![]() |
|
||
|
|
The data represented in the table above reflect FHLMC’s entire universe of standard Treasury-based hybrid ARMs (non-IO, non-prepay penalty) that are still within their initial reset period. The coupon classes represent the gross weighted-average coupons on the pools, aggregated into 25 basis point buckets, with the amount of outstanding balance reported for each bucket. Looking at model performance on one-month speeds for the different versions, it is clear that the new 5.2a model is much more accurate across all collateral types, and the performance is very strong. Prepayment speeds on hybrid collateral dropped from Oct. to Nov. by around 10% on average across coupons for both the 3/1 and 5/1 hybrids, and by about 5% and 12% on average for 7/1 and 10/1s, respectively. Both models responded to the drop, but version 5.2a was much closer to actuals across the board. The overall model error for version 5.2a was 1.7 CPR across coupon and collateral types, while the error for 5.1f was 3.0 CPR across coupon and collateral types. This comparison demonstrates that version 5.2a uses its new functional form and new model features to achieve a greater level of accuracy in predicting prepayments when compared to the previous version. Clients may go to http://dynamic.ad-co.com/performance/ to link directly to the performance reports used to generate these numbers and analyze the data in more detail. Page 2 of 2
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||