This paper represents the first attempt to estimate an explicit, structural model of credit rationing that simultaneously explains both bank and consumer behavior. Cheating by a small number of individual loan applicants induces banks to grant credit to only a proportion of observationally identical consumers. By comparing the value functions of dynamically optimizing agents, we develop a discrete choice model of the consumers' decision whether or not to apply for credit and of the banks' decision whether to grant the loan. The model is then estimated, using cross sectional data as a nonlinear ordered, sequential logit. Our results shed considerable light on the criteria used by banks in assessing credit applications. They also reveal failings in Hansen-and-Singleton-style time series implementations of the life cycle hypothesis that may explain the latters' poor statistical performance. In particular we find very strong dependence of preferences upon demographic characteristics.
