The theme of this release concerns our model for bonds with semi-annual payments, M0. This model differs from the other SPV models in that it uses a simplified version of the prepayment function which does not incorporate the delivery option or any burnout measure. Moreover, the model is estimated based on a sample comprised solely of semi-annual bonds. Our primary task is to test whether the recommendations we have outlined for the use of this model seem appropriate. These are as follows:
1. Since the M0 model does not capture burnout, it should only be used for bonds with similar pool factor values. When we developed the model, we found that a very large percentage of the semi-annual bonds are old and rather burned out. Consequently we recommend that the model should only be used in the pricing of semi-annual bonds which opened before January 1993. Moreover, the estimation sample is also restricted to only include semi-annual bonds that opened before this date.
2. The M0 model is used in the pricing of all semi-annual bonds that opened before January 1993 - both the ones with and the ones without published debtor distributions. In cases where we do not know the breakdown of borrowers, we assume that all loans are concentrated in the loan group consisting of the smallest loans.
3. Semi-annual bonds that opened after January 1993 are priced within the DMBS model. When we have debtor information we use this; otherwise we assume that all loans are concentrated in the loan group consisting of the smallest loans.
We start by introducing an alternative version of the M1 model which is estimated on the basis of a larger estimation sample than the regular M0 model since we disregard the restriction on the opening date. We term this model M0_Ext. A comparison of this model to the regular M0 model leads to the conclusion that the regular M0 model slightly outperforms the extended. Moreover, we find that the best prepayment estimates from the M0 model are obtained by assuming that all loans are concentrated in the group consisting of the smallest loan sizes – group 12. Next the M0 model is compared to the DMBS model. As for the M0 model we establish that the best results for DMBS model are obtained by concentrating all loans in loan group 12. Comparing the single loan group versions of the two models we reach the following conclusion; for the in-sample bonds, i.e. the ones that are in the estimation sample of the M0 model, the M0 and DMBS models perform equally well, while the DMBS model outperforms the M0 model out-of-sample.
Thus point 1 in the list above seems to hold while point 2 and 3 are more debatable.
Another interesting topic of this release is an extensive analysis of the M1 model. The motivation for this analysis is that the parameters of the M1 model have changed significantly compared to the parameters of the corresponding model from the previous release. We explore how the changes affect the estimated prepayment rates, and we also try to resolve why the changes have emerged.