This paper provides evidence on the relationship between differential treatment of minority borrowers and their mortgage market outcomes. Using data from a field experiment that identifies differential treatment matched to real borrower transactions in the Home Mortgage Disclosure Act (HMDA) data, we estimate difference-in-difference models between African American and white borrowers across lending institutions that display varying degrees of differential treatment. Our results show that African Americans are more likely to be in a high-cost (subprime) loan when borrowing from lenders that are more responsive to them in the field experiment. We also show that net measures of differential treatment are not related to the probability of African American borrowers having a high-cost loan. Our results suggest that differential outcomes are related to within-institution factors, not just across-institution factors like institutional access, as previous studies find.
We simulate changes to metropolitan area home prices from reforming the Mortgage Interest Deduction (MID). Price simulations are based on an extended user cost model that incorporates two dimensions of behavioral change in home buyers: sensitivity of borrowing and the propensity to use tax deductions. We simulate prices with both inelastic and elastic supply. Our results show a wide range of price effects across metropolitan areas and prospective policies. Considering behavioral change and no supply elasticity, eliminating the MID results in average home price declines as steep as 13.5 percent in Washington, D.C., and as small as 3.5 percent in Miami-Fort Lauderdale, Florida. Converting the MID to a 15 percent refundable credit reduces prices by as much as 1.4 percent in San Jose, California, San Francisco, California, and Washington, D.C., and increases average prices in other metropolitan areas by as much as 12.1 percent (Miami-Fort Lauderdale). Accounting for market elasticities produces price estimates that are on average thirty-six percent as large as standard estimates.