Analysing different models within the same panel data

[Dejan Toscano]

I'm currently analysing the effect of the borrowers gender on the performance on financial service providers.
So far the advice I got on this forum has helped me greatly, however I'm still struggling with one question.

I defined the overall performance of an FSP as four dependent variables, which are

• portfolio at risk
• operating expense ratio
• operational self sufficiency and
• return on assets

The predictors stay the same within all four models.
Now after testing whether I can apply RE or FE (with the Mundlak devide Mr. Wooldridge recommendet in this post), I found RE to be consistent for two of those models, and inconsistent for the other two, implying I have to use FE (or at least that I can't use RE) for the latter.
I'm wondering, can I run RE on two models, FE on the other two and still make a general statement concerning the performance?
My intution tells me yes as results for both models should be consistent, however I haven't found any literature concerning this and I don't want to make an obvious mistake.
Any literature recommendation is also appreciated!

Thanks in advance!

[Phil Bromiley]

You'll increase the chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Also, assume we're not from your area. I don't even know the structure of your data - it seems like a panel but what the unit of observation is within that panel is unclear.

You should look at xthybrid and the Mundlak estimators users have provided for estimating such models. I'd be a bit uncomfortable comparing fe and re estimates on your structural models even if the Hausman is not significant. Fe and re are really estimating different things.

[Dejan Toscano]

Thank you Phil, your reply is already enough for me. I've taken some more time to read about the different functions of fixed and random effects estimators and came to the same conclusion as you did. I did not include any any code as I thought this may be a question more on the theoretical side and thus not needed, I'll keep it in mind however. Thanks again for taking the time to reply!