Estimating Conditional Mortgage Delinquency Transition Matrices (WP-2020-05)

Abstract

In this paper, we outline alternative methods for constructing transition probabilities conditional on both borrower-/loan-specific and macroeconomic factors. We define the transition states in terms of monthly delinquency and prepayment behavior in which payment behavior is measured in terms of days past due over a reoccurring, fixed-length (monthly) billing cycle: a discrete-time model design that allows us to better capture consumers’ decisions to continue servicing their debt, delay payment, prepay, or default. We find that for a large sample of active, first-lien, single-family, owner-occupied mortgages from 2004-2013 the assumptions of a Markov chain do not hold and by conditioning the estimates of the transition probabilities on loan-specific and macroeconomic factors, we generate more accurate out-of-time forecast over a time horizon typically used in practice (i.e., 24-months ahead) and are significantly more accurate during periods of changing economic conditions as was observed during the 2008-2010 financial crisis.

Authors

Qingqing Chen, Dennis Glennon and Amos Golan