One-way fixed effect model with clustered error

[Claire Noh]

Dear All,

I have a panel dataset, which is unbalanced, consisting of about 200 unique firms over 15 years.
Based on my research questions, I want to conduct multiple regression analyses, using fixed effect models (and the results of Hausman test also support FE instead of RE).

To control for firm- and year- specific heterogeneities, I initially taught about using two-way fixed effects both at the firm and year level. However, one of my independent variables includes GDP, which is constant across all firms for each year. So, if I include the year-specific dummy in the model (along with the firm-specific dummy), VIF value on GDP becomes extremely high (greater than 100).

In this case, can anyone suggest some options that I can take?
I am thinking about controlling only for the firm-effect with clustered error at the year-level. Does this make sense?
Any suggestions would be appreciated.

Thank you!

[Carlo Lazzaro]

Claire:
standard errors should be clustered on -panelid- (which is -firm- in your case).
Your model already accounts for -firm- fixed effect, if you -xtset- your data as:

xtset firm year

That said, I would only add -i.time- to your predictors.
As a aside, please note that you cannot compare via -hausman- -fe- vs -re- specification with default standard errors and then decide to invoke cluster/robusr standard errors, otherwise -hausman- test outcome may be unreliable.
Hence, you should decide before teh abovementioned comparison if your model deserves robust/clustered standard errors or not. If yes, the -hausman- test should be replaced by the user-written command -xtoverid- (tyoe -search xtoverid- from within Stata to spot and install it).
Last bust not least, please note that the best way to get more) helpful replies is sharing what you typed and what Stata gave you back via CODE delimiters (see the FAQ on this and other posting-related topics. Thanks).