Prediction and Modelling of Mortgage Prepayment Risk in a Low Interest Rate Environment Using Time-Varying Parameters

In this paper we estimate the prepayment risk of American mortgages on the residential market. Since the housing prices increase, whereas the mortgage interest rates decrease, we believe that correlations and variable coefficients differ before and after the financial crisis of 2008. The currently used survival analysis, MNL model and Markov switching model fail to take the fluctuation of parameters over time into account, but show differences between distinct intervals. We use a time-varying Markov switching model with dynamic parameters and a generalized auto-regressive score to investigate the presence of time-dependency. Both the results of five state and two state model are not significant, meaning that we find no evidence of timedependency. Overall we conclude that we are unable to improve upon the currently used MNL model in terms of mortgage prepayment estimation.