Özet:

Mortgage-backed securities (MBS) are structured financial products that are produced via securitization of mortgage loans. Due to the nature of securitization, all risks of mortgage loans are transferred from originators to MBS investors. Prepayment and default risks of mortgages lead to uncertainty in MBS cash flows and create a complex problem for valuation of these instruments. Therefore, estimating these mortgage termination risks has become the focus of valuation of MBS collateral pools. This study explores two questions by using a publicly open dataset provided by Fannie Mae. First, two machine learning algorithms (Random Forest and Multinomial Logit Regression) are used for classification to predict whether a mortgage loan is likely to be prepaid, defaulted or current. Afterwards, Competing Risks Cox Regression Analysis is performed to see determinants of when mortgage termination risks are likely to happen. It is found that not all mortgage borrowers behave optimally in their prepayment and default decisions. Therefore, in addition to refinancing incentive and negative equity which depend on variations in prevailing mortgage interest rates and housing prices, heterogeneity in mortgage borrowers’ behaviors and loan characteristics, and also local economic factors are significantly important in estimating mortgage termination risks. It is worth noting that prominence role of mortgage payment delinquencies in particularly predicting defaults emphasizes the essential need of monitoring payments by servicers to keep safety of MBS investors and financial markets.

Anahtar Kelimeler: