-xtreg, re- and year and industry fixed effects

[Marie Anache]

I am running a panel regression with -xtreg, re- and would like to account for both year and industry fixed effects. For this purpose, I created the variable "ff_12", which places each firm in one of the 12 fama & french indusries. My problem is that my interaction term includes the dummy variable "cont_ind". This dummy is equal to 1 when a firm operates in a "controversial industry" and 0 when not. Including i.ff_12 within the same model is therefore not ideal - because both are correlated.

My current code is:

xtset panel_id year
xtreg erm_index_w L.c.ECSR_w##cont_ind roa_w leverage_w i.bigfour i.crisis_dummy i.ff_12, re vce(robust)

Is there any way to deal with this? And could you confirm that xtset panel_id year already serves the purpose of controlling for year fixed effects?
panel_id currently corresponds to the firm id.

[Andrew Musau]

In linear models, you can include fixed effects using indicators. So yes, you can have industry indicators and time indicators.

And could you confirm that xtset panel_id year already serves the purpose of controlling for year fixed effects?

It does not. You need to explicitly include these using indicators.

My problem is that my interaction term includes the dummy variable "cont_ind". This dummy is equal to 1 when a firm operates in a "controversial industry" and 0 when not. Including i.ff_12 within the same model is therefore not ideal - because both are correlated.

Is "cont_ind" a variable of interest? If it is simply a control, then it does not matter that it is collinear with the industry indicators. Its effect is already accounted for by the industry indicators if it drops out. Finally, you need to ensure that a random effects model is justified in the first place. The command below will help with this.

ssc install xtoverid, replace
help xtoverid