Validating mortgage prepayment forecasts using a dynamic bivariate-choice regression method

Tingting Ji

Abstract

ABSTRACT

This paper illustrates the validation of a mortgage prepayment forecasting model using a dynamic bivariate-choice regression method. The results demonstrate that the dynamic bivariate-choice regression method can significantly enhance the forecasting of mortgage prepayment speeds for various mortgage products, as compared with the traditional logistic regression method. This is because the dynamic bivariate-choice regression method can incorporate the effects of time-dependent variables (eg, house price or interest rate) on prepayment. Validation is performed by using parallel testing and incorporating the dynamic features of interest rate, house price and borrower demographics. Backtesting and risk sensitivity analysis are also performed to compare the model forecasts from the two parallel methodologies against historical model performance.