Hi,I am implementing a vanilla pricer for Heston model based on fft as explainedin Peter Carr "Option Valuation Using the Fast Fourier Transform", 1999.I checked that I get correct prices for calls of all strikes (I checked them against the web pricer in http://kluge.in-chemnitz.de/tools/price ... _price.php whichcomputes the analytical prices).The problem is that I have to use a value for N (in fft) at least equal to 2^14=16384in order to get a relative error with respect to analytic prices < 1%.I read in Peter Carr "Option Valuation Using the Fast Fourier Transform", 1999 thata value of N=2^10=4096 should instead be sufficient.Do you have any idea to help me?Bye

You can use the Fractional FFT which is more flexible than the straight FFT.Hope that this helps,K.

Carr uses a spacing eta of 0.25 with 4096 points. Which spacing are you using?The most complete article on FFT Error Control is the one of Lee: Option pricing by Transforms Methods.Actually I don't use Lee results but I just apply what Carr suggests. I have some problems for long dated options with "high vol of variance" and low "speed of mean reversion" (i.e. when Heston model becomes unstable for most of the numerical methods).RegardsStefano

The problem for long dated options and high vol of variance arises from jumps in the complex logarithm. There was an article in the September issue of Wilmott dealing with this problem "Not-so-complex logarithms in the Heston model" by Kahl and Jäckel. This might solve your problem.

I am giving a look to that paper. I have always ignored the problem of the complex log ... ;-)By the way, which discretization scheme do you use when you do a monte carlo for the Heston model ?