[Scilab-users] How to start with several variables in AUTOMATIC function of GROCER

[Eric Dubois]

Dear Samuel.

The error can be explained easily: one of the main principles of the
program automatic is to find the best model that passes specification
tests, that is tests that the residuals can be considered as white noise
and that the model is stable (through standard Chow tests). And the Chow
test estimates a number of coefficients equal to twice the number of
exogenous varaibles (that is 12 coefficients with only 9 points!)

As explained in chapter 15 of  the manual (page 8-9), available on my web
site (http://dubois.ensae.net/grocer.html), you can overcome this
difficulty by reducing the set of specification tests. For instance:
--> automatic(nc, x,'test=doornhans')
works.

This is at the expense of the philisophy of the program...

Éric.

2013/2/19 Samuel Enibe <samuel.enibe at unn.edu.ng>

> Dear sir,
>
> I would like to employ the AUTOMATIC function in GROCER toolbox to perform
> automatic regression starting with a rather LARGE set of variables and
> ending with the few that are most significant. However, I have noticed the
> following problems using the example attached:
>
> 1. Programme works with only 3 initial variables specified (see line 30 in
> attached file), which works, but line 31 with only 6 starting variables
> complains of TOO MANY EXOGENOUS VARIABLES
> 2. When it runs with 3 starting variables, the R2 value is not specified.
> How can I get this calculated and used.
> 3. The help files included with the AUTOMATIC package does not seem to
> answer these questions.
>
> ANY ASSISTANCE WILL BE GREATLY APPRECIATED
>
>
> --
> Samuel Ogbonna Enibe
> BEng (Nig), MSc (Reading, England), PhD (Nig)
> Professor of Mechanical Engineering
> Director, National Centre for Equipment  Maintenance & Development
> University of Nigeria, Nsukka, Nigeria
> Tel: +2348063646798
> Email: samuel.enibe at unn.edu.ng
>      enibesam at yahoo.com
> _______________________________________________
> users mailing list
> users at lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.scilab.org/pipermail/users/attachments/20130219/c8df97c0/attachment.htm>