Forecasting using SAS Forecast Server, SAS/ETS, and more

Adjusting Standard Errors in Non-linear Seemingly Unrelated Regression

Reply
Occasional Contributor
Posts: 6

Adjusting Standard Errors in Non-linear Seemingly Unrelated Regression

Hi,

I am estimating a non-linear seemingly unrelated regressing using proc model, as follows:

proc model data=temp2w;
   Y1= a0 + ((gamma))*(rho1)*X1+ (1-(theta))*(rho2)*X2;
   Y2= b0 + (1-(gamma))*(rho1)*X1+ (theta)*(rho2)*X2;
   fit  dPIFO dPIDOM/ sur;
run;
quit;

I need estimates of a0, b0, gamma, theta, rho1, and rho2. The model estimates the parameters as I would expect. Is there a way to adjust standard errors for clustered observations in SAS? My dataset consists of a panel of firms, and there is reason to suspect that observations are not independent within a firm.

SAS Employee
Posts: 25

Re: Adjusting Standard Errors in Non-linear Seemingly Unrelated Regression

Unfortunately, PROC MODEL doesn't support the estimation of parameters in a system of equations with random effects.  I'm not aware of any other SAS procedures that can address this problem either.  We'll look into adding this ability to a future version of MODEL.

Occasional Contributor
Posts: 6

Re: Adjusting Standard Errors in Non-linear Seemingly Unrelated Regression

Thanks for your reply! I don't think I need random effects, however. I want the fixed effects estimates the model produces, but I want to adjust the standard errors for possible intra-cluster correlations. For example, proc genmod allows for the "repeated' command which adjusts the standard errors for non independence within the groups. Is there any way to do this in proc model?

SAS Employee
Posts: 25

Re: Adjusting Standard Errors in Non-linear Seemingly Unrelated Regression

PROC MODEL can correct the standard errors for heteroscedasticity using the HCCME= option on the FIT statement.  This option does not allow you to specify the structure of the clustering of the data, but it may give you results closer to what you're looking for.

Ask a Question
Discussion stats
  • 3 replies
  • 775 views
  • 0 likes
  • 2 in conversation