Debt structure instability using machine learning

Abstract

Applying a machine-learning algorithm to a large sample of U.S. public firms, we document that more than 30% of the firms substantially alter debt structures in a year, even when leverage ratio is stable, when short-term debt is trivial, and when little cash outlay is required for operations. The instability of debt structure reveals new costs of financial constraints: compared to high-credit-quality firms, low-credit-quality firms have to change debt structure more frequently to accommodate their financing needs, even with increased borrowing costs; low-credit-quality firms lack the opportunity available to high-credit-quality firms to reduce borrowing costs through switching debt instruments.