Hi All,I foud it from Jaekel P- Weiner increments backed out path construction with Brownian bridge(BB) method driven by ?Sobol? sequence shows better convergence properties for the numerical integration of differential equation using Monte carlo simulations. Can we implemment BB with pseudo random numbers instead of sobol sequency, in this case is there is any significant improvement in covergence over standard MC simulations(with out BB).Regards,

Hi outrun,With reference to your secod point( in previous post). Exatly I am looking for possibility of mix single and multistep path that are mutually consistent. I am happy/excited to know that it is possible and I will explore the same for the portfolio VaR(n day) computaion( with asian options).Assume I have to two methods in MC simulations1. Multi step scenatio for all the products2. Mix of single amd multi step paths- (possible with BB)For both the methods I need pseudo random numbers.With my initial researh work I found that slightly higher number of computaions( additions, multiplications etc..) are required for the second method than the first one.This may be due to due to constrution of weiner process using BB and calculation of weiner increments driven by BB, where as in first we can use directly pseudo random numbers as weiner process increments.As I am using these concepts to compute the Portfolio VaR, which is the tail phenomena. Is method 2 is really make any difference to the convergence VaR numbers as compared to the first method. Kindly give your valuable inputs,Regards,

I think you're using a sledgehammer to crack a nut.I'd simulate the Asian seperately, and add the two results.Sure, you're losing the correlation benefit, but so what, it's VaR and wrong anyway.Plus you can claim you're being prudent by overstating the number.Bear in mind too that you can't use Cholesky with Sobol, and pseudo randoms with BB usually defeats the purpose of bridging.I'd recommend a read of Glasserman.

LOL, that's the kind of guy I am!Now you mention geometric, you're insane if you don't use control variates for Asians, they'll converge like gangbusters as P Hagan used to say.

Hi TinMan,outrun, Tanx for your suggestions, I will definitely look in to Glasserman and also geomentric based asian option pricing.TinMan, I will considered your inputs as a prgamatic method. But I am developing this for the product, with your inputs, I will give user two option to select pragmatic method or multi step Monte Carlo( using Cholesky->pseudo random numbert / PCA->Sobol->BB etc...). As part of pragamatic method can you suggest any other appraximations.Regards,Jeevan

QuoteOriginally posted by: TinManBear in mind too that you can't use Cholesky with SobolWhy is that ?