LDA quantification of 10-K Risk-Factors and the Information Content of Textual Reporting

Abstract

We use machine-learning to determine the information content of the Item 1A Risk Factors section of S&P 1500 10-Ks. We identify and quantify 30 risk-factors and show a strong positive relation between levels of and contemporaneous changes in risk-factors and proxies for the associated risks. Typically, 28% of cross-firm variation in a risk-proxy is explained by cross-firm variation in the associated risk-factor. Risk disclosure is not found to be forward-looking. Item 1A’s informativeness has not declined through time despite previously documented increases in boilerplate content, stickiness and redundancy. Indices of operating and financing risk help explain asset and equity volatility.

Keywords: risk-factors, textual analysis, 10-K, risk indices

JEL Classification: G10, G38, K22, M41