Valuation Commentary - October '06

Valuation Consequences of the New ARM Prepayment Model in v5.2a
Part II: Non-Agencies Take off their Mask
by Alex Levin

Version 5.2a features two non-agency ARM models, a non-agency prime model, and a sub-prime model. These models have been built using our unified framework. Much like their agency counterparts, they are designed within the Active-Passive (APD) structure, 4 sources of prepayments (turnover, refinancing, cashout, and cure) and the same set of tuning knobs. The Spread-at Origination (SATO) effect plays a prominent role in our sub-prime model, whereas the weighted average original loan size (WAOLS) and weighted average original LTV (WAOLTV) are the corrective factors of refinancing, turnover and cashout speed multiples. It is clear that it is impossible to compare fairly prime and sub-prime instruments without accurate assessment of their credit differences. In this short article, we review valuation aspects of non-agency prime ARMs and sub-prime ARMs as a consequence of the prepayment behavior.

Jumbos Are Faster, but Not Cheaper

Prime fixed-rate borrowers of jumbo size tend to exercise the refinancing option more efficiently. Although a same-OAS or same-Price valuation exercise can be carried out with ease, the most popular task in the primary market is to compute a well-defined rate spread offered to jumbo borrowers. This important practical measure facilitates the lending of jumbo loans based on the difference in exercise efficiency (given similarly adjusted credit) as captured by a prepayment model. In particular, we may start with comparing jumbos and agencies having the same coupons and GWACs; this modeling setup adjusts jumbo ARMs’s credit to that of agency ARMs (assuming there is no difference in credit of the underlying loans).

To illustrate the solution, we first measure the current coupon rate; on September 22, it was close to 5.5 for the Fannie Mae 5/1 with 2/2/5 caps. Using the market price of 100-04 and our new agency hybrid model, we compute LOAS of 20.7 bps. We now fix this LOAS and employ it for all calculations for non-agency prime ARMs. For each combination of WAOLS and WAOLTV, we can iterate for GWAC until the same prices are achieved as in the agency case.

This modeling exercise results in a somewhat counter-intuitive finding: same-OAS, same-Price GWAC is lower for jumbo ARMs than for agency ARMs by 5-15 bps. In other words, jumbo hybrids should be