|
Hello,
I've posted a couple similar-sounding questions and not received a reply,
so I'm sorry if this is starting to get irritating... I just thought I'd
try once more.
I am trying to run a 3-equation SUR (cost function and 2 share equations)
on panel data (50 firms, 180 time periods, though the data is
unbalanced). I am wondering what my estimation options are in SAS.
First of all, my data (say, data=test) is currently arranged as follows
(just an example):
firm time lnCost lnOutput lnPrice1 lnPrice2 ShareInput1 ShareInput2...
1 1 1234 12 234 345 345 454
1 2 1457 13 323 235 454 235
1 .....
2 1 3456 20 234 456 124 347
2 2 3454 22 346 124 136 359
.....
This us what I've thought of/tried so far:
1) Proc Model with a "by" statement: (pseudo-code)
Proc Model data= test;
Variables blah blah blah;
parms blah blah blah;
restrict asNecessary;
lnCost= alpha + beta1*lnOutput + beta2*lnPrice1 + beta3*lnPrice2+
beta4*lnOutput*lnPrice1 +beta5*lnOutput*lnPrice1+ beta6*lnPrice1*lnPrice2;
Share1= beta2+ beta4*lnOutput + beta6*lnPrice2;
Share2= beta3+ beta5*lnOutput + beta6*lnPrice1;
fit lnCost Share1 Share2/ITSUR;
by Firm;
run;
I've actually tried this, and it gives separate intercept and slope
parameter estimates for each firm, which i believe means that i've got 50
firm-specific models, rather than 1 model for the industry. i would like
to use the slope parameter estimates to calculate elasticities for the
industry, though, and i'm not sure if i can just average these slopes, or
if that is not quite right. if i were able to run a normal fixed effects
model (the way i understand them), i could get differing intercepts
(especially if i did not require the intercepts of the share equations to
equal a paramter from the cost equation) across firms, but a single
estimate for the slope parameters that would be valid industry-wide.
2) Use proc model, but transform the data so that it is in deviations
from the mean. I.e. calculate for each variable and for each firm the
value of (variable value in time t for firm i -minus- mean value across t
for firm i of the variable). this would (I think?) create a within
estimation, but i'm not quite sure how my data should then be arranged for
SAS to acknowledge multiple firms, nor am I ceratin of how exactly to
program it in Proc Model. Do you think it'd look pretty much the same as
above, just without the "by" statement, and using data suitably
transformed?
And finally, if you have any thoughts about my data layout and whether it
needs to be transposed or stacked differently to run these models, I'd
love to hear them!
Thanks so much for your time and help,
Lindsay
| 
