0
$\begingroup$

I am new to Stata and I am using a panel data of 203 firms over 4 years over different countries for my thesis. Time-invariant variables are owner type dummies, an ownership concentration variable, industry dummies, and possible country dummies. Rest are time variant including a country level variable which I want to use as an interaction term. I wonder how to proceed from now as I don't know what test is appropriate?

A paper I am following with similar constructs neatly uses 3 different models. The first model is seen in the picture, CG and F are firm-level vars, so is this firm fixed effects? In model 2 they introduce the country-level variable. And then interacts country and firm level in model 3. Is this something I could do? It seems appropriate since I also want to use the country factor as a moderator.

However, recent advances in my field also argue for using a system GMM, since my variables rarely fit the strict exogeneity assumptions. Would it be appropriate to do the 1-3 model as above, then lastly comparing those static models with the dynamic?

enter image description here

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

First of all: Welcome.

Second, of course it is appropriate. Without knowing the full background of your question, it seems like there is solid reasoning to include both types of models, especially in a thesis. As a thesis supervisor, I always encourage my students to consider alternative models and approaches (if there is an argument for it, of course). I'd at least discuss the different model types and then argue why I committed to one of the models.

For the rest of your question I recommend to take a deeper look into the Hausman test as well as into Greene "Econometric Analysis". Without knowing your dataset, we can't really give you advice on how to build your model.

Regards, Richard

Improve this answer
$\endgroup$
1
  • $\begingroup$ Hi Richard, thank you and I appreciate the reply. So I tested the FE estimation on my baseline model, but it dropped many variables as they are time-invariant. The Hausman just tested the non-omitted control variables, so it doesn't really help me. The RE estimation resulted in insignificant results. However, I did not test with interaction terms. This is a bit worrying. Any ideas? $\endgroup$
    – Kristian Pal
    Commented Feb 26, 2019 at 10:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.