Panel Data Regression

[Marijke EUR]

Hi All,

I would like to get help on a multivariate regression that I am running on a panel data set. I'm doing a research on the effect of shareholder activism on firm performance. I run a multivariate regression with dependent variable ROA and independent variable a dummy for target and several control variables. However, the outcomes of my regression are unusual. According to my professor are the independent variables to significant and this is a problem. I am struggeling with where it went wrong and I would like to get some advice on the steps that I might have forgotten. I ran the following tests, see attachment for results:

1) Test for non-normality
sfrancia
2) I find non-normality and winsor the data at the 10% level. The data is still not nonmally distributed after but looks better
winsor2
sfrancia
3) use Hausman test and find fixed effects should be used
xtreg ROA Target logMV BVMV Lev, fe
estimates store fixed
xtreg ROA Target logMV BVMV Lev, re
estimates store random
hausman fixed random
4) I check if there are time-fixed effects and find there are
xtreg ROA Target logMV BVMV Lev i.Year, fe
testparm i.Year
5) I perform a Wald test for heteroskedasticity and control accordinlgy with robust
xttest3
6) I test for serial correlation performing a Woolridge test and find there is -> vce()
xtserial
7) Latstly, I check for multicolinearity -> no
vif
8) I perform a regression using xtreg fe vce(robust)

According to my professor the results on the regression are too significant and he worries that something is wrong. I don't understand why it is a problem that the data is significant?
I wonder if someone could see if I made a mistake or forgot a specific test.

Help would be appreciated!! Thanks in advance

Kind regards, Marijke

[Carlo Lazzaro]

Marijke:
just one comment on your code and related post-estimation test:
how could you perrform a -hausman- test with robust standard errors?

[Marijke EUR]

Hi Carlo,

I don't fully understand what you mean. I run the test by using the following code:

xtreg ROA Target logMV BVMV Lev, fe
estimates store fixed
xtreg ROA Target logMV BVMV Lev, re
estimates store random
hausman fixed random

[Carlo Lazzaro]

Marijke:
in your code you used robust standard errors, that -hausman- does not support.
Hence (although that might not be your core problem), you've seemingly imposed the -robust- option after the -hausman- outcome: this is not quite correct.
You should have used the user-written programme -xtoverid- (type -search xtoverid- from within Stata to get it) instead of -hausman-.

[Marijke EUR]

Thank you for your fast response!
I am I right if i then should use:
xtreg ROA Target logMV BVMV Lev, fe vce(robust)
estimates store fixed
xtreg ROA Target logMV BVMV Lev, re vce(robust)
estimates store random
xtoverid fixed random

If I do this test i get the error:

varlist not allowed
r(101);

[Carlo Lazzaro]

Marijke:
you should simply run:

xtreg ROA Target logMV BVMV Lev, re vce(robust)
xtoverid

[Marijke EUR]

When I run the test I get a p value of 0.0006 meaning that I should use the fixed effect model, right?
Then I guess the regresssion xtreg ROA Target logMV BVMV Lev i.Year, fe vce(robust) should be correct. Do you see another mistake? I wonder why it is a problem that my control variables (logMV BVMV Lev) are very significant? I do not understand what is wrong with the regression output.

Apologies for all the questions, I really appreciate your help!

[Carlo Lazzaro]

Marijke:
as per -xtoverid- oucome, you should actually go -fe-.
I do not think there's anything wrong with your regression code and output; the latter simply reflects your data.
However, I've just noticed that you omitted -i.year- from your previous regression code.
If your omission was accidental and you were willing to re-run -xtoverid- including -i.year-, please note that -xtoverid- does not support factor notation; however, there's a workaround based on the -xi- prefix:

xi: xtreg ROA Target logMV BVMV Lev i.Year, fe vce(robust
xtoverid