The Evolution of Risk Factor Disclosures through a Structural Topic Model

Using an unsupervised structural topic model, we examine the topical evolution of risk factorspublished in 72,000 US companies’ 10-K reports over a period ranging from 2006 to 2021. Wethereby update the existing literature with a contemporary examination of the risk reportinglandscape. Accounting for temporal and industrial effects, our model allows us to identify themost common risk types published within an industry, pinpoint the differences in prevalenceand specificity across industries and measure their evolution over time. Identifying 24 types ofrisks, our research reveals that cybersecurity (IT), corporate growth strategies, R&D, tax,environmental, regulatory and stock market related risks have significantly gained inprevalence at the expense of more traditional competition, sales, human capital and supplychain risks. We reject the premise that firms are increasingly disclosing similar risk types aswe provide statistical evidence that the industry in which a firm operates significantlyinfluences the prevalence and disclosure of risk factor topics over time. Finally, we show thathalf of the risk factor topics significantly gain in industry-specificity over our sample period.These findings suggest that firms have to some extent been listening to the SEC’s call to reportmore material risk factors that reflect the current risk climate and the firms’ activities. Ourpaper contributes to the emerging literature on the specific content and economic determinantsof risk factor disclosures as well as the growing body on textual analysis methods through theuse of a novel unsupervised machine learning model