Valuation Commentary - July 2006

TBAs, SATO, and Fannie 7.0
by Alex Levin

Those who follow AD&Co’s weekly market analyses should have noticed a recent richening of FNCL 7.0. In just 6 weeks, Libor OAS (LOAS) fell from +13 bps to – 36 bps. What caused this significant change? How credible is it? Do major brokers agree with us?

The price of this coupon has dipped slightly, from 102.50 to 102.34, remaining somewhat consistent with the small current coupon’s rise from 6.19% to 6.28%. It is easy to see that the LOAS drop is caused by a dramatic change of the assumed WALA for the TBA, not the price. The median age assumption has quickly gone from 40 months down to 8 months as follows: on June 9, median WAM was 319; the next week it became 332; on June 30, it already rose to 346, etc. Out of 11 major brokers contributing to Bloomberg’s VALL screen, only two currently believe that moderately seasoned pools will be delivered into TBA 7.0; six brokers thought so back in June.

Clearly, newer premium pools should trade at a much lower price than the seasoned ones; yet the brokers’ median for LOAS that we compile regularly, but do not disseminate, has gone down by only 10 basis points from 0 to -10. Therefore, the market makers apparently don’t think TBA 7.0 has richened to the same extent as AD&Co. Who is correct?

TBA Assumptions and SATO

Since AD&Co does not make the market and has no access to the full brokers’ pool inventory, we analyze the market using the median assumptions on WAC and WAM loaded from Bloomberg. We consider this method to be objective and practical but, admittedly, not very accurate.

To avoid spurious valuation effects when modeling the TBAs, we exclude the SATO effect (normally presented in our prepayment model). The SATO effect suggests that, all else equal, pools originated above prevailing market rates may be credit-impaired and will be temporarily slower in refinancing than the usual zero SATO pools.

It is imperative to know the exact age to implement SATO; operating on an average or median can make the use of SATO misleading. Indeed, let us assume that two cohorts are candidates for delivery: the 1-yr old and the 5-yr old. The median age is, therefore, 3 years, but the market rates 3 years ago have nothing to do with either of the cohorts. Hence, we can’t use the SATO effect in such a case.

On the other hand, there are situations when the TBA age is certain, and pools originated at some positive SATO level are going to be delivered. It seems, at first glance, we are now facing this exact situation with FNCL 7.0. Assuming that the 8-mo median age, indeed, matches the origination of FNCL 7.0, and looking back at the mortgage rates in November of 2005, we conclude that TBA-deliverables were originated at a SATO exceeding 1.0%. Once the SATO effect is activated in our model prepay speeds drop and the valuation results line up with those of the brokers (Exhibit 1).

Exhibit 1. Comparative valuation of FN TBAs as of July 21, 2006

It is clear that brokers use some sort of SATO adjustment in their models when evaluating TBA 7.0. The Bloomberg median speed, a modest 469 PSA, confirms this conclusion. Other TBA assumptions found in some dealer reports, such as average loan size, don’t seem prominent. It seems that we should stop hesitating and begin using the SATO effect for all TBAs (perhaps, until we see another bi-modal cohorts again). However,…

Does SATO Tell the Truth?

As Dan Szakallas points out in this month’s Model Performance section, the SATO effect is fading for agency pools. The actual speeds for recently originated FNCL 7.0 are much higher than most models anticipated. Historical speeds for the FNCL7N Bloomberg cohort have been in the 1500 – 2000 PSA range and stay there even with the current-coupon exceeding 6.0% for 4 straight months.

In the AD&Co model, recently originated 7s are much slower than actuals when the full SATO effect is used (SATO tuning = 1.0), but moderately faster than actuals when it is omitted altogether (SATO tuning = 0.0). The best current SATO tuning seems to be 0.33 (Exhibit 2), but it was around 0.6 in 2005, 0.75 in 2004 and close to 1.0 in 2002.

Exhibit 2. Recent prepayments of the 2005Q4 FNCL 7 (in CPR)

Month 
Actual
 Model Model
Model

 

 

SATO = 0

SATO = 1

SATO = 0.33

March 2006

35.82

61.05

13.85

35.53

April 2006

38.31

52.86

16.76

39.42

May 2006

39.71

49.40

17.70

39.00

June 2006

37.20

41.66

18.79

36.84

August 2006 CPR, and even the long-term PSA forecasts, depend on SATO and range from 16.9 CPR / 508 PSA (full SATO effect) to 35.4 CPR / 620 PSA (void SATO effect).

From the option-adjusted duration (OAD) stand-point, most brokers report numbers in the 2.0-2.1 yr range, visibly longer than the 10-yr swap-based empirical duration of 1.42 yr (www.lehmanlive.com). AD&Co’s OAD is reported as a short 1.12 yr (void SATO effect), but it would be 2.0 yr if the SATO is fully accounted for.

Conclusions

The SATO effect for agency MBS is now more dead than alive, but its inclusion drives today’s valuation of TBA 7.0. It seems that using no adjustment for SATO is the better evil, and the coupon comes out much richer than most dealer models show.