“Cure” Effects and Mortgage Default: A Split Population Survival Time Model

Abstract

Using non-agency securitization data consisting of mortgages originated between 1991 and 2007, we find that fraction of defaulted mortgages increases from 10.8% in the pre-crisis period (July 2007) to 19.6% in the post crisis period (July 2009). This paper then applies a split population hazard model, or widely known as a mixture cure model in biometrics literature, to jointly predict incidence (probability) and latency (hazard rate) of mortgage default, and more specifically to analyze the right-tail characteristics of the survival distributions of the sample mortgages. Our results show that negative equity is a highly significant factors driving both default incidence and latency for both adjustable rate mortgages (ARMs) and fixed rate mortgages (FRMs). We also show that borrowers’ credit scores have weaker predictive effects on the latency (survival hazard) risks, but both low FICO and subprime borrowers have higher probability to default in an adverse market condition, ceteris paribus. Borrowers with low credit scores have high default probabilities but with a longer time to default (hazard rate). They are more likely to hang onto their mortgages even if they are “underwater” (negative equity). In terms of the “cured” rate, we show that subprime borrowers with underwater (negative equity) FRM mortgages, and prime ARM borrowers have relatively higher “cured” rates in the mortgage sample. More policy experiments can be conducted using the mixture cure model to test the effectiveness of selected financial assistance programs in the future.