Infinite calculation while calibrating 2F model...

[Toyin Akin-3]

Hi all,

 

Using your Bermudan example I duplicated the code for the numerical calibration of your HW

model (TreeSwaption) and instead replaced the object with a G2 model. I also changed the calibration code (in the example file) so that the max iteration was 100 and the accuracy was 1e2

 

After 30 minutes the G2 model is still calibrating.

 

The CIR model does calibrate, but only under one of the method types (numerical or analytical, I forget which!!). The error message for the method which does not calibrate is a bracketing root problem.

 

Does this imply that the work on the G2 model is as yet unfinished?

 

Thanks in advance,

Toy.

[Sadruddin Rejeb-3]

Hi Toyin,

> The CIR model does calibrate, but only under one of the method types
> (numerical or analytical, I forget which!!). The error message for the
> method which does not calibrate is a bracketing root problem.
The problem should lie with the tree pricer. I've noticed some instability
with trees generated for the CIR model... I'll investigate this after 0.3.0,
since the goal for this release is just to have BK and HW working (most
standard models).

> After 30 minutes the G2 model is still calibrating.
If you calibrate on swaptions, this is normal. The only pricer that works
with the G2 model is the cap analytical pricer. Bidimensional trees are
still experimental.

> Does this imply that the work on the G2 model is as yet unfinished?
Oh yes. We need an efficient bidimensional tree (which I'm working on) and
I'm investigating an analytical swaption pricer (cf. the book of
Damiano/Brigo). So let's be patient...

Regards,
Sad

--
GMX - Die Kommunikationsplattform im Internet.
http://www.gmx.net

[Toyin Akin-3]

Hi Sadruddin.

Two things in which I would like some clarification (for my own mind!!).

1) The BlackScholesProcess class's drift function returns -r*x. Shouldn't
this be just r?

2) For your Fitting parameter (CIR class), the formula for the value()
function differs greatly from that of Damiano/Brigo-2001  pg93 (3.77).
However I just observed this by eye and thus you may have simplified the
formula..

Regards,
Toy.

[Sadruddin Rejeb-3]

Hi,

> 1) The BlackScholesProcess class's drift function returns -r*x. Shouldn't
> this be just r?
Copy and paste error... shame on me.
Well, actually it is (r - \sigma^2/2), because I had in mind the process of
ln(S) under risk-neutral measure. It is fixed now.
 
> 2) For your Fitting parameter (CIR class), the formula for the value()
> function differs greatly from that of Damiano/Brigo-2001  pg93 (3.77).
> However I just observed this by eye and thus you may have simplified the
> formula..
I ran some tests a while ago, and I'm quite sure it showed that the formula
is correct. If you can point out a specific error, please tell me.
(If I remember, the formula in the Damiano-Brigo book is incorrect, there is
a
2 missing).

Sad

--
GMX - Die Kommunikationsplattform im Internet.
http://www.gmx.net

[Toyin Akin-3]

Hi Sad,

Yep, I too noticed the addition of the 2 factor.
Well that's cleared that up.

Thanks in advance,
Toy.

----- Original Message -----
From: "Sadruddin Rejeb" <[hidden email]>
To: <[hidden email]>
Cc: <[hidden email]>
Sent: Monday, April 15, 2002 11:19 AM
Subject: Re: [Quantlib-users] Re: Infinite calculation while calibrating 2F
model...


> Hi,
>
> > 1) The BlackScholesProcess class's drift function returns -r*x.
Shouldn't
> > this be just r?
> Copy and paste error... shame on me.
> Well, actually it is (r - \sigma^2/2), because I had in mind the process
of
> ln(S) under risk-neutral measure. It is fixed now.
>
> > 2) For your Fitting parameter (CIR class), the formula for the value()
> > function differs greatly from that of Damiano/Brigo-2001  pg93 (3.77).
> > However I just observed this by eye and thus you may have simplified the
> > formula..
> I ran some tests a while ago, and I'm quite sure it showed that the
formula
> is correct. If you can point out a specific error, please tell me.
> (If I remember, the formula in the Damiano-Brigo book is incorrect, there
is
> a
> 2 missing).
>
> Sad
>
> --
> GMX - Die Kommunikationsplattform im Internet.
> http://www.gmx.net
>