Proc Panel with regression weight

[ibsulkim]

Hello,

I have a panel dataset with a number of firms over different dates. Assume the left hand side variable in the panel regression is called y. I am trying to run a panel regression using the inverse of the std.dev(y) as regression weight (and include time and firm fixed effects). Could anyone help how I can implement this in SAS?

I have been using proc panel to include the time and firm fixed effects as below, and I was hoping to add "weight" option in the code (similar to what we do in proc reg), but it seems there is no such option available for proc panel.

proc panel data=data PRINTFIXED;
id firm_id date;

model y   =  x1 x2 x3   / fixtwo ;

output out=est residual=residual predicted=predicted;
run;

Any help would be greatly appreciated.

[Ksharp]

Better post it at Forecasting forum too, since proc panel is under SAS/ETS .

[SteveDenham]

This is just a wild-eyed guess, but you might look into using the INSTRUMENTS statement, with 1/sd defined in the dataset as a variable that gets included via the INSTRUMENTS statement.  Best bet would be to follow @Ksharp 's advice and post in the Forecasting community.

SteveDenham

[STAT_Kathleen · Accepted solution]

PROC PANEL does not support WEIGHT statement as PROC REG does and there is no way to weight observations in estimation in PROC PANEL. However, if your concern is heteroskedasticity and you wanted to correct for heteroskedasticity, then there are alternative options you could consider other than to apply weights. PROC PANEL supports HCCME = option on the MODEL statement which allows you to obtain heteroskedasticity corrected covariance matrix of parameters. This option is discussed in the following section of documentation:

http://go.documentation.sas.com/?cdcId=pgmsascdc&cdcVersion=9.4_3.4&docsetId=etsug&docsetTarget=etsu...

and details of HCCME:

http://go.documentation.sas.com/?cdcId=pgmsascdc&cdcVersion=9.4_3.4&docsetId=etsug&docsetTarget=etsu...

In addition, SAS/STAT procedure PROC MIXED can also be used to fit some panel data models and PROC MIXED procedure does support WEIGHT statement similar to that in PROC REG. The following usage note provides some example codes to fit fixed and random effects models in PROC MIXED and PROC TSCSREG/PANEL:

http://support.sas.com/kb/22/851.html

I hope the above information is helpful.

Kathleen

[sbxkoenk]

Hello,

To build upon the reply of @STAT_Kathleen 

... if your concern is heteroskedasticity and you wanted to correct for heteroskedasticity, ...

I have just come across this interesting (recent) PROC PANEL usage note:

Usage Note 67322: Heteroscedasticity and cluster correction of standard errors using the PANEL procedure

67322 - Heteroscedasticity and cluster correction of standard errors using the PANEL procedure (sas....

Cheers,

Koen

[ibsulkim]

Thank you very much Kathleen for prompt help and for sharing the links - they are all very helpful.
Proc mixed seems a solution in my case, and thanks for the link that directly compares between proc panel and proc mixed.

All the best.