Predicting future financial performance of banks from management’s tone in the textual disclosures

Predicting bank performance is important for investors and regulatory authorities. Previous research on non-financial firms has shown that augmenting the numeric information in the financial statements with textual information from the rest of the annual reports led to a more accurate prediction of performance. The researchers have generally eschewed the use of narrative in textual disclosure for building better bank prediction models. Moreover, very few of these studies dealt with endogeneity problem that is pervasive in corporate settings. This study employed natural language processing (NLP) for extracting tone of disclosures and used it along with other financial data in the predictive models. The models were estimated using GMM to deal with the endogeneity. Our results suggested that in addition to quantitative financial information, textual data were a valuable source of additional information for predicting the future performance of banks. Moreover, dealing with endogeneity was found necessary for obtaining better predictions by leveraging textual information.