Robust Standard Errors, Fixed Effects - Panel Data

[guneet robin]

I am using a panel data model of ~120 brands from year =1 to 8. All these brands belong to a particular parent brand also. Understandably, data will be characterized by serial correlation and heteroskedasticity. Using the Breush Pagan Test & Hausman Specification and also based on theory, I believe that I have to use fixed effect specification for my model. But I am curious that if I have accounted for heterogeneity in my model by including fixed effects, why do I need robust standard errors?


Also, how will the robust standard errors differ if I use mother brand fixed effects i.motherbrand instead of i.brand in the model?

Thanks in advance,
guneet

[Carlo Lazzaro]

Guneet:
-you may check yourself if robust standard errors (SEs) (I assume that you mean -vce(cluster idpanel)- SEs) differ from default SEs;
- where either I.brand_fixed_effects or I.brand makes more sense probably depends on your research purposes. Anyway, an advisable first step is to skim the existing literature and find out what others did in the past (in terms of regression models) when dealing with the same research topic.