Background
While affordability has always been viewed as a functional objective for housing, affordable mortgage financing has increasingly become a desired policy outcome in and of itself, particularly in the wake of the subprime market boom and bust. The primary model of mortgage pricing underlying subprime lending, efficiency pricing, is based on the notion that borrowers who pay more for a mortgage represent a greater risk to the market. In this regard, the “affordability” of a mortgage, or whether or not a borrower receives a high cost (subprime) mortgage is driven completely by economic factors stemming from borrower risk17. However, a growing body of evidence suggests that mortgage markets are not “dichotomized;” that many of the borrowers receiving subprime financing (perhaps as much as 30-50 percent) may have been able to qualify for prime mortgage financing18. In fact, recent research on Community Reinvestment type loan programs (affordable financing alternatives targeting underserved borrowers) may have actually “crowded out” higher cost lending in certain neighborhoods where such alternative lending programs were present. Further, there is evidence that low income borrowers receiving these “affordable” mortgage financing strategies have much lower rates of delinquency and default than similar borrowers receiving high cost financing (matched on credit and income risk characteristics)19.
This analysis tests the extent to which MRB lending volume may have crowded out higher cost lending20 during the peak of the subprime mortgage boom (2006). This analysis suggests that some of borrowers who received MRB subsidized loans may have otherwise received high cost loans. This is an important finding, given that the yard stick by which MRB is typically assessed is the prime rate; cost benefit analysis of the MRB program are based on the interest rate savings from the MRB subsidized loan to the prime rate. If the MRB strategy prevented borrowers from receiving high cost loans, the monetary savings from the reduced interest rate and fees would result in a much more substantial “benefit” calculation. Further, as we look for alternative financing strategies for borrowers post the subprime boom, it is essential to consider the extent to which MRB mortgages may be able to provide affordable financing to borrowers who would have otherwise been targeted for high cost loans (which are to a large extent, no longer available).
17Cutts, Amy & Robert Van Order. 2005. “On the Economics of Subprime Lending,” The Journal of Real Estate Finance and Economics, 30(2): 167-196.
18Carr, James H. & J. Schuetz. 2001. “Financial Services in Distressed Communities: Framing the Issue, Finding Solutions”, Fannie Mae Foundation, August.
19Ding, Lei, Roberto Quercia, Janneke Ratcliffe & Wei Li. 2008. Risky Borrowers or Risky Mortgages: Disaggregating Effects Using Propensity Score Matching. Working Paper, Center for Community Capital. University of North Carolina.
20Based on the HMDA LAR files for 2006, a loan is considered “high cost” if any cost information is reported, signaling that the APR on the mortgage is more than 3 percentage points above the comparable U.S. Treasury rate. The use of the new LAR high cost information (available to researchers and the public after 2005) is in line with other recent analyses of high cost lending. While the pricing data reported to HMDA is not able to confirm whether or not a loan is subprime, the reporting threshold (3 percent above US Treasury) was designed so that 98 percent of prime loans would not qualify under the threshold; thus while not a perfect measure for subprime lending, it represents an improvement over other geographically disperse measures for subprime lending. The primary weakness of the LAR high cost indicator is the difficulty comparing high cost information from year to year, as the APR on the loan is related to changes in the yield curve over time (see Avery, Breevort and Canner 2006 for a discussion of these and other limitations of the HMDA higher priced data).
MRB Loan Activity & High Cost Loan Activity
First, MRB and high cost lending volume in 2006, by county are compared. Only counties located within Metropolitan Statistical Areas (MSA’s) in Ohio, Indiana and Florida are included (as these are the only counties for which accurate data on MRB and high cost loan volume is available), resulting in 120 counties. In addition to MRB loan volume, other lending strategies (FHA, GSE and privately sold) are considered, as well as the structure of the lending institutions in the county (percent independent). Appendix A4 includes summary statistics for the Ohio counties included in this analysis. The following observations are made from the
descriptive comparison:
• An increase in the proportion of independent lenders in a county is associated with an increase in the share of high cost loans in the county, ranging from less than 20 percent of loans as high cost to 25 percent of mortgages with high cost reported.
• As found in other research, GSE and FHA DIRECT purchase volume is associated with a decrease in high cost market share. The top counties (in the upper 75th percentile) in terms of GSE or FHA market share have less than 20 percent of their total loan volume in high cost loans, compared with nearly 30 percent of total loan volume in high cost loans for the lowest counties (bottom 25th percentile) in terms of GSE and FHA lending activity.
• An increase in the proportion of loans sold to private investors or insurance companies is associated with an increase in high cost lending in a county.
