Mortgage lending, sample selection and default

Traditional models of mortgage default suffer from sample-selection bias because they do not control for the loan approval process. This paper estimates a sample-selection-corrected default model using the 1990 Boston Federal Reserve loan application sample and the 1992 Federal Housing Authority (FHA) foreclosure sample. A single-equation FHA default model appears to suffer from substantial selection bias, but the bias primarily arises from the omission of credit history and other variables that are only in the application sample. Therefore, default models that contain detailed information on applicants may not suffer from substantial selection bias. Finally, a test for prejudice-based discrimination is developed and conducted, but the findings are inconclusive.