16RT14 A transitions-based framework for estimating expected credit losses

This paper presents a framework for estimating losses for residential mortgage loans. At the core is a transitions-based probability of default model which yields directly observable cash-flows at the loan level. The estimated model includes coefficients on unemployment, Loan to Value ratio and interest rates, all of which allow a macroeconomic scenario to be fed through the model and impact loans’ probability of default and cure. Other loanlevel covariates such as bank, Buy-to-Let status, and vintage also impact loans’ transition probabilities. Loss Given Default is also modelled over a three-year horizon combining loan-level collateral information with macroeconomic house price forecasts. The breakout of flows from the stock of defaults allows the impact of loan modifications on recovery rates to be modelled. Unlike other models of mortgage credit risk, this framework allows a hysteresis effect of the time spent in default on the probability of loan cure to be modelled explicitly. In Ireland, an increase in the time spent in default from three months to one year leads to a decrease in the probability of loan cure from 30 to 12 per cent.