Robust standard errors in HPMIXED

[Cuneyt]

Dear SAS Users,

I have a large panel data set with ~1 million firms observed over 7 years. I am using fixed effect for firms, but random effects for years as I have time-invariant independent variables. Given the size of the dataset, I am using PROC HPMIXED. My code is below. How can I obtain robust standard errors for this model? Any help will be greatly appreciated.

proc hpmixed data=mydata;
	class firm year;
	model y = x firm / solution;
	random intercept / subject=year;
run;

[jiltao]

I am not aware of a way to obtain robust standard errors in PROC HPMIXED. For data this size, the p-value is very likely to be significant, regardless of whether the standard errors are robust or not. Many analysts are not too concerned about p-values in situations like this.

Thanks,

Jill

[sbxkoenk]

• Don't you have SAS/ETS with PROC PANEL?
• Or SAS Econometrics in Viya with PROC CPANEL (CPANEL procedure is the CAS-enabled PANEL procedure)?

There's a ROBUST option to compute robust variances

There's a HAC option to specify a heteroscedasticity- and autocorrelation-consistent (HAC) covariance 

There's a HCCME=0 | 1 | 2 | 3 to specify a heteroscedasticity-corrected covariance matrix estimator (HCCME) 

There's a CLUSTER option that corrects covariance for intra-cluster correlation 

[EDIT] HPMIXED is multi-threaded indeed, but CPANEL is multi-node and multi-threaded on each node. That's because CPANEL leverages the CAS-engine in Viya (this is distributed computing on in-memory data).

BR, Koen