Relative to the persistent delinquency typically observed in consumer credit, mortgage forbearance during the COVID-19 pandemic was surprisingly short-lived. To understand why, we develop a rich structural model and estimate it on credit panel microdata. Next, using the model and survey data, we identify key roles for both selection into mortgages by relatively patient households and the failure of anticipated income losses to materialize. Even taking selection in account, had losses occurred, forbearance participation would have been persistent. Thus, the power of forbearance, a frequent macroeconomic policy, hinges on borrower selection, shock realizations, and revisions to expectations following those realizations.