Hello everybody.
Generally, this class is ok, so you could use it for instance in FD engines
or trees. But you should keep in mind next hints:
- your dt must be the same as time step in lattice method, otherwise the
result will be unreliable, cause of difference in accuracy (check it among
plain vanilas)
- the better choise is to use Bicubic interpolation of implied vol surface
- if your interpolation method is smooth the dy sensitivity won't be
generally high, but i guess it's better to use lattice spot step for dy.
Nevertheless, there are still some problems:
1) The first one is speed. The derivative calculation is performed by FD
valuation, which is too slowly.
Despite FD i plan to use first and second derivative of bicubic
interpolation.
2) The second is tails. It's necessary to use extrapolation to constant
values, which still save variance
increasing. Have anybody worked with this problem and could give some
advices?
Regards.
--
View this message in context: http://www.nabble.com/LocalVolSurface-class-tp20896055p20896055.html
Sent from the quantlib-dev mailing list archive at Nabble.com.
|
