MF1F NTL Support
Locked
 
|
Hi,
I added high precision floating point support to the markov functional model making use of NTL (and supplementary boost functions). This allows for a better numeraire fit to long term (like say 50y) constant maturity calibration sets. NTL is only used for some intermediate results, the interface does not change. While there are other, faster yet dirty ways to stabilize the calibration (like the AdjustYts option) this is probably a good way to produce benchmark results and check convergence in the numerical parameters. I updated the code on my github as well as the doc on ssrn giving an example. More generally, is high precision arithmetic a topic anyone came across in some context he or she wants to share ? Peter ------------------------------------------------------------------------------ Master Visual Studio, SharePoint, SQL, ASP.NET, C# 2012, HTML5, CSS, MVC, Windows 8 Apps, JavaScript and much more. Keep your skills current with LearnDevNow - 3,200 step-by-step video tutorials by Microsoft MVPs and experts. ON SALE this month only -- learn more at: http://p.sf.net/sfu/learnmore_122712 _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
|
|
Hi Peter,
I've used gcc's type __float128 to calculate high precision reference prices for some stoch vol models but I'd to change multiple parts of QuantLib to get it working. Just changing the typedef for Real in types.hpp was not enough.
regards Klaus
On Thursday, December 27, 2012 02:26:55 PM Peter Caspers wrote: > Hi, > > I added high precision floating point support to the markov functional > model making use of NTL (and supplementary boost functions). This allows > for a better numeraire fit to long term (like say 50y) constant maturity > calibration sets. NTL is only used for some intermediate results, the > interface does not change. While there are other, faster yet dirty ways > to stabilize the calibration (like the AdjustYts option) this is > probably a good way to produce benchmark results and check convergence > in the numerical parameters. I updated the code on my github as well as > the doc on ssrn giving an example. > > More generally, is high precision arithmetic a topic anyone came across > in some context he or she wants to share ? > > Peter > > ---------------------------------------------------------------------------- > -- Master Visual Studio, SharePoint, SQL, ASP.NET, C# 2012, HTML5, CSS, MVC, > Windows 8 Apps, JavaScript and much more. Keep your skills current with > LearnDevNow - 3,200 step-by-step video tutorials by Microsoft MVPs and > experts. ON SALE this month only -- learn more at: > http://p.sf.net/sfu/learnmore_122712 > _______________________________________________ > QuantLib-dev mailing list > https://lists.sourceforge.net/lists/listinfo/quantlib-dev ------------------------------------------------------------------------------ Master Visual Studio, SharePoint, SQL, ASP.NET, C# 2012, HTML5, CSS, MVC, Windows 8 Apps, JavaScript and much more. Keep your skills current with LearnDevNow - 3,200 step-by-step video tutorials by Microsoft MVPs and experts. ON SALE this month only -- learn more at: http://p.sf.net/sfu/learnmore_122712 _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
|
|
In reply to this post by Peter Caspers-4
thanks, Klaus. I tried to make the 80 bit double run on msvc, that
wasn't fun (and did not work / I gave up at some point). gcc quad
precision sounds good, is it fast ? NTL seems attractive in that it
runs on many platforms without customization and with (to be tested)
identical results, and you can push the precision even higher
(actually I can run a (rather extreme) calibration test case which
fails with 113 bit mantissa, but runs fine with 150 bit). MPFR seems
to be another possible solution, also with support in boost as a
user defined float type, however I did not try.
Peter
Am 27.12.2012 17:57, schrieb Klaus Spanderen:
------------------------------------------------------------------------------ Master HTML5, CSS3, ASP.NET, MVC, AJAX, Knockout.js, Web API and much more. Get web development skills now with LearnDevNow - 350+ hours of step-by-step video tutorials by Microsoft MVPs and experts. SALE $99.99 this month only -- learn more at: http://p.sf.net/sfu/learnmore_122812 _______________________________________________ QuantLib-dev mailing list [hidden email] https://lists.sourceforge.net/lists/listinfo/quantlib-dev |
| Free forum by Nabble | Edit this page |