Seemingly unrelated regression (SUR)

[Tom Jay]

Dear Stata community,

I have two questions.

1. Are there alternatives to seemignly unrelated regressions (SUR) that are more "novel", i.e. is SUR outdated? I am asking this because the theoretical as well as the applied literature I can find on SUR seems to be rather old. I am thinking possibly along the lines of GSEM, cluster analysis,...

2. I plan to use SUR with unbalanced equations, on which I found McDowell's 2004 article in the Stata journal. Is there a later version on this or did development on SUR with unbalanced equations stop in 2004?

Do not get me wrong. I have no issue with using a "classic". I just want to make sure I do not overlook any more robust techniques.

Thanks!

Tom

[Joro Kolev]

Mechanically -sem- supersedes -sureg- because it can do everything sureg does and some more.

I do not think that -sem- can deal with unbalanced equations, so you are back to McDowell's work, which is brilliant work by the way and should have gotten a lot more interest than it got.

The literature on unbalanced Seemingly Unrelated Regression is a bit thin, you can check also this key article:
Schmidt, P. (1977). Estimation of seemingly unrelated regressions with unequal numbers of observations. Journal of Econometrics, 5(3), 365-377.

[Tom Jay]

Thank you, Joro!

And when I want to include categorical variables I arrive at GSEM?

Addtionally, can I use GSEM when some of my dependent variables have heteroskedasticity?

Thanks again,

Tom

[Joro Kolev]

Yes Tom, when you have a nonlinear model, you need -gsem-, and yes, it can calculate robust to heteroskedasticity variances with -gsem-.

But it cannot deal with unbalanced equations, so if you want to take this issue seriously, you need to do what McDowell did and use -xtgee-, which is also suitable for some nonlinear models.

Thank you, Joro!

And when I want to include categorical variables I arrive at GSEM?

Addtionally, can I use GSEM when some of my dependent variables have heteroskedasticity?

Thanks again,

Tom