• Of most importance to this analysis, MRB lending activity is associated with a decrease in high cost market share. The top counties (in the upper 75th percentile) in terms of MRB lending as a percent of total lending activity in the county have an average of 19.3 percent of their total loan volume in high cost loans, compared with high cost loan proportion of 27.4 percent for counties with low to no MRB loan activity (bottom 25th percentile).
Multivariate Analysis of High Cost Lending and MRB’s
While the descriptive analysis presented demonstrates an association between counties with high MRB loan activity and low High Cost lending activity in 2006, it is essential to simultaneously account for county characteristics and other lender (and market share) characteristics to determine the true effect from MRB loan activity. A sophisticated “multi-level” modeling procedure was used to conduct this multivariate analysis.
Predicted probabilities for key variables from the final model are provided in Appendix A6. While complete results and models are available from the principal researcher, a summary of the findings is provided below.
• Of most importance to this analysis, even after controlling for census tract, county level and other lender characteristics, the probability of a borrower in a county receiving a high cost loan is
significantly lower (6.3 percent) for borrowers purchasing homes in counties with a high proportion of MRB loan activity. Low income borrowers purchasing homes in counties with the highest MRB loan activity (15 percent of total loan volume is MRB mortgages) have a 19 percent probability of receiving a high cost mortgage in 2006, compared with 25 percent probability for borrowers purchasing homes in counties without any MRB loan activity.
• While FHA and GSE lending activity in a county is also associated with a decrease in the probability of a borrower receiving a high cost loan (5.5% and 6.2% respectively), the effect of MRB activity is the most substantive loan activity effect.
• Other factors, including lending structure (proportion of independent lenders) and county level factors (most notably, the median purchase price of homes in a county) are also predictors of high cost lending. However, even after controlling for these factors, MRB loan activity in Ohio, Indiana and Florida is still substantially associated with a reduction in high cost lending in 2006.
Change in the Predicted Probability of Low Income Borrower Receiving a High Cost Mortgage
County Demographics
Note: Only significant variables, at p<.05, are presented in the table. The percentages represent a change in the predicted probability of a borrower receiving a high cost mortgage, based on changes in county demographics and county mortgage market characteristics from their minimum to maximum observed values in the sample.
Appendix A: Supplemental Tables
Appendix A1: Indicators of MRB Originating Lender Regulatory Structure & Size, All States
Mortgage Company Subsidiary 24 10.96% 329 23.09% 5,319 19.82%
Independent Mortgage Company 60 27.40% 346 24.28% 7,733 28.81%
1Each lender-state unit is counted individually, thus while there are 181 unduplicated lenders, there are 219 lender units total participating in the MRB programs in Ohio, Indiana and Florida.
2Lending Institution Size is based on the number of branch offices and/or the asset size if available. This information is not available for all non-bank lenders (in particular, broker institutions and some
independent institutions); 161 of 219 lender county units have this information available.
Appendix A2: Mortgage Borrower Need, Comparison by Program in Indiana, Ohio & Florida 2004-2006 Indiana 33,492 51,000 55,000 42,000 42,000 34,000 31,332
Ohio 37,320 54,000 57,000 45,000 44,000 35,000 34,272 Florida 32,856 69,000 65,000 45,000 66,000 34,000 31,200 Median Loan to Income Ratio
Indiana 5,839 290,227 88,017 40,530 54,892 127,501 4,546
% LAR 2.01% 30.33% 13.96% 18.91% 3.57%
Ohio 15,469 488,019 146,052 49,456 83,633 190,120 11,388
% LAR 3.17% 29.93% 10.13% 17.14% 5.99%
Florida 5,581 1,169,371 278,464 47,769 282,229 204,923 4,495
% LAR 0.48% 23.81% 4.09% 24.14% 2.19%
Note: Loan Application Register (LAR) loans in the sample include owner occupied, home purchase, first lien originated mortgages only. The sample does not include loans purchased by institutions that were not originated by the institution.
