Posted by
Kucherenko, Sergei on
Aug 09, 2010; 11:30am
URL: http://quantlib.414.s1.nabble.com/Kooql-tp7085p7089.html
Dear Kakhkhor,
>It is surprising to see that with only 32768 path QMC results were so accurate. I was under the impression that QMC looses its efficiency at higher dimensions.
The notion that "QMC looses its efficiency at higher dimensions" is incorrect. For majority of financial problems QMC ( with Sobol' sequences ) will be much more efficient than MC ( in roughly sqrt(N) times where N is the number of sampled points/paths ) regardless of nominal dimensions because financial problems have low effective dimensions. For details see e.g.
http://www.broda.co.uk/gsa/wilmott_GSA_SK.pdf ,
http://www.broda.co.uk/gsa/qmc_finance.pdfRegards,
Sergei Kucherenko
________________________________________
From:
[hidden email] [
[hidden email]] On Behalf Of Kakhkhor Abdijalilov [
[hidden email]]
Sent: Monday, August 09, 2010 11:25 AM
To:
[hidden email]
Subject: Re: [Quantlib-users] Kooql
Dear Mark,
I did some work on parallel implementation of LS. The results will be
made public after the testing is completed (if everything works well,
hopefully).
I just read "GRAPHICAL ASIAN OPTIONS" paper and have several questions.
In section 5 where the numerical results are discussed, 32768 path
were used to price the option at each volatility level and the target
price was computed on CPU using 2^22 path. But how the control variate
was used? Was the Sobol sequence randomized somehow?
It is surprising to see that with only 32768 path QMC results were so
accurate. I was under the impression that QMC looses its efficiency at
higher dimensions. Could it be because Asian payoff becomes less
volatile as the number of averaging dates increases? Also, how CPU
performance was measured? Did it use QuantLib's Asian pricing engine?
Regards,
Kakhkhor Abdijalilov.
------------------------------------------------------------------------------
This SF.net email is sponsored by
Make an app they can't live without
Enter the BlackBerry Developer Challenge
http://p.sf.net/sfu/RIM-dev2dev_______________________________________________
QuantLib-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/quantlib-users------------------------------------------------------------------------------
This SF.net email is sponsored by
Make an app they can't live without
Enter the BlackBerry Developer Challenge
http://p.sf.net/sfu/RIM-dev2dev
_______________________________________________
QuantLib-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/quantlib-users