Monthly observations and quarterly fixed effects

[Dominique Descat]

Hi everyone,

Hope you are doing well and maybe you have some handful insights for the following.

My paneldata consists of monthly observations over x years for y firms.
My variable of interest is also monthly.

When I include time fixed effects (monthly) my variable of interest is dropped due to collinearity with the time fixed effects. Which makes sense.
If I include quarterly time fixed effects instead of monthly fixed effects; my variable of interest is not dropped by stata.
However I want to ask you:
- If I include quarterly fixed effects in case of monthly observations; is my model then still valid?
- Why can or can't I use this approach of using a time fixed effect that spans a larger time period than the unit of observation? What is the reasoning behind it?

Maybe you know by accident a paper that touches upon this issue.

Thank you!
Regards, Dominique

[Maxence Morlet]

Hi Dominique,

If you take a look at Correia's (2017) reghdfe command, it will give you details about the nesting of fixed-effects; this is probably relevant for you. Collinearity might arise if the fixed-effects are nested, but I am not too sure about that (Sergio Correia explains it a lot better!).

Your time FE will now no longer capture monthly shocks but quarterly shocks, so depending on the field you work in, they may pick up less variation (in finance for instance, shocks are very punctual and returns are mean-reverting). What you may want to do is to interact these time FE with your unit FE. Since time FE are now in quarters, it should not be too much of a problem regarding computing difficulty. This will allow time shocks to have an idiosyncratic effect on each panel.

[Dominique Descat]

Thank you Maxence! I indeed will interact the time FE with my unit FE and I will look at the relevant literature to decided whether quarterly fixed effects are appropriate. Good to know that my model will still be valid when using quarterly FE. Thanks!