Seemingly unrelated regression & within transformation

[Guest]

Hello,

I am currently estimating labor demand elasticities by means of a panel dataset on firms. In this context, I conduct a seemingly unrelated regression (command: sureg, version: 14) for my system of three equations. In order to account for firm fixed effects in my analysis, I want to run SUR on within-transformed data. As I have not found a possibility to combine SUR and within transformation by an explicit command, I decided to first within-transform my data manually and then run the sureg command. I think my procedure is correct if the data do not contain any missing values which render corresponding observations useless for the estimation. But my data contain many missing values. Depending on whether I delete missing observations before or after the within-transformation, the sets of within-transformed data differ since the firm-specific means change. I checked both possibilities and the results differ greatly. I think the procedure should be consistent with a common FE estimation using xtreg ... ,fe but I am not sure whether Stata here first within-transforms the data or first deletes observations with missing values. So my questions is should I first delete missing observations and then subtract the firm-specific mean or should I first subtract the firm-specific mean and then exclude missing observations from the SUR estimation.

It would be great if you could help me. Thank you in advance!

[daniel klein]

If you are going for listwise deletion of missing values, you need to restrict the sample first, then do all calculations based on the complete matrix, i.e. the dataset without any missing values. Moreover the estimation of the n firm specific means should be taken into account when determining the degrees of freedom. I am not sure whether this makes a difference in the case of sureg, since it uses the standard normal distribution for significance tests of the model parameters.

Best
Daniel