PROC Surveyreg, PROC GLM, or PROC Mixed?

[therock]

Hi Everyone,

I am trying to replicate the following statement:

“All specifications include year and industry fixed effects and standard errors are heteroskedasticity robust, clustered at the firm level.” (Costello, 2013). My problem is I am getting three difference coefficients on the model. I need to which option is correct.

I have three option:

1) proc surveyreg data=sample ;
cluster firm;
class industry year;
model DV = IND1*IND2 IND3 IND4 / noint adjrsq solution;
run;

2) proc glm data=sample ;
absorb industry;
class year;
model DV = IND1*IND2 IND3 IND4 year / noint solution;
run;

(Could not find a way to do clustering in Proc GLM)

3) proc mixed noclprint data=sample;
class firm industry year;
model DV = IND1*IND2 IND3 IND4 industry year / noint solution;
random intercept/subject=firm;
run;

(Could not get R-square in Proc Mixed)

Any help or insight will be very much appreciated. Thank you in advance. 

[PaigeMiller]

“All specifications include year and industry fixed effects and standard errors are heteroskedasticity robust, clustered at the firm level.”

That's really not enough to go on. Perhaps if I read the entire paper, I could advise further, but I'm not going to read the entire paper.

Nevertheless, if there is clustering at the firm level, this would suggest using PROC SURVEYREG. I note that you did not put year and industry into the model — putting them in the CLASS statement but not in the MODEL statement really doesn't do anything.

[therock]

That is what almost all paper mentioned in the Finance field. 

By putting the Class variables in model statement, SAS tells me that: The X’X matrix has been found to be singular. Essentially one or more of my independent variables are the exact same. If I do not include the class variables in the model statement, it works just fine.

[PaigeMiller]

---

@therock wrote:

By putting the Class variables in model statement, SAS tells me that: The X’X matrix has been found to be singular.

---

This always happens when you have CLASS variables. It does not indicate you did anything wrong. It does not indicate a problem with the analysis, which (if everything is done correctly) is the correct analysis. Everyone who has ever used CLASS variables in a SAS model with the SOLUTION option has gotten this message.