Omitted variables when running with xtreg

[Emma Ien]

Hello!

I'm using panel data including the following variables,
(y variable: abnormal return, cumulative abnormal return)
(x variable:GovScore log_BM log_Size CAR AR EnvirScore ESGScore Size Lev SocialScore)
These variable are all focused on a short event window for like a month, hence, especially for the score variable, it maintains the same, hence, that means it is time-invariant?

My goal was to carry out regressions and do the Hausman test to see if I should use fixed effect or random effect. And then hopefully to move on to find the relation between return and ESG scores.

But when I did the code:

This happened

I've read across in the thread of using a 'hybrid model' but few questions seems to encounter problems at the early stage of just doing the xtreg coding, hence, does it mean there is fundamentally something wrong with my dataset (my dataset is strongly balanced) and it should be better applied as an OLS model? But this would be very wired since I am carrying out an event study.

And when I did normal regressions, this was the result, does that mean the result is insignificant no matter if it was panel data or not?

This relates to a previous question posted on the server:
https://www.statalist.org/forums/for...edicted-return

Thank you very much,
Emma

[Emma Ien]

reddit post reference: https://www.reddit.com/r/stata/comme...ng_with_xtreg/

[Carlo Lazzaro]

Emma:
the issue is probably more trivial: you ran two different models.
As far as -xtreg,re- is concerned, have you run -xttest0- as a postestimation test to investigate whether a panel-wise effect actually exist?
Eventually, since you have a relevant number of panels, you should impose clustered standard errors (and switch to the community-contributed command -xtoverid- instead of -hausman- to test which specification is the best one for your dataset).

[Emma Ien]

Emma:
the issue is probably more trivial: you ran two different models.
As far as -xtreg,re- is concerned, have you run -xttest0- as a postestimation test to investigate whether a panel-wise effect actually exist?
Eventually, since you have a relevant number of panels, you should impose clustered standard errors (and switch to the community-contributed command -xtoverid- instead of -hausman- to test which specification is the best one for your dataset).

Hi Carlo,

Thank you for your reply, since I'm quite new to stata, could you kindly go into a bit more detail?

This is what I coded.

xtset com_id date, daily

* LM test for the penal wise effect
xtreg CAR ESGScore Lev log_Size log_BM
xttest0
invalid syntax
r(111);

Could anyone hint why is this invalid?

Thank you,
Emma

[Emma Ien]

Hello!

I'm using panel data including the following variables,
(y variable: abnormal return, cumulative abnormal return)
(x variable:GovScore log_BM log_Size CAR AR EnvirScore ESGScore Size Lev SocialScore)
These variable are all focused on a short event window for like a month, hence, especially for the score variable, it maintains the same, hence, that means it is time-invariant?

My goal was to carry out regressions and do the Hausman test to see if I should use fixed effect or random effect. And then hopefully to move on to find the relation between return and ESG scores.

But when I did the code:

This happened
[ATTACH=CONFIG]n1608776[/ATTACH]

I've read across in the thread of using a 'hybrid model' but few questions seems to encounter problems at the early stage of just doing the xtreg coding, hence, does it mean there is fundamentally something wrong with my dataset (my dataset is strongly balanced) and it should be better applied as an OLS model? But this would be very wired since I am carrying out an event study.

And when I did normal regressions, this was the result, does that mean the result is insignificant no matter if it was panel data or not?
[ATTACH=CONFIG]n1608775[/ATTACH]

This relates to a previous question posted on the server:
https://www.statalist.org/forums/for...edicted-return

Thank you very much,
Emma

Sorry for the previous confusion, this is what I meant when I talking about omitted variables. Since the CAR represents cumulative abnormal return and being generated from the same return as AR(abnormal return). Just wondering what could go wrong? Other than this regression, the others work just fine.

[Emma Ien]

Emma:
the issue is probably more trivial: you ran two different models.
As far as -xtreg,re- is concerned, have you run -xttest0- as a postestimation test to investigate whether a panel-wise effect actually exist?
Eventually, since you have a relevant number of panels, you should impose clustered standard errors (and switch to the community-contributed command -xtoverid- instead of -hausman- to test which specification is the best one for your dataset).

Hi Carol,

Thank you for the reply!

My results are all significant for the xttest0.

And for my event windows, it is 57 days (instead of previously mentioned one month), hence I would still be using clustered standard errors?

Thank you,
Emma

[Carlo Lazzaro]

Emma:
as per your post #1 it seems that you have 139 panels: that's why it is advisable to invoke clustered standard errors.

[Emma Ien]

Hi Carlo,

Thank you for your reply, since I'm quite new to stata, could you kindly go into a bit more detail?

This is what I coded.

xtset com_id date, daily

* LM test for the penal wise effect
xtreg CAR ESGScore Lev log_Size log_BM
xttest0
invalid syntax
r(111);

Could anyone hint why is this invalid?

Thank you,
Emma

see: https://www.statalist.org/forums/for...d-syntax-r-111

[Carlo Lazzaro]

Emma:
sorry, no.