the problem of using the out-of-the-monety options to calibrate by FFT

[liuwenbo]

HI,
i want to calibrate the model of BNS,VG,Heston model to the market options data of HongKong .Firstly ,i have tried to calibrate the heston model. but there was a problem.
I find a matlab code from http://www.theponytail.net/CCFEA/lect02/lect02ivsurfcal.m because he use the call options data to calibrate. But i want to use the data of out-of-the-money which recommed by Carr(1999)"option valuation using the fast fourier transform". i modified the matlab code.But I was confronted with a prblem.I find the computed options price by FFT in the moneyness(ln(K/S)) interval [-0.4 0,2] is very sensitive to the parameter alpha.
firstly,i choose a proper alpha for short maturity options,the price in that intervals may be is reasonal, but the same alpha will induce weird price ,such as 1e+30, to long maturity options. so the optimization work can not work well if i calibarate one day data,because there were a lot of short an long term options in a certain day,and alpha should be constant.
secondly ,after choosing a proper data for short maturity options, if i calibarate short options only, the output parameters from matlab's function lsqnonlin is the same as the input initial parameters.
can you help me?
zip file contains: (1) one day's stock index options data of HongKong stock market
(2) my modified matlab code for out-of-the-money options
(3)two refered workingpapers
thanks very much!!!

 

[Pulco]

How do you choose your value for alpha ? This is a crucial point if you want to obtain accurate prices.