Q. AD&Co.’s prOAS concept sounds attractive, but does the MBS market actually use it?
A. We have received a lot of praise from both industry practitioners and academics. We have formulated well what Wall Street has known for years: MBS-specific risk is caused by prepayment uncertainty and is directional. Every prepay model is just a good guess leaving investors with enough risk to demand compensation: the OAS.

We have observed and quantified many interesting market facts. Here is one of them: using critical turning points of a very volatile 2003 market, we compared predictions of price changes for the entire Trust IO market with actual moves. Four times out of five, the prOAS model predicted moves better than the OAS model did.

Another notable fact that is hard to miss– sharp rate drops always inflate OAS for the MBS premiums; sharp surges do same for the discounts. This stubborn directionality can only be explained by common prepay model concerns: refinancing or turnover.

Q. Why are empirical durations reported by brokers shorter than AD&Co.’s prOAD for premium MBS?
A. Many brokers report empirical sensitivity to a point of the yield curve (such as the 10-yr swap), which practitioners interpret as “empirical durations.” Consider, for example, high-premium MBS such as FN7.5 or FN8.0. AD&Co.’s effective duration will be normally seen in a 1.5-1.7 yr range. During the first half of 2005, brokers reported near-zero and even negative empirical exposure to the 10-yr rate. These two views may not be as contradictory as they seem. Premium MBS are driven by interest rates that are much shorter than the 10-yr. With the curve being flattened, a positive exposure to the 2-3-yr sector may be translated into a seemingly negative exposure to the 10-yr rate. Hence, the reported number represents a methodological confusion that should not be compared with the option-adjusted duration.

A more accurate assessment of empirical duration is required for a well-based judgment. One possible approach would be to produce a complete set of empirical key rate durations first and then add them up.

 

 

>>>