Key Conclusions:

Using a newly developed additive semiparametric model, this paper investigates how the implicitly-subsided affordable housing credit in the secondary mortgage market is distributed over lower income homebuyers. GSE and HMDA data for the 20 largest MSAs are used.

1. The partial-linear (PLR) semiparametric model does a better job, compared with linear and quadratic models, in controlling the non-linear effects of borrower’s credit risk factors. The PLR model significantly improves the goodness of fit and it reduce the estimation bias that is found in a linear model.

2. Detailed PLR analysis is conducted for each of the 20 metropolitan areas. The results suggest that neighborhoods with a higher ratio of African-Americans are more likely to be under-represented and a neighborhood's racial component has a greater effect in suburban areas than that in center cities. It is also found that the GSEs purchase disproportionately numbers of lower income loans in relatively affluent neighborhoods. Higher frequency of investor loans and FHA/VA activities also contribute to the spatial mismatch.

3. The paper investigates the non-linearity of the effects of borrower's risk factors on the GSE lower income purchases, using graphic presentations of the semiparametric results. It is the graphical representation of these non-linear components that provides a new and useful tool for analyzing mortgage risks.

• Request a hard copy of the full paper
• Return to index of working papers