The informativeness of risk factor disclosures: estimating the covariance matrix of stock returns using similarity measures

While risk factor disclosures in 10-K filings have been criticized by practitioners as generic and boilerplate, recent studies indicate that these risk reports can be informative. This study contributes to the ongoing discussion by investigating whether risk factor disclosures contain valuable information that can be used to improve the estimation of the covariance matrix of stock returns. In particular, we examine the 10-K and 10-Q filings of firms listed in the Standard & Poor's 100 index from 2006 to 2020. We compute cosine similarity measures to compare risk factor reports and use them in linear regressions to estimate the covariance matrix of stock returns. Our estimators using risk report data outperform well-established sample-based estimators, such as the shrinkage estimator of Ledoit and Wolf. This indicates that risk factor disclosures are informative and contain information that is not already reflected in historical stock prices. This information can be used to improve portfolio selection and thus generate economic value.