A Non-Parametric Prepayment Model and Valuation of Mortgage-Backed Securities

Abstract

A non-parametric technique called generalized additive model (GAM) estimation is particularly useful in high-dimension non-parametric estimations and in situations that involved mixed parametric and non-parametric specifications. The relationship between prepayment rates, and variables such as the age of the mortgage, the ratio of the mortgage coupon rate and prevailing interest rates, and expected and unexpected burnouts is highly non-linear, and it is difficult to capture these relations with parametric functions. Decomposition of pool burnouts into expected and unexpected components improves the model fit and has important pricing implications. The prepayment model estimated here fits the data significantly better than other models in use, and illustrates the factors that affect prepayments and the prices of mortgage-backed securities.