This happens with basic structural models if some of the variances are very
close to zero. The problem is that the precision of the numerical hessian is
limited (e.g. 2 or 3 decimals). Very often this problem can be avoided if
you rescale your data (e.g. if you have data in logs, try multiplying the
logged data by 100). Also, with basic structural models you will improve the
numerical stability of the estimation if you concentrate out one of the
variance parameters. This trick was introduced by Harvey and is utilized in
the STAMP pagkage. See my example of estimating and AR(1) model in the state
space chapter of the 2nd Edition of Modeling Financial Time Series (the
example in the first edition contained a small error). Finally, you may also
get better estimates if you use the outer product gradient version of the ML
covariance.
I have started a tutorial on basic structural models on the additions page
of the MFTS 2nd Edition web page
http://faculty.washington.edu/ezivot/MFTS2ndEditionAdditions.htm
. I will be updating this with more hints on estimation sometime this month.
ez
-----Original Message-----
From: s-news-owner@lists.biostat.wustl.edu
[mailto:s-news-owner@lists.biostat.wustl.edu] On Behalf Of Carole Birrell
Sent: Sunday, August 27, 2006 8:19 PM
To: Splus Mailing list
Subject: [S] Hessian Matrix
Dear SPLUS users,
I am using the SPlus version 7.0 Finmetrics and in particular the SsfFit
function. This function has obviously changed from version 6.2 , and now
seems to check for a Hessian matrix which is not invertible.
I am simulating data from a multivariate basic structural model and then
attempting to estimate the parameters for this model.
I am now getting warnings saying that "Hessian failed to invert. in:
SsfFit(all9, Augmentdata, "EstAugSV", trace = T)".
Do I need to alter the precision somewhere, if so how do I do that?
If not, do you have any advice as to what I should do?
regards,
Carole Birrell
University of Wollongong.
--------------------------------------------------------------------
This message was distributed by s-news@lists.biostat.wustl.edu. To
unsubscribe send e-mail to s-news-request@lists.biostat.wustl.edu with the
BODY of the message: unsubscribe s-news
|