Re: [Quantlib-dev] Monte Carlo

[Andrea <mar...@go...>]

On 20/09/10 10:36, ANIMESH SAXENA wrote:
> Hi,
>    I had been work on a monte carlo design (with Stochastic Volatility -
> Heston Model). My basic idea was to allow almost any exotic structure to
> be priced (well 90% of them) using stochastic volatility coz
> GeneralizedBlackScholes process is not practically useful. I designed
> this from a trader / quants perspective. The design is very poor - I Admit!
> 
> I have tried to generalize the heston process, and using what I call
> hooks (different exotic payoffs) the process will price them. The
> process generates the paths, (cholesky decomposition used to generate
> correlated random numbers). Halton sequence used for fast convergence.
> The hook just uses the process to get the paths generated and updates
> the payoff. So any new exotic structure added will just need a new hook. 
> As of now I have skipped the discounting portion, and is purely used by
> me for personal trading. I do the discounting manually using HJM later.

Interesting. I think I need to dive deeper into your code.

> 
> I would like some suggestions from anyone who wants to take try this.
> Maybe with lots of changes to design this methodology can be put into
> QuantLib. This will help in solving the tight coupling of engines and
> process issue.

I did try to solve a similar problem.
Decoupling engine and products and I contributed the folder "mcbasket" under ql\experimental.
I have to admit that it only works for a GeneralizedBlackScholes but it is super easy
to add new payoffs.
Basically what you call Hook, there is called PathMultiAssetOption (this deals with the dates) and
PathPayoff (the actual payoff formula).

> 
> Link is a google doc to attached code (zip file). It's in Mac (XCode)
> gui, but well can be compiled on any platform easily. (No dependencies
> on Boost etc)
> 
> Thanks for your time,

Same here

Andrea