August 23rd, 2005, 8:41 am
QuoteOriginally posted by: pyatskiI have a tree that I use for valuing a Bermudian callable bond. I am satisfied with the accuracy of the prices (which is within 1 cent per $100 of notional). But the accuracy of the duration is not adequate (about .50 a year). Are there any "tricks" to increase the accuracy of the duration estimation?Unfortunately not really! Greeks (delta and even more gamma) are numerically instable in trees if you use a simple approach (shift the yield curve by 1bp and recompute the price). This was documented for the Cox-Ross-Rubistein model by Pelsser and Vorst ("The binomial model and the greeks", Journal of Derivatives, 44-49, spring 1994) and for the Hull-White model in Semi-explicit Delta and Gamma for European swaptions in Hull- White one factor model.If you are interested by Bermudan swaption in the gaussian HJM framework (extended Vasicek or Hull-White in particular) you can also have a look at the sequels of the second article that are devoted to Canary and Bermudan swaption using semi-analytical / numeric integration techniques (not tree)A semi-analytical approach to Canary swaptions in HJM one-factor modelBermudan swaptions in Hull-White one-factor model: analytical and numerical approachesBy avoiding the tree, the greeks are more stable.