 |
 |
Prepayment Performance
New Burnout Tuning Recommendations
Prepayments slightly decreased from Dec. to Jan., the first overall
drop in a months. The holiday season appears to have had its yearly
effect. Both discounts and premiums posted lower speeds than in previous
months, though the drop was larger for FNMA pools than for GNMA pools,
5.3% and 3.8%, respectively. Below are actual pool CPR speeds of selected
coupon buckets, and corresponding model forecasts.
|
|
FNMA 15
|
FNMA 30
|
GNMA 15
|
GNMA 30
|
||||
|
Coupon
|
Actual
|
Model
|
Actual
|
Model
|
Actual
|
Model
|
Actual
|
Model
|
|
5.0
|
5.3%
|
12.4%
|
3.0%
|
3.6%
|
2.4%
|
4.7%
|
9.0%
|
6.3%
|
|
5.5
|
25.5%
|
37.0%
|
4.3%
|
7.9%
|
11.3%
|
22.9%
|
3.2%
|
5.5%
|
|
6.0
|
39.5%
|
56.6%
|
30.5%
|
36.4%
|
25.4%
|
34.7%
|
21.5%
|
25.2%
|
|
6.5
|
42.0%
|
49.0%
|
52.6%
|
58.5%
|
33.3%
|
35.8%
|
45.9%
|
49.7%
|
|
7.0
|
44.1%
|
32.6%
|
56.3%
|
59.9%
|
32.8%
|
31.6%
|
54.3%
|
53.5%
|
|
7.5
|
43.4%
|
27.3%
|
56.7%
|
49.7%
|
31.7%
|
28.7%
|
55.5%
|
47.6%
|
The AD&Co. model performed best for GNMA 30-yr pools in Dec., where it was slow by an average of only .3% CPR across all coupon types. For GNMA 15-yr pools the model was fast by an average of 2.8% CPR, due largely to the 5.5 and 6.0 buckets, as the model was accurate on the premiums. The model was fast for FNMA 15-yr pools by an average of 2.5% CPR across all coupon types, but there were differences across coupon buckets. The model was fast on discounts, but slower on premiums. Much like last month, prepayments increased only slightly across the last 4 coupon buckets, resulting in a very flat curve. The model was also fast for FNMA 30-yr pools by about 2.1% CPR. It was fast across both premiums and discounts. Overall, the model performed well across all buckets and collateral. Looking ahead, we expect prepayments to remain mostly steady through Feb., and increase beginning in Mar. as housing turnover usually increases coming out of the winter months.
Over the last year, we have closely analyzed how our model performs for different collateral, coupon buckets, and origination years. Recently, we found that for our 30-yr conventional and GNMA models, our burnout parameter has been too strong; we were predicting that some pools were burning out faster than they actually were, resulting in slower prepayment forecasts in older premiums. We looked at different time periods (last year, last 6 months, and last 3 months) to see how we could utilize our tuning >>>
