Deep Learning for Mortgage Risk by Justin Sirignano, Apaar Sadhwani, Kay Giesecke :: SSRN

We develop a deep learning model of multi-period mortgage risk and use it to analyze an unprecedented dataset of origination and monthly performance records for over 120 million mortgages originated across the US between 1995 and 2014. Our nonparametric estimators of term structures of conditional probabilities of prepayment, foreclosure and various states of delinquency incorporate the dynamics of a large number of loan-specific as well as economic and demographic variables at national, state, county and zip-code levels. The behavior of mortgage risk can vary strongly depending upon the geographic region. Moreover, the relationship between factors and mortgage risk is often highly nonlinear. Higher-order interactions between multiple factors are prevalent.