Refinancing transitions and equity extraction among CRA mortgage borrowers

Abstract

Refinancing transitions can alter both the long-term cost and the sustainability of a homeownership tenure. However, because few datasets allow researchers to follow homeowners from one mortgage to the next, little research exists regarding the extent and nature of refinancing transitions. This article uses uniquely rich data from the Community Advantage Program (CAP) to examine the refinancing decisions of a sample of low-income borrowers with 30-year fixed-rate purchase mortgages. While the majority of refinancing CAP borrowers secured lower-cost prime loans, a minority refinanced into adjustable-rate mortgages and into products with above-prime interest rates. The empirical analysis documents the refinancing transitions made by CAP borrowers and explores the role of equity extraction in the refinancing decision. Refinancing motivated by a desire to secure a lower interest rate is shown to be substantively distinct from refinancing that includes equity extraction. This difference carries over into borrowers’ selection of refinancing products, as borrowers who extract equity transition more frequently into non-prime FRM and ARM products than borrowers that rate refinance.

Keywords:

• refinancing
• homeownership
• CRA

Notes

¹The analysis dataset for this article includes loans originated by 15 lenders. These lenders include five of the ten largest banks and eight of the 50 largest banks in the United States. The remaining seven banks include mid-size regional banks, several of which were acquired in the years following their participation in CAP. These lenders originate CAP mortgages in 22 states. The originating banks span the US, although concentrations of CAP originations are present in the Southeast and Midwest (and in North Carolina and Ohio specifically).

²Among the CAP loans in the analysis dataset, 89 percent qualified for purchase under the first criteria.

³See Freddie Mac's Primary Mortgage Market Survey: http://www.freddiemac.com/dlink/html/PMMS/display/PMMSOutputYr.jsp?year = 2008

⁴The interest rates reported for the prime market by PMMS exclude borrowers’ mortgage insurance payments.

⁵This sample reflects a random subset of all non-rural mortgages in the CAP portfolio that were originated during this period. The sample is drawn for data collection due to the costs associated with in-person interviews, which collect information about the refinanced mortgage. This sample is discussed further in the empirical analysis section.

⁶This definition of cash-out refinancing includes 82 refinances in which the borrower both extracted equity and secured a lower interest rate. As a result, this measure may overstate the presence of cash-out refinancing to the extent that some borrowers who extracted equity refinanced primarily to secure a lower interest rate.

⁷Among borrowers whose rates had reset prior to the 2005 interview, the current interest rates averaged 5.96 percent, compared with introductory rates that averaged 5.82 percent. This average rate, however, likely does not capture rate resets on any 2/28 or 3/27 hybrid ARMS, as most refinancing occurred within 2 to 3 years of 2005 interview.

⁸See also Deng, Quigley, and Van Order (2000).

⁹The correlation coefficient comparing the rate spread measure with the measure of the call option defined by equation (2) exceeds 0.95. Because these variables are virtually collinear, the rate spread measure is used in the empirical analyses for its ease of interpretation.

¹⁰Home value estimates are derived at multiple time points using Fannie Mae's automated valuation model.

¹¹The model was also estimated using the loan age in months and the loan age squared to define the baseline. The substantive findings are similar across models.

¹²This third wave also defines the sample used in this analysis, as the costs associated with in-person data collection required that a sample of CAP borrowers be drawn. In-person interviews were collected for a random sample of non-rural mortgages in the CAP portfolio. Of the 1,284 completed interviews, 1163 contained complete information on all measures.

¹³Interpretation of the estimated coefficients in MNL models must be performed with care, as the sign of a coefficient may differ from the sign of the underlying marginal effect under certain conditions. Derivation of the marginal effects confirms the sign of the estimated coefficient (and the substantive interpretation) for all of the variables that reach significance in the results presented. The coefficients are presented rather than the marginal effects because the hazard structure of the underlying dataset requires that some assumption about time be imposed in order to derive a meaningful marginal effect.

¹⁴Home values were estimated at three time points using Fannie Mae's automated valuation model. The values for intervening months were linearly interpolated using these estimates.

¹⁵It is also possible that some of the non-prime ARM refinances represent transitions into exploding ARMs or hybrid products with undisclosed prepayment penalties or other fees. While these characteristics cannot be observed in the CAP dataset, they were common among non-prime ARM products during this period.

¹⁶The primary exception is Courchane, Surette, and Zorn (2004), who document transitions between prime and subprime mortgages through refinancing. While the focus of the analysis is on the ability of subprime borrowers to refinance into prime credit, the survey documents that 13 percent of prime borrowers who refinance transition into subprime products. Pennington-Cross and Chomsisengphet (2007) examine equity extraction and refinancing out of subprime mortgages, but focus on mortgage termination and do not observe the subsequent product.