Reject inference in consumer credit scoring with nonignorable missing data

► New approach to deal with missing data in a model of consumer credit scoring. ► Parameter estimates change significantly if we take refused credit into account. ► Forecasts of defaults derived from the new method improve upon standard ones.