 |
 |
the value of CMO, for the same OAS and market conditions. If a CMO class is much longer than its collateral, the curve’s steepening (without change of the MBS current rates) extends the tranche’s life and impairs its value relative to the collateral. When the curve flattens or inverts, gains in discounting are limited by a shorter life. Therefore, on average, the CMO’s price profile will exhibit negative convexity to twist. Short classes behave in the opposite way and are positively convex. Similarly, valuation of caps found in floaters relies on the interaction between prepay speeds and the odds of hitting the cap. When the curve steepens, the cap moves into the money and applies to a larger surviving balance. The losses on resets won’t be offset by much smaller gains induced when the curve flattens or inverts.
Here is the action guide: When selecting the number of factors in an interest-rate model, ask yourself what type of instruments represents your main investment focus. If it is MBS pass-throughs (including ARM pools, IOs, POs, and MSRs), use the Hull-White model with confidence. And, if you license AD&Co.’s OAS model, you will take advantage of the blazing speed of backward induction. If most of the instruments are CMOs, I recommend you think and learn more about AD&Co’s two-factor normal model. Calibrated to the same swaption matrix, this model may shed light on the true value hidden by single-factor modeling. And it won’t increase your processing time because, with CMOs, you are bound to Monte-Carlo anyway. AD&CO.’s OAS system supports 4 term structure models and lets you conduct your own tests and compare results.
1. Gauge the accuracy. The AD&Co. OAS system reports an estimate of pricing accuracy. If you quote price and compute OAS, “RMSE” is the OAS accuracy. Otherwise, it is the price accuracy. The actual achieved accuracy is somewhat better; RMSE stands for simple sampling standard deviation of values, but it suits the purpose. Review your accuracy and processing time and adjust your set-up accordingly. Reduce the number of paths if the accuracy is redundant. Increase it if the accuracy is deficient.
2. Set a different number of paths for different Intex instruments. Run more paths for agency CMOs, less paths for non-agencies - not because of accuracy, but because of ability. In a typical portfolio batch, setting 100 Monte-Carlo paths for non-agency CMOs may be the most one can afford.
