18
National Survey of Mortgage Originations
Critchfield, Dey, Mota, and Patrabansh
The smaller share of completely rural borrowers with scores of 740 or greater suggests that loans
in completely rural areas were likely to have higher interest rates because the best rates are typically
offered to borrowers with scores above 740. To test this conjecture, we examine whether there
were statistically significant differences in Freddie Mac’s Primary Mortgage Market Survey (PMMS
®
)
spread across the analysis groups (metro, non-metro, and completely rural).
19
We consider both the
observed difference in PMMS spreads as well as an adjusted difference based on a regression model
that accounts for borrowers, property, and loan attributes:
PMMS_Spread
i,c
= α +
ρ
CR
CR
c
+
ρ
NM
NM
c
+ βX
i
+ ε
i,c
The equation models the PMMS spread of the loan originated to borrower i in analysis group c
(PMMS_Spread
i,c
) on indicator variables of whether analysis group c is completely rural
(CR
c
) or non-metro (NM
c
), as well as a vector of borrower, property, and loan attributes (X
i
).
The coefficient estimates for
ρ
CR
and
ρ
NM
indicate the average difference in the PMMS
spread for loans originated in completely rural and non-metro counties relative to metro counties.
Exhibit 4 displays the coefficient estimates and indicators of statistical significance for the sample of
all mortgages, and then separately for purchase and refinance mortgages. This exhibit also provides
the borrower, property, and loan attributes that are used as control variables for estimating the
differences between counties.
20
Borrowers in completely rural counties indeed paid a slightly higher
interest rate than borrowers in metro areas,
21
and the spread over the PMMS rate for borrowers in
completely rural counties was 14 basis points higher than metro areas with a statistically significant
difference. The results imply that completely rural borrowers paid 16 basis points more than
non-metro borrowers. While 14 basis points appears small for the average of completely rural
counties, a mortgage of $100,000 with a 4.00 percent rate that moved to 4.14 percent would cause
the monthly payment to rise $8.10, or $97.20 per year, and cost the borrower $2,916.00 over the
life of the loan. For purchase mortgages, the differences between completely rural and metro areas
was 24 basis points and statistically significant, but the refinance mortgages difference was not
significant. Notably, in this regression framework, the PMMS spread does not vary significantly
with most borrower characteristics, whereas the purchase flag, loan amount, property type, and
credit score are statistically significant predictors of the interest rate spread.
Geographic Differences in Borrowers’ Experiences
and Knowledge
In examining differences in mortgage borrowers’ experience and knowledge by geography, we
use the framework of the PMMS equation and exhibit 4 to examine borrowers’ self-reported
satisfaction, knowledge, and lender selection. More specifically, exhibit 5 shows the results of the
NSMO asking borrowers how satisfied they were with several aspects of their mortgage and the
19
Freddie Mac publishes the average PMMS rate by mortgage term on a weekly basis. The PMMS spread is calculated as
the dierence between the actual note rate of a mortgage and Freddie Mac’s PMMS average prime oer note rate for that
term at that time. This spread indicates how expensive a mortgage is compared to the average mortgage of similar term
taken out in that week. The spread used in this article is unbounded while the spread in the NSMO public use file is
bounded for privacy considerations.
20
Some of the controls, such as race, ethnicity, age and gender, are characteristics that lenders do not or cannot use in
loan pricing models. We include them to account indirectly for unmeasured characteristics that may be correlated with
these controls.
21
The R-squared for the models ranged from 0.06070 to 0.08412.