Would constant annual variables across firms in a year be fully absorbed by the year dummy?

[Mona Yaghoubi]

Hi all,
I am running the following regression using reghdfe written by @Sergio Correia:

reghdfe f.Dependent U c.U#c.Q Q $xlist i.year, absorb(firmcode) vce(cluster firmcode year)

where the dependent variable is the firm’s debt ratio and U and Q are my variables of interest and are constant for all firms in a year, and as you know, c.U#c.Q is the interaction between U and Q. I have 20 years of unbalanced data for 5123 firms.

When I run my regression, one-third of the year dummies (the latest years) will be omitted because of collinearity; they will be only omitted when I add U, Q or their interaction.

Would this assumption be correct that because I put the i.year at the end after all other variables, Stata will automatically only absorb the part of the year fixed effect that has not been already absorbed by the U or Q variables (or the other annual macro variables like GDP)? Or would U and Q be fully absorbed by the year fixed effect?

I appreciate your help.
Thank you.
Mona

[Jared Greathouse]

Yep. No constants are allowed with TWFE models


EDIT: Constant within years? Then no. At least I don't think so.

[Andrew Musau]

I don't think that the coefficients on the year dummies are of any use to you, so you should absorb them as well. More importantly, doing this will let you know directly what right hand side variables are collinear with the year effects. With the dummies, you need to change base levels to figure this out.

reghdfe f.Dependent U c.U#c.Q Q $xlist, absorb(firmcode year) vce(cluster firmcode year)

[Joro Kolev]

Everything that is changing only across time (GDP, Inflation, any macro factors) will be absorbed by a full set of time dummies.

Everything that is changing only across the cross sectional units will be absorbed by the firm fixed effects.

[Mona Yaghoubi]

thank you all