Monte Carlo VaR Error of estimation

Swing · February 23rd, 2014, 4:09 pm

Hello,OK this might be a quite simple question for you...I'm performing a monte carlo to calculate value at risk (with a 3 dimension risk factor)Now, I would like to calculate the error of the estimation of the VaR with respect to the number of simulations (drawing a graph of estimation error wrt N of simulation)Do know how I can do that please? the formula for the error on the VaR ?I'd very much appreciate your contributions..Many Thanks.

chewwy · February 24th, 2014, 9:06 am

you could consider a binomial distribution to you a confidence interval - note that for the x^th percentile, there's an x% change that each sim falls above/below it...

Swing · February 24th, 2014, 9:46 am

Hi,Could you please explain more your idea?Thanks

chewwy · February 24th, 2014, 1:06 pm

Sorry, I suppose I meant beta not binomial. See here for more info - http://www.vosesoftware.com/ModelRiskHe ... to_run.htm

Swing · February 24th, 2014, 1:11 pm

I have a pretty akward result, I'm getting different levels of VaR with Sobol sequences (-220kUSD) and with pseudo-random numbers (-310KUSD)Do you please see any possible explanation for this..Thanks.

Swing · February 24th, 2014, 1:24 pm

Thanks Chewwy, I'll check this out !

Swing · February 24th, 2014, 3:21 pm

Any help about the difference between the results with Sobol sequences and Pseud-randoms ?

quartz · February 24th, 2014, 6:19 pm

QuoteOriginally posted by: SwingAny help about the difference between the results with Sobol sequences and Pseud-randoms ?How do you check Sobol' for convergence? How many samples? Using any library/tool? Any randomisation? Sobol' can be tricky... more often than not.

Swing · February 24th, 2014, 6:34 pm

Thanks quartzThere was an error in the generation of the sobol sequences, my bad..