Serving the Quantitative Finance Community

 
User avatar
fearless
Topic Author
Posts: 0
Joined: October 9th, 2003, 6:45 pm

Monte Carlo Variance Reduction Comparison

August 25th, 2004, 12:38 pm

Aside from a few variance reduction techiques that are most suitable for some securities, how would you pick among antithetic, stratified and quasi-random sequence techniques? You can't pick the one with the smallest error term since the standard deviation of the results is a function of the paths that are randomly selected. I have read somewhere that if you pick the right technique the most complex problems can be simulated by 200 runs. Is there any truth to this?
 
User avatar
quantie
Posts: 20
Joined: October 18th, 2001, 8:47 am

Monte Carlo Variance Reduction Comparison

August 25th, 2004, 7:36 pm

QuoteOriginally posted by: fearlessAside from a few variance reduction techiques that are most suitable for some securities, how would you pick among antithetic, stratified and quasi-random sequence techniques? You can't pick the one with the smallest error term since the standard deviation of the results is a function of the paths that are randomly selected. I have read somewhere that if you pick the right technique the most complex problems can be simulated by 200 runs. Is there any truth to this?There are cases where one of (av,ss,qrs) will deliver more bang than the others and there are places where using something mayn't necessarily improve your results. (decrease the variance of your estimate)I don't know if your statement on 200 replications is true but I don't quite see how that could be the case. AV is inexpensive since you really aren't generating numbers all over again so its cheap to do it, its always nice to use, except when the payoff function is not monotonic then AV can actually increase the variance (certain kind of barriers)
 
User avatar
pj
Posts: 11
Joined: September 26th, 2001, 3:31 pm

Monte Carlo Variance Reduction Comparison

October 23rd, 2004, 12:59 am

Realistically? 200 runs is a myth unless you have an exceptionally good control variate.As for number generator choices: your mileage might vary but I never use antithetic sampling, and rarely use anything other than Sobol' numbers.Best regards,pj
 
User avatar
fearless
Topic Author
Posts: 0
Joined: October 9th, 2003, 6:45 pm

Monte Carlo Variance Reduction Comparison

October 24th, 2004, 4:01 pm

Quote...but I never use antithetic sampling, and rarely use anything other than Sobol' numbers Is there a reason behind your choice (it didn't work for you) or just never had the opportunity? Also, my understanding is that Sobol numbers are a bit complicated. Do you agree?