Prev: MACROS IN SAS.
Next: Merge - Hash
From: Lambros_17 on 22 Apr 2010 05:30
Hi all

I have an unbalanced cross section (firms) time series (years)
dataset. I would like to treat both firm and year effects as fixed and
allow for a time-wise autoregressive and cross-sectional
heteroscedastic covariance. In particular, I would like to allow for
the following error covariance structure: AR(1) within firms and
heteroscedasticity across firms.

The code I came up with is the following:

PROC MIXED DATA=mydata;
CLASS firmid year;
MODEL y = firmid year x1 x2 x3 /SOLUTION;
REPEATED /SUBJECT=firmid TYPE=AR(1) GROUP=firmid;
RUN;

The problem is that SAS cannot estimate this. The answer I get is
"Convergence status: Warning, did not converge". However, when I try
to estimate the model by employing only the fixed effects (delete the
REPEATED line) or only the aforementioned error covariance structure
(delete the CLASS line and firmid year from the MODEL line) everything
is OK.

How can I estimate the model (preferably in PROC MIXED) by employing
simultaneously the fixed effects and the aforementioned error
covariance structure? Is something wrong with the code I wrote?

Any help is greatly appreciated, as I have a deadline approaching…

Thanks in advance,
Costas
From: BruceBrad on 23 Apr 2010 00:45
Are you actually trying to fit both a fixed and random time effect? Is
such a model identified? The documentation on the new Panel procedure
(or the procedure itself) might be useful. Have a look at the sections
on the Parks method and the Dynamic Panel Estimator.

From: Ryan on 24 Apr 2010 08:18
On Apr 22, 5:30 am, Lambros_17 <clamb...(a)gmail.com> wrote:
> Hi all
>
> I have an unbalanced cross section (firms) time series (years)
> dataset. I would like to treat both firm and year effects as fixed and
> allow for a time-wise autoregressive and cross-sectional
> heteroscedastic covariance. In particular, I would like to allow for
> the following error covariance structure: AR(1) within firms and
> heteroscedasticity across firms.
>
> The code I came up with is the following:
>
> PROC MIXED DATA=mydata;
>         CLASS firmid year;
>         MODEL y = firmid year x1 x2 x3 /SOLUTION;
>         REPEATED /SUBJECT=firmid TYPE=AR(1) GROUP=firmid;
> RUN;
>
> The problem is that SAS cannot estimate this. The answer I get is
> "Convergence status: Warning, did not converge". However, when I try
> to estimate the model by employing only the fixed effects (delete the
> REPEATED line) or only the aforementioned error covariance structure
> (delete the CLASS line and firmid year from the MODEL line) everything
> is OK.
>
> How can I estimate the model (preferably in PROC MIXED) by employing
> simultaneously the fixed effects and the aforementioned error
> covariance structure? Is something wrong with the code I wrote?
>
> Any help is greatly appreciated, as I have a deadline approaching…
>
> Thanks in advance,
> Costas

Costas,

First thing that sticks out to me is your REPEATED statement--Why
haven't you included your interval variable, "Year"?:

"REPEATED Year / ..."

I don't know enough information to say what might be causing this
problem. Are you potentially overspecifying your model? For instance,
are you sure you need to allow for heterogenous variances?

HTH,

Ryan
 | 
Pages: 1
Prev: MACROS IN SAS.
Next: Merge - Hash