Re: [R] Need help to estimate the Coef matrices in mAr

From: Arun Kumar Saha <>
Date: Wed 20 Sep 2006 - 10:11:36 GMT

Dear Spencer,

Thank you very much for your attention on my problem. According to your advice I did some home work on this problem, but unfortunately I could not solve my problem.

Suppose I have a dataset of length 300 with 2 variables. And I want to fit a VAR model on this of order 2.

I went through the function mAr.est and got understand that, here 'K' is a matrix with (300-2) rows and 7 columns. the first col. consists only 1, next two columns consist of lagged values of two variables with lag-length 2, next two col. consist of lagged value with lag length-1, and next two cols are for lag-length-0.

Next, they add additional a 7-7 matrix to K. For this matrix diagonal elements are the square root of sum of square of elements of K (col. wise) and rest of the elements are 0.

I feel that this matrix, that is added to K, is the key matrix for any type of modification that you prescribed. Therefore for experimental purpose I put NA against one of its off-diagonal elements. But I got error.

However I cannot understand why they put such figures for diagonal and off-diagonal elements of that matrix.

Can you suggest me any solution more specifically?

Thanks and regards,


On 9/4/06, Spencer Graves <> wrote:
> Have you tried 'RSiteSearch("multivariate autoregression",
> "functions")'? This produced 14 hits for me just now, the first of
> which mentions a package 'MSBVAR'. Have you looked at that?
> If that failed, I don't think it would be too hard to modify
> 'mAr.est' to do what you want. If it were my problem, I might a local
> copy of the function, then add an argument accepting a 2 or
> 3-dimensional array with numbers for AR coefficients to be fixed and NAs
> for the coefficients. Then I'd use 'debug' to walk through the function
> line by line until I figured out how to modify the function to do what I
> wanted. I haven't checked all the details, so I don't know for sure if
> this would work, but the function contains a line 'R = qr.R(qr((rbind(K,
> diag(scale)))), complete = TRUE)' which I would start by decomposing,
> possibly starting as follows:
> Z <- rbind(K, diag(scale)
> I'd figure out how the different columns of Z relate to my problem, then
> modify it appropriately to get what I wanted.
> Another alternative would be to program it from scratch using
> something like 'optim' to minimize the sum of squares of residuals over
> the free parameters in my AR matrices. I'm confident I could make this
> work, even if the I somehow could not get it with either of the other two.
> There may be something else better, e.g., a Kalman filter
> representation, but I can't think how to do that off the top if my head.
> Hope this helps.
> Spencer Graves
> Arun Kumar Saha wrote:
> > Dear R users,
> >
> > I am using mAr package to fit a Vector autoregressive model to my data.
> But
> > here I want to put some predetermined values for some elements in
> > coefficient matrix that mAr.est going to estimate. For example if p=3
> then I
> > want to put A3[1,3] = 0 and keep rest of the elements of coefficient
> > matrices to be determined by mAr.est.
> >
> > Can anyone please tell me how can I do that?
> >
> > Sincerely yours,
> > Arun
> >
> > [[alternative HTML version deleted]]
> >
> > ______________________________________________
> > mailing list
> >
> > PLEASE do read the posting guide
> > and provide commented, minimal, self-contained, reproducible code.
> >

Arun Kumar Saha, M.Sc.[C.U.]
S T A T I S T I C I A N    [Analyst]
Transgraph Consulting []
Hyderabad, INDIA
Contact #  Home: (91-033) 25558038
                Office: (91-040) 30685012 Ext. 17
                  FAX: (91-040) 55755003
               Mobile: 919989122010

	[[alternative HTML version deleted]]

______________________________________________ mailing list
PLEASE do read the posting guide
and provide commented, minimal, self-contained, reproducible code.
Received on Wed Sep 20 20:27:31 2006

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.1.8, at Sat 30 Sep 2006 - 17:30:08 GMT.

Mailing list information is available at Please read the posting guide before posting to the list.