Stata is omitting FE variables

[Yash Sureka]

Hi,

I am trying to use panel data and regress excess returns on my independent variable, as well as many other control variables. In addition, I am also including monthly fixed effects and state fixed effects. However, upon conducting the regression, Stata omits all my state fixed effects "because of collinearity". My sample is using US, China, Japan, and South Korea, and I have 3 state FE variables (one for china, japan and korea each), which equals to 1 when it is the respective country, 0 otherwise. Could someone please help me understand why stata is omitting these variables and how I can prevent it from happening?

Thanks a lot!

[Ariel Karlinsky]

Your'e probably explicitly putting i.country in the varlist, and also estimating a panel FE with xtreg, fe. The two cannot co-exist.
But without sharing output, we can only guess. Please look at FAQ for an explanation as to how.

[Yash Sureka]

Hi Ariel,

This is the result I am getting. I haven't even included my control variables yet. Govt2 is my independent variable.

R-sq: Obs per group:
within = 0.0171 min = 31
between = 0.1555 avg = 61.5
overall = 0.0063 max = 112

F(1,241) = 4.19
corr(u_i, Xb) = -0.0439 Prob > F = 0.0417

------------------------------------------------------------------------------
d_exret | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
govt2 | -2274.634 1110.717 -2.05 0.042 -4462.586 -86.68147
china | 0 (omitted)
japan | 0 (omitted)
korea | 0 (omitted)
_cons | 739.6928 283.3876 2.61 0.010 181.46 1297.926
-------------+----------------------------------------------------------------
sigma_u | 4530.4097
sigma_e | 4413.2578
rho | .51309662 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(3, 241) = 47.23 Prob > F = 0.0000

[Ariel Karlinsky]

Per FAQ, enclose your output in [CODE] delimiters. Also, you don't show the command.
But from what you have posted I can see that I was right. When you use xtreg, fe you are already including country FE (assuming your panel unit is country). Thus they are omitted.

I suggest you read up on Panel estimation in stata. This is a great, example full resource:
https://www.princeton.edu/~otorres/Panel101.pdf

[Carlo Lazzaro]

Yash:
I do share Ariel's helpful comments and source (I would also add the following reference, conceived for Stata users: https://www.stata.com/bookstore/micr...metrics-stata/).
No wonder about your results, as -fe- estimator cancels out time-invarying predictors (such as nations, in your example), getting rid of both observed and unobserved heterogeneity related to time-invariant predictors. This can shelter you from omitted variable bias (if related to time-invariant predcitors), but comes at the cost of letting the coefficients of time-invariant predictors unestimated.
I would also take a step aside: did you investigate whether -fe- specification is the right one for your data (see -hausman- if you're intended to use default standard errors [SEs]; see -searchy xtoverid- to install a wonderful user-written programme intended to replace -hausman- if your SEs are not in the default mode)?
As an aside, please use CODE delimiters for sharing what you typed and what Stata gave you back (see the FAQ on this and other posting-related topics). Thanks.

[Yash Sureka]

Thank you very much, Ariel. I was able to fix my output.