Shifting to the new version 5.1c prepayment model, we can see that the model performed well across all collateral types, with speeds being very close to actuals.

	FNMA 15		FNMA 30		GNMA 15		GNMA 30
Coupon	Actual	Model	Actual	Model	Actual	Model	Actual	Model
4.5	11.2	9.9	9.5	9.9	14.2	13.3	15.5	12.6
5.0	14.4	12.2	12.6	12.6	16.4	14.7	17.0	15.2
5.5	17.9	16.8	19.2	17.5	19.4	19.2	22.5	22.3
6.0	21.8	19.8	27.7	28.5	20.7	21.1	30.3	29.9
6.5	22.0	21.2	32.9	32.0	21.3	21.4	34.2	33.6
7.0	23.6	22.4	33.4	33.9	20.8	21.1	33.6	32.0
7.5	24.2	23.8	31.75	35.82	22.9	20.5	32.1	29.4

The model performed the best for the FNMA 30-year pools, just off by a balance-weighted average of -0.5 CPR. The FNMA 15-year model also performed reasonably well, as it was off by a balance weighted average of -1.6 CPR across coupons, mostly in the discounts. Both GNMA models performed much better than version 4.3.4. The GNMA 15-year model was off by a balance-weighted average of –0.9 CPR and the GNMA 30-yr model was only off by a balance-weighted average of -0.8 CPR across coupons. This shows that 5.1c is more accurate at forecasting speeds for the lower coupon GNMA discounts that seem to prepay at faster rates than their FNMA counterparts.

The September speeds showed the effects of both the rising pattern of interest rates from the lows seen at the end of June and the usual seasonal downturn in home sales seen in the fall months. These factors combined for the most significant drop in prepayments seen in many months. Looking ahead, we can expect October prepayments to drop noticeably once again since the seasonal factor will dampen speeds even more, and because mortgage rates hit a high for 2005 in the second week of October. This will cause another sizeable drop in discounts as well as a 15% or so decrease in cuspy premiums. We could be looking at a few months of very low prepayments if rates stay at their current levels into the winter months.