Bank risk aggregation with forward-looking textual risk disclosures

Abstract Approaches based on financial statements are important to the field of bank risk aggregation. However, previous studies only used numerical data recorded in financial statements to aggregate bank risk. Time lags in numerical data can bias risk aggregation results. Thus, this paper first incorporates forward-looking textual risk disclosures reported in financial statements into bank risk aggregation. In doing so, we overcome the drawback of risk aggregation resulting from using only historical numerical data, and in turn, we obtain more reasonable aggregate risk results. In our experiment, based on 812 pieces of numerical risk data for each type of credit, market and operational risk and based on 36178 summary headings of textual risk disclosures drawn from 1224 Form 10-K filings of 153 U.S. commercial banks for 2010-2017, we aggregate credit, market and operational risks for U.S. commercial banks. In comparing total risks with and without forward-looking textual risk disclosures, our empirical results show that disregarding forward-looking textual risk disclosures overestimates the total risk of 2010 to 2013 while underestimating the total risk of 2014 to 2017.