The Contracts Not Written: Measuring Covenant Tightness with LLM-Generated Synthetic Counterfactuals
Abstract
Syndicated loan covenants exhibit substantial heterogeneity even among observably similar borrowers. We ask whether this variation reflects economically meaningful differences in contracting outcomes shaped by bargaining power or is largely incidental once observable risk is accounted for. Using a new measure of covenant tightness, defined as a contract's position within a deal-specific distribution of feasible covenant packages generated by large language models and validated against independent programmatic methods, we document a sharp asymmetry across borrower credit quality. Among investment-grade borrowers, tighter covenants are associated with significantly higher loan spreads but do not predict subsequent violations. Among speculative-grade borrowers, tightness has no incremental pricing power but strongly predicts violations. This cross-outcome asymmetry is difficult to reconcile with a residual risk interpretation and instead reflects two distinct contracting regimes: negotiated outcomes driven by bargaining power for investment-grade borrowers, and lender risk management and proximity to binding constraints for speculative-grade borrowers. More broadly, we show how counterfactual contract distributions can be used to measure contract design in settings with substantial contractual heterogeneity.
Cite
@unpublished{rettl0the,
title = {The Contracts Not Written: Measuring Covenant Tightness with LLM-Generated Synthetic Counterfactuals},
author = {Daniel A. Rettl and Malcolm Wardlaw}
}
Replication
https://github.com/MalcolmWardlaw/covenant-tightness-counterfactuals