Motivated by the recent dramatic decrease in option adjusted spreads for especially 6% 2041 and 7% 2041 bonds we introduce a new specification of the burnout function in the DMBS model. 

The current model generates too high prepayment rates relative to the actual prepayment rates for high coupon bonds resulting in large negative option adjusted spreads. The high estimated prepayment rates are caused by very high prepayment gains for high coupon bonds in the current market.

We find that a plausible explanation for the general overestimation of prepayment rates in the current DMBS model is too little burnout effect for pool factors above 10%. Moreover, the current approach of using outstanding amount as weight in the estimation reduces the effect of bond series with low pool factors on the estimated parameters since they typically have low outstanding amount as well.

Burnout describes the tendency of bonds with high historical prepayments to prepay less than similar bonds without high prepayment history. The burnout effect is modelled by dampening the calculated loan group specific gain.

With effect from the quarterly update of the DMBS model 19 July 2010 we modify the model by introducing a new specification of the burnout function. Until now the burnout effect has been modelled by a normal distribution function, but this will be changed to a power function where the exponent is the only parameter. The modification implies a more aggressive dampening of estimated prepayment rates at high pool factor levels compared to the current model.

Our analyses show that prepayment estimates from the new model are generally more in line with actual prepayments for high coupon bonds. Considering calculated key figures for the last two years the new model generates higher and more stable option adjusted spreads, albeit still negative for high coupon bonds, as well as option adjusted durations compared to the current DMBS model. 

In addition to the introduction of a new burnout specification, we change the estimation procedure for the prepayment model by using equal weights for all observations.

Furthermore, we have made a slight adjustment to the Preliminary Redemption model in which the arrival intensity of preliminary redemption is dependent on the level of prepayment predicted by the prepayment model. This extended adjust model was analyzed in the April 2010 version of the ScanRate Prepayment and Valuation Service.

The full analysis of the changes to the DMBS model is available for ScanRate Prepayment and Valuation Service (SPV Service) customers in the July 2010 release.