Date:         Wed, 19 Jun 1996 23:58:37 GMT
Reply-To:     Michael Haid <tmhaid@GSBVAX.UCHICAGO.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Michael Haid <tmhaid@GSBVAX.UCHICAGO.EDU>
Organization: The University of Chicago
Subject:      HELP: Panel Data and PROC MIXED

Hi everybody,
 
I need help on PROC MIXED! The problem set-up is as follows: I have panel data
for 51 firms over 18 years). I would like to estimate the following model:
 
model y = x1 x2 x3
 
Now I would like to fit the following models:
 
1) Model 1: Simple OLS on pooled data,
2) Model 2: Model 1 + Groupwise Heteroscedasticity,
3) Model 3: Model 2 + Cross-Secional Correlation of the Error Terms,
4) Model 4: Model 3 + Autocorrelation within the Groups,
5) Model 5: Model 4 + Autocorrelation between Groups,
6) Model 6: Model 5 + allowing for firm-specific slopes for x1.
 
Model 1 is no problem. For model 2, I have the following:
 
----------------------------------------
proc mixed data=a;
    class firm year;
    model y = x1 x2 x3 / s;
    repeated / type = un(1) group=firm;
run;
----------------------------------------
 
Is this the right solution? Model 3 assumes that error terms in the same yeare
are correlated. In the variance covariance matrix which consists of 51x51
submatrices (18x18) that means the diagonals of the off-diagonal matrices are
also non-zero. How do I specify a model like this in PROC MIXED? How do I
specify Model 4-6? Model 4 assumes that the error terms of each firm are
autocorrelated. Model 5 assumes in addition that the error terms of one firm
are correlated with past error terms of other firms (serial correlation).
 
Right now, I am very desperate since I4d like to finish my dissertation within
the next two months. I appreciate any help.
 
Please send any comments/solutions to:
 
tmhaid@gsbvax.uchicago.edu
 
Thank you very much.
--Michael Haid