Business Services Industry

Third party originators and mortgage prepayment risk: An agency problem?

Journal of Real Estate Research, The, 1999 by LaCour-Little, Michael, Chun, Georgory H

Data

Two sets of data are used. A national mortgage loan-servicing firm who prefers anonymity provided the first data set, consisting of loan level information on 16,974 fixed-rate mortgage loans originated during calendar year 1992 at an average note rate of 8.45%. Collateral is highly diversified geographically, with the largest concentrations of loans from New York, California and Illinois. About half are thirtyyear and half fifteen-year term. Both conforming and nonconforming loans are included. Loan payoff behavior is tracked through 1997 and defaulting loans are excluded. By year-end 1997, about 80% of these loans had prepaid. While additional data on adjustable rate mortgages (ARMs) was available, defining the borrower's incentive to refinance in the case of ARMs is much more difficult given annual rate adjustments. Accordingly, ARM loans were dropped from the final analysis and will be the topic of later research. Exhibit 1 gives summary statistics on the loan level data set including geographic distribution and payoff experience by year.

Since loan level data was from a single servicing firm, the question arises whether results may be generalized to the broader mortgage market. To test for similarities to the overall mortgage market, a second data set from the commercial provider

Mortgage Information Corporation (MIC) was examined. MIC compiles delinquency, default and prepayment data in cross-tabular format based on a population of some twenty-four million loans nationwide. Since loan level information is not available in this data set, we confine our analysis to a comparison of mean prepayment speeds by loan product and origination year stratified by TPO versus retail origination channel.

Results

Initially, separate analyses were performed on the thirty-year and fifteen-year, conforming and non-conforming, loan types. Each exhibit first shows the descriptive statistics for that particular product type, at the loan level, and then the results of the logistic regression. Exhibits 2-9 show results for thirty-year conforming FRM, the fifteen-year conforming FRM, the thirty-year jumbo FRM and the fifteen-year jumbo FRM, respectively. Exhibits 10-11 show the results for all loan types, but separately estimated for TPO versus retail loan types (controlling for loan term and conforming loan status through additional indicator variables, JUMBO and 15YEAR). The difference in the magnitude of the coefficient on the variable INCENTIVE between TPO and retail loans provides some indication of the relative sensitivity of each borrower type to refinancing opportunities. We expect TPO loans to be more responsive to refinancing incentives relative to retail loans.

While there are differences among product types, regression results are remarkably consistent, with most parameter estimates significant at the 99% level. MONTH is consistently positive, ranging in value from 0.07-0.26. MONTHSQ is consistently negative, indicating the time has a non-linear effect on refinancing probability. As expected, INCENTIVE is consistently positive, ranging in values from a low of 0.25 (for the fifteen-year conforming product) to a high of 1.35 (for thirty-year nonconforming FRM). These results confirm the usual finding that prepayment probability increases with loan age and, of course, increases dramatically as the spread to market rate increases (as the option to prepay is in-the-money). Moreover, the magnitude of the coefficient on INCENTIVE is larger in the case of the thirty-year loans, compared to the fifteen-year loans, consistent with the notion that borrowers obtain greater benefits from refinancing as remaining loan term increases. The effect of loan size (ORIBAL) is not consistent, and is negative (but not statistically significant) in the case of thirty-year non-conforming FRM. Likewise, the effect of income is not statistically significant in three out of the four regressions and, in the case in which it is statistically significant, the magnitude is too small to be of any economic significance. The indicator variable for a third party originator (THIRDPTY), however, is large, positive and highly statistically significant across all regressions, ranging in value from 0.45-3.98.


 

BNET TalkbackShare your ideas and expertise on this topic

Please add your comment:

  1. You are currently: a Guest |
  2.  

Basic HTML tags that work in comments are: bold (<b></b>), italic (<i></i>), underline (<u></u>), and hyperlink (<a href></a)

advertisement
advertisement
  • Click Here
  • Click Here
  • Click Here
advertisement
Click Here

Content provided in partnership with ProQuest

Most Recent Business Articles

Most Recent Business Publications

Most Popular Business Articles

Most Popular Business Publications