Simultaneity bias in mortgage lending: A test of simultaneous equations models on bank-specific data

This study uses simultaneous equations models and single-equation models to test for simultaneity bias in mortgage refinance data compiled by a regional bank. The purpose of the study is to assess the claim that single-equation models of the lending decision produce biased and inconsistent parameter estimates of endogenous mortgage terms. Bank-specific data are analyzed to avoid bias resulting from uncontrolled policy, training, or underwriting differences across banks. Importantly, the data contain all variables the regional bank identified as important factors in explaining its loan disposition results. After controlling for applicants' debt, income, credit history, and requested loan term, I find that the race coefficient in single-equation models is biased upward, while the loan-to-value ratio coefficient is biased downward, although both biases are insignificant. Overall, the results suggest that simultaneous equations models are preferable to single-equation models in tests for discrimination, and can be used to determine the extent of race coefficient and loan-to-value ratio coefficient bias in single-equation models.