other supervised learning techniques like Convolutional Neural Networks (CNN), Support Vector Machines (SVM) and Linear Regression. We also combine the word sentiment polarities along with their count as our weighted feature vector.