This work applies natural language processing (NLP) methods to analyze Sect. 7 (Management Discussion and Analysis) of the 10-K filings for a large set of U.S. financial companies in the period 2005–2022 in order to build an index of financial institution confidence. We perform sentiment analysis using the FinBERT pre-trained financial language representation model for financial text mining, and aggregate the yearly sentiment values of each financial institution to build our index. We show a positive correlation between our index and the main events that characterize the US economic cycle.