Appendix A3: Borrower & Area Characteristics, Base Model
Variable Mean Std. Dev. Min Max
Borrower Characteristics
Credit Score 676.98 65.13 450 850
Credit Score (log) 6.51 0.10 5.98 6.75
FHA/VA Loan (Dummy) 0.57 0 1
Gross Monthly Income 3122.64 995.52 742 9986
Monthly Income (Log) 8.00 0.32 6.61 9.21
Debt Ratio >41% (Dummy) 0.41 0 1
Loan to Value Ratio 0.96 0.08 0.21 1.29
Minority Status (Dummy) 0.16 0 1
Days Since Purchase 875.15 280.90 459 1583
Days Since Purchase (Log) 6.73 0.31 6.13 7.37
Indiana (Dummy) 0.19 0 1
Florida (Dummy) 0.21 0 1
Ohio (Dummy 0.60 0 1
County Economic Indicators
Unemployment Rate (04-06) 5.09 1.08 2.10 13.10
Change in Unemployment -0.74 0.45 -2.50 0.70
Median Income (2005) 58452.61 5509.46 34950 67350 Median Income (log) 10.97 0.10 10.46 11.12 Income Change (05-06, thousands) 1.37 1.85 -12 12.6 MSA Economic Indicators (n=19,348)
Median Sale Price (2007, thousands) 151.34 48.99 78.90 365.50
% Change in Sale Price (1st Qtr 07 to 1st Qtr 08) -0.09 0.05 -0.22 -0.01 Census Tract Characteristics
Population (in thousands) 4234.70 1.57 121.00 19904
Population (log) 8.35 0.45 4.80 9.90
Percent in Urban Areas 0.74 0.41 0.00 1.00
Percent with >= High School Ed 0.81 0.09 0.35 0.99
Percent Owner Occupied Housing 0.73 0.14 0.02 1.00
Average Year of Tenure 1989.36 4.06 1969 1999 Tract AMI as Percent of County 0.94 0.23 0.11 3.13
Percent Minority 0.15 0.18 0.01 1.00
Note: Unless indicated, the number of observations is 24,456 for all variables, except minority status.
The number of observations with minority status reported is 23,328.
Appendix A4:
Loan Performance as of March 2008, MRB Loans Closed 2004‐2006
Ohio Indiana Florida
Ever Late > 60 days or Delinquent 12.72% 21.80% 15.55%
Ever Late > 60 days (past 24 months) 10.12% 17.78% 13.92%
Bankrupt 0.91% 1.28% 0.41%
Foreclosed 3.71% 7.58% 2.69%
Paid in Full 2.97% 4.51% 10.16%
Days Since Purchase 846 976 918 n 15469 5839 5581
Appendix A5: Descriptive Statistics by County located in MSA's (Ohio)
WASHINGTON 17% 5.30 0.90 11.05 34% 10.82 102% 17% 23% 7% 11% 5%
WOOD 14% 5.20 0.93 11.70 39% 10.98 101% 23% 52% 2% 9% 2%
Economic Indicators Lender Indicators
Appendix A6: Predicted Probabilities- Receiving a High Cost Loan, Low Income Borrowers1
Co-Borrower 23.98% 17.13% -6.85%
Female 23.98% 23.69% -0.29%
Neighborhood Characteristics
State Dummy: Indiana 23.98% 22.88% -1.10%
State Dummy: Florida 19.54% 23.98% 4.44%
Census Tract Denial Rate 21.27% 26.91% 5.64%
Census Tract Loan Volume (%) 24.53% 12.44% -12.09%
Census Tract Population (log) 23.88% 24.08% 0.20%
Census Tract % ≥ Highschool 24.56% 23.41% -1.15%
Census Tract %Minority 24.09% 23.87% -0.22%
Census Tract % County AMI 24.91% 23.08% -1.83%
County Level Variables Economic Indicators
Change in Unemployment (06/05, %) 23.64% 24.33% 0.69% 22.63% 26.10% 3.47%
MSA Median Purchase Price (2006) 26.54% 21.59% -4.95% 29.25% 11.89% -17.36%
Median Income (2006, log) 24.60% 23.37% -1.23% 27.57% 22.12% -5.45%
County Loan Volume (% of State) 24.95% 23.03% -1.92% 25.28% 17.16% -8.12%
Lender Indicators
Bank Regulation (% Independent) 22.68% 25.33% 2.65% 17.72% 31.21% 13.49%
Market Share GSE Purchase 24.47% 23.50% -0.97% 26.41% 20.94% -5.47%
Market Share FHA Purchase 24.67% 23.30% -1.37% 26.33% 20.13% -6.20%
Market Share Private Sec/ Purchase 23.60% 24.37% 0.77% 22.92% 26.85% 3.93%
Market Share: MRB Originations 24.65% 23.32% -1.33% 25.34% 19.06% -6.28%
Average Lender Localness (% of State) 24.46% 23.50% -0.96% 25.28% 20.93% -4.35%
1All continuous variables are held at their means to compute predicted probabilities, unless otherwise specified. Binary variables are held at their modal value, as follows: State = Florida (Indiana = 0; Ohio=0); Black=0; Hispanic=0;
Asian=0; Co-Borrower=0; Female=0. Predicted probability at base = 23.98%
Appendix B: Lender Survey