Re: FdDividendAmericanOption
Posted by Joseph Wang on Dec 30, 2004; 12:07am
URL: http://quantlib.414.s1.nabble.com/FdDividendAmericanOption-tp3453p3459.html
More data:
* Bermudan options also seems to fail when you run it with small grids.
* The failure appears independent of dividends (i.e. it still fails when
you get rid of dividends).
* The failure appears to be independent of finite differencing models
* The numerical instability doesn't seem to happen when you have values
that are very much
in the money or out of the money
* As I mentioned before, the code fails when interest rates are being
set to zero which suggests
that the problem involves calculating the second derivative
The one thing that seems to pop out is the BSM operator which takes as
input the log of the gridSpacing_. Knowing nothing about the code, it would
seem to me that the BSMOperator is expecting a linear grid, and putting in
a logarithmic grid would appear to do bad things. It looks like I'm
going to
have to crack open Hull and White and check the derivation from first
principles.
Can anyone who knows more about the code, let me know if I'm on the right
track? This does seem to be a pretty major bug, unless I'm missing
something.