Risk Factors That Matter: Textual Analysis of Risk Disclosures for the Cross-Section of Returns

Abstract

I use machine learning to extract the risk factors identified by firms in their annual disclosures, quantify firms' exposure to each risk, and construct mimicking portfolios that proxy for each risk. These portfolios are priced and contain information beyond what is accounted for by common multi-factor representations. A second-level factor model categorizes risks into systematic and idiosyncratic and helps construct pricing factors. These firm-identified risk factors perform at least as well as classic models when pricing a broad set of assets. These results suggest that firm-side risks, as in production-based models, are valuable in pricing the cross-section.

Keywords: Cross-Section of Returns, Factor Models, Machine Learning, Big Data, LDA, Text Analysis, NLP