Is there any way to absorb time dummies in xtreg, fe?

[Jeon Joan]

Hello

I want to do

code : xtreg depvar indepvar i.day, fe vce (r)

with the panel data. (day is a time variable)

But the data is too big, so I can not use the above code

Is there any way to absorb time dummies in - xtreg, fe - like - areg, absorb() - ?

I used reghdfe and it is not available because the coefficient and t-value are too different compared with xtreg.

thank you for reading

[FernandoRios]

I would revise again the results comparing reghdfe and xtreg, fe
by construction both methods should give you the same point estimates with only small differences in standard errors when clustering by the panel id variable
can you post the results and commands you are using to see if there is any syntax mistake?
fernando

[Jeon Joan]

I wanted to do
CODE : reghdfe, absorb(week week#city) vce (cl id) compact pool(1)
but when I do that I get the message "Warning: variance matrix is nonsymmetric or highly singular" and strandard errors do not appear.
Do you have any reason to guess that the coefficients and t-values of the two codes are significantly different?

[FernandoRios]

Hi Jeon
Now that you show the results, it does puzzle me why are the estimations so different.
Couple of thoughts come to mind.
1. Drop the pool(1) and compact option and see if that changes anything.
2. I wonder what is the structure of your panel data. is it City by week? if that is the case, then you are "overfitting" the model. Both of your specifications are controlling for each combination of city and week.
3. Because you are controlling for all City-week combinations, this is not a two way fixed effect model, but only one way fixed effect. Perhaps you can try the following:
egen cityweek=group(city week)
areg margin brand (all other x's variables), absorb(cityweek)

HTH
Fernando

[Jeon Joan]

dear FernandoRios

I'm too late.
Thank you for your advice.
I looked up and found that I was making a big mistake.
I did not know that id should be included in absorb ().
Solved the problem
Thank you
:D