Modelling prepayment risk in residential mortgages

The current low interest rate environment has triggered refinancing incentives in the residential mortgage sector. An unscheduled return of a part or the full outstanding principal constitutes a risk from the perspective of financial institutions providing mortgages. On the one hand, prepayments affect the Asset and Liability management of a bank. On the other hand, prepayments lead to interest rate risk. Given the magnitude of the residential mortgages on the balance sheet of a bank, it is of vital importance to obtain insight in the actual maturity of the mortgages provided by financial institutions. This thesis looks into different models that can be used to predict current and future prepayment rates. The most widely used prepayment models are option theoretic models, multinomial logit models and competing risk models. Aside from these models this thesis investigates the applicability of the Markov model as a prepayment model. Important determinants of prepayment include borrower specific characteristics, loan specific characteristics and macro-economic variables. Variable selection procedures are used to identify the most important risk drivers. Models are compared based on a wide range of in- sample and out-of-sample performance criteria to determine the model that is most appropriate for predicting prepayment rates. Another important feature that the models should be capable of incorporating is the recent credit crisis of 2008. Tests on parameter stability are conducted to determine the possible presence of structural breaks in prepayment models.