Seemingly Unrelated Regression Easy but Urgent

[David Benkert]

Hi Guys,

I have a paper to submit on Monday and the advisor had recommended I check the robustness of my time series by running the SUR.
My current data setup is simplified below:

date	fund	turnover	qe
1	1	12	1
1	2	13	1
1	3	14	0
2	1	15	1
2	2	13	1
2	3	12	0

I had previously run the time-series model by using the "if" command such as reg turnover qe if fund == 1 for instance

Question:
1. How should my data look like to run a SUR to compare all three fund's turnover against qe at the same time?
2. Can I run a particular command without reshaping my data?

my actual time-series reg code looks something like this if it helps:

reg CBF_totturnover CBF_turnover_lag CBF_cap_lag CBF_ret_lag set_turnover_y int_CBF_qe1 int_CBF_qe2 int_CBF_qe3 dumm_qe1 dumm_qe2 dumm_qe3 msci_ret_y dumm_qe_uk dumm_qe_eu dumm_qe_jp dumm_coup06 dumm_gfc dumm_edc dumm_coup14 dumm_pre_qe dumm_post_qe if CBF_id == 1

I believe it should not be too hard but I can't seem to find it online and really really need your help - I'm under an extreme time constraint so the pressure is killing me.


Thank you in advance!

[Clyde Schechter]

This does not sound to me like an application for SUR. Perhaps your advisor meant seemingly unrelated estimation? This would be a very typical application for that.

To do that, your data organization need not be changed from what you used for the original regressions. The approach would look something like this:

reg outcome predictor_vars if fund == 1
estimates store fund1

reg outcome predictor_vars if fund == 2
estimates store fund2

reg outcome predictor_vars if fund == 3
estimates store fund3

suest fund1 fund2 fund3

If you then additional wish to perform cross-fund contrasts of coefficients, you can use the -test- command following -suest-. See -help suest- for more details.

[David Benkert]

Thank you very much Clyde!
I will perform the analysis and get back to my advisor.

Ps.: Your answers have contributed by far the most of my STATA learning curve and really honored to have you personally answer my question.