In our previous blog, we showed the difference, or ‘spread’, between the average contract interest rate for jumbo and conforming loans during the last 17 years, without adjusting for credit risk, property location, or scale economies. As a follow-up, this blog estimates the adjusted jumbo-conforming spread by controlling for the major loan, borrower, and property characteristics that affect mortgage interest rates, such as loan size, credit score, loan-to-value (LTV) ratio, debt-to-income (DTI) ratio, and home location.
Figure 1 displays the average contract interest rate in 2018 (first six months) compared to 2009 by loan origination amount, expressed as the difference between the loan amount and the local-market conforming loan limit. The blue line in the figure shows that the interest rates in 2018 declined subtly with the loan amount until the conforming loan limit was reached. Then the average rate dropped abruptly by 27 basis points before starting a gradual decline again. The chart shows an inverse relationship between the interest rate and the loan origination amount. The general trend reflects various fixed-costs of origination; in other words, the fixed-cost per dollar of the loan size declines as the loan size increases. Similar to 2018, the interest rates in 2009 declined gradually with the loan amount until the conforming loan limit was reached. However, then the rates took a sharp 85 basis point rise. The figure implies that the historical trend of mortgage rates spiking above the conforming loan limit has reversed and in 2018 the jumbo loan is cheaper than conforming.
As pointed out in the previous blog, the recent jumbo-conforming spread may have been influenced by the lower credit risk attributes of jumbo loan borrowers and risk-based pricing, and differences in scale economies and property location. Thus, to control for these effects we employed a regression specification derived from Hendershott-Shilling to estimate the impact. We use the CoreLogic loan level mortgage data for Q1 2001 through Q2 2018 to estimate the effect of jumbo status on the contract interest rate for conventional 30-year fixed-rate home-purchase loans. Mortgage rate is expressed as a function of loan jumbo status, loan size, credit scores, LTV ratios, DTI ratios, condo-coop status, state location of property, and origination week.
Figure 2 plots the estimates for the jumbo-conforming spread from the regression equation ran for each quarter. The overall pattern of adjusted spread followed the pattern of unadjusted spread. Our analysis found that the contract interest rates on jumbo mortgages remained slightly lower than the rates on conforming mortgages for 2018Q2 even after controlling for credit risk, location, and scale economies. However, the spread narrowed from negative 33 basis points (with no adjustments for differences in attributes) to negative 5 basis points for the adjusted estimate in Q2 2018. As expected, due to risk-based pricing, the regression results show that the mortgage rate is influenced by credit score, LTV ratio, and DTI ratio.
Figure 3 reports the coefficient estimates for both the unadjusted and adjusted regression models for Q2 2018 and Q2 2009. The results show the mortgage rate rises as the credit score declines in both years. The excluded category of loans, those with credit score above 780, has the lowest default risk in the sample, so it serves as the benchmark. For a borrower with credit score between 640 and 680, the loan interest rate is higher by 45 basis points than for a borrower with credit score above 780 in Q2 2018. Similarly, as the loan amount rises, the interest rate falls because of the declining average fixed cost per dollar of originating and servicing loans.
This analysis illustrates the importance of controlling for loan, borrower, and property characteristics when comparing the contract rates on conforming and jumbo loans. Doing so shows that jumbo loan rates, which appear to have been 33 basis points lower than conforming rates before making any adjustments for differences in attributes, were actually priced very close to conforming loans: we estimated jumbo loans were only 5 basis points below the rates on comparable conforming loans during 2018Q2.
 For regression details see Patric H. Hendershott and James D. Shilling, “The Impact of the Agencies on Conventional Fixed-Rate Mortgage Yields,” Journal of Real Estate Finance and Economics, vol. 2, no. 2 (June 1989). We revised the model by adding some other relevant variables that were not in the original model such as credit score and DTI ratio.
 We only used 30-year, fixed-rate, first-lien, owner-occupied, purchase-money one-family loans.
 Various unobserved variables could affect the estimated spread. For example, our data set does not include points paid by the borrower; differences in the number of points typically paid by a jumbo and a conforming borrower could result in an effective spread that was larger or smaller than 5 basis points during 2018Q2.
©2018 CoreLogic, Inc. All rights reserved