We study the aggregate consequences of dynamic lending relationships in a model of heterogeneous banks facing financial frictions. We estimate the model’s loan demand system on administrative loan-level data: the market power implied by the estimated strength and persistence of relationships yields a long run reduction in credit of 5.9%. Relationships amplify the negative real effects of credit supply shocks, but mute those of negative credit demand shocks. In a financial crisis which destroys 25% of bank net worth, for example, loan volume drops more than twice as much in our baseline model than in a competitive analog with no relationships, but banks recapitalize faster.

We develop a simple model of concentrated lending where lenders have incentives for evergreening loans by offering better terms to firms that are close to default. We detect such lending behavior using loan-level supervisory data for the United States. Banks that own a larger share of a firm's debt provide distressed firms with relatively more credit at lower interest rates. Building on this empirical validation, we incorporate the theoretical mechanism into a dynamic heterogeneous-firm model to show that evergreening affects aggregate outcomes, resulting in lower interest rates, higher levels of debt, and lower productivity.