Panel Data Regression: Omitted time variable when doing fixed effects

[Marius Karlsen]

Hi All,

A bit urgent post for me as I am submitting my paper the upcoming week :/

I have a panel dataset from 2013 to 2021 years trying to explain the ESG scores effect on excess returns by using the Fama French 5 factor model. When trying do fixed id and time effect regression I experience that some of the years gets omitted because of collinearity.

code:
xtreg Excess_Return MRP SMB HML RMW CMA EIK_TOT i.YEAR, fe


Output:
. xtreg Excess_Return MRP SMB HML RMW CMA EIK_TOT i.YEAR, fe
note: 2017.YEAR omitted because of collinearity.
note: 2018.YEAR omitted because of collinearity.
note: 2019.YEAR omitted because of collinearity.
note: 2020.YEAR omitted because of collinearity.
note: 2021.YEAR omitted because of collinearity.

Fixed-effects (within) regression Number of obs = 1,125
Group variable: ID Number of groups = 125

R-squared: Obs per group:
Within = 0.3285 min = 9
Between = 0.0029 avg = 9.0
Overall = 0.2834 max = 9

F(9,991) = 53.86
corr(u_i, Xb) = 0.0004 Prob > F = 0.0000

------------------------------------------------------------------------------
Excess_Ret~n | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
MRP | 1.152452 1.052882 1.09 0.274 -.9136828 3.218587
SMB | 1.076407 .8548503 1.26 0.208 -.6011176 2.753932
HML | -.1150325 .3378545 -0.34 0.734 -.7780249 .5479599
RMW | .8958188 1.177744 0.76 0.447 -1.415339 3.206977
CMA | .7731153 .1285679 6.01 0.000 .5208187 1.025412
EIK_TOT | -.0054181 .0492966 -0.11 0.913 -.1021558 .0913197
|
YEAR |
2014 | .0541499 .0246891 2.19 0.029 .0057011 .1025988
2015 | -.0250039 .0279268 -0.90 0.371 -.0798064 .0297986
2016 | -.0395275 .0665556 -0.59 0.553 -.1701336 .0910786
2017 | 0 (omitted)
2018 | 0 (omitted)
2019 | 0 (omitted)
2020 | 0 (omitted)
2021 | 0 (omitted)
|
_cons | .0052125 .0266213 0.20 0.845 -.0470281 .0574531
-------------+----------------------------------------------------------------
sigma_u | .0834369
sigma_e | .18167517
rho | .17418405 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(124, 991) = 1.90 Prob > F = 0.0000

When changing the "EIK_TOT" variable to a different variable 2016 also get omitted.

Is there some way I can get rid of this issue and possibly fix it?

Please let me know if there is something more I can provide to more easily help with the inquiry.

Thanks in advance for any help you can provide to help me!

/// Marius

[Carlo Lazzaro]

Marius:
welcome to this forum.
This omission usually happens because the predictor is perfectly collinear with the fixed effect.
Please also note that urgent issues (as legally perceived by the OP) do not give you any fast lane on this forum.
Conversely, it would be more helpful sharing an excerpt/example of your dataset via CODE delimiters (as per FAQ).
In addition (but it is clearly not my business), I'm a bit surprised that you discover this issue just one week before the scheduled submission on your paper.

[Marius Karlsen]

Carlo:
Tanks for your help and my apologies about the comment that this was urgent.
It makes sense that omission usually happens because the predictor is perfectly collinear with the fixed effect. I will look into this.

Thanks.