The theme of this SPV concerns preliminary redemption modelling. Once a week the Danish mortgage institutions publish the amounts already redeemed by the borrowers in each mortgage backed bond for the next unpublished payment date. The preliminary redemption model utilises this information to improve the mortgage model forecast of the prepayment rate on the next unpublished payment date.
A relative prepayment intensity is used to model the arrival pattern of the early redemptions over the period between publications. The current relative prepayment intensity is a function of time since the last publication only. However, as shown in this theme the redemption data indicates that the functional form of the arrival pattern also depends on the level of prepayment.
Observed intensities are often close to exponential. However, while the exponential curvature is very pronounced in a high prepayment rate scenario, the actual intensity is often closer to linear when low prepayment is observed. To incorporate this dependency into the preliminary redemption model an extension of the current intensity is proposed. The intensity is extended such that the amount of exponential curvature depends on the level of prepayment.
Preliminary redemption models with the original and extended intensities are estimated on each quarter during the last five years using one and five years rolling windows of estimation data. The forecasting abilities of the models are compared, and examples of the estimated intensities and their impact on option adjusted values and sensitivity measures are given.
Two main conclusions are made from the analysis. First, using only one year of estimation data leads to unstable parameter estimates over time. In turn this may lead to significant jumps in option adjusted values and sensitivity measures when switching between estimated models. Second, results indicate that the model based on the extended intensity yields marginally better forecasts of future prepayment rates.
In continuation of the theme from the previous release, concerning prepayment modelling for interest only bonds, we have this time conducted a follow up on the topic which can be found in the ‘Results’ chapter. Last time the analysis was concentrated on the DMBS model, but this time we focus on the M1 model as well. We mainly examine the effects from including interest only bonds in the estimation sample. We find that this extension leads to significant changes in the parameters of the M1 model, while the effects are only minor for the DMBS model. Different analysis show that we obtain higher forecasting performance for both models if we include the interest only bonds in the estimation. Subsequently, the estimation sample has been extended.