Breaking the Word Bank: Measurement and Effects of Bank Level Uncertainty

Abstract

Banks differ from non-financial firms as banks must communicate to both regulators and shareholders. Also, unlike non-financial firms, banks possess opaque and complex balance sheets and are the main providers of credit to the real economy. In this paper, I propose a new index to detect the idiosyncratic uncertainty banks face at the bank-quarter level by applying natural language processing techniques to earnings conference call transcripts. The index reveals which banks at a given quarter signal more uncertainty about their balance sheets. Higher uncertainty is associated with lower lending and higher liquidity the next quarter, suggesting active management of uncertainty. The active management of uncertainty is more pronounced during periods of high aggregate volatility and for banks with more skin in the game. Using loan level data and firm fixed effects, I control for demand-side factors and find that higher bank level uncertainty is associated with lower loan issuances the following quarter.

Keywords: Uncertainty, Banking, Credit, Natural Language Processing, Machine Learning

JEL Classification: G21, G3, E5, D8