[hiba naeem]

Cralo,

Kindly suggest me what should I do now?

[Carlo Lazzaro]

Hiba:
perhaps you can consider a Mundlak approach (-search Mundlak-).

[hiba naeem]

Thanks again Carlo,
I am reading this approach but there are some confusions.

in the first step of this approach we have to compute panel level means of time varying covariates. now in my case we will take all independent variable as time varying?

[Carlo Lazzaro]

Hiba:
see https://blog.stata.com/2015/10/29/fi...dlak-approach/

[hiba naeem]

Carlo,

Following the given 3 steps I have performed this test on my data, this test also confirmed that fixed effect model is appropriate. Now what next? what will be now command of regression because after this text when I run xtreg with fe option the problem of omitting dummy is still there. I have used following commands:

bysort companyid :egen fsmean = mean (fs)
bysort companyid :egen flmean = mean (fl)
bysort companyid :egen fgmean = mean (fg)
bysort companyid :egen fcfmean = mean (fcf)
quietly xtreg roa fcf fl fs fg fsmean flmean fgmean fcfmean , vce(robust)
estimates store mundlak
test fsmean flmean fgmean fcfmean

and this resulted into rejection of null hypothesis so now will I use this command as final regression with fixed effects?

xtreg roa fcf fl fs fg fsmean flmean fgmean fcfmean , vce(robust)

am I right?

and where I will use dummies in this command?

[Carlo Lazzaro]

Hiba:
the Mundlak test is consistent with -hausman- test outcome, as the both point you to -fe- specification; as we have already discussed, -fe- specification wipes out time-invariant predictors (-sector- in you case) ans there's nothing you can do about that.
If you strive to get -sector- coefficient estimated, the only approach I can envisage is the Mundlak's correction (see -search mundlak-, if you haven't installed this user-written programme on your computer, yet). In brief, Mundlak's correction relaxes the -re- assumption of no correlation between the panel-wise effect and the time-varying predictors by including the individual mean of the time-varying predictors themselves.
Hence, you can include them in the right-hand side of the panel data regression equation and go -re- instead of -fe-.
That said, I would check the literature in your research field and see whether Mundlak's correction has been applied in the past to panel data regression analysis.

[hiba naeem]

Carlo,
will it be:
xtreg roa D1 D2 D3 fcf fl fs fg fsmean flmean fgmean fcfmean, re

if yes, then this can be done without using Mundlak test.

can we use LSDV in my case?

[Carlo Lazzaro]

Hiba:
I would go:

xtreg roa D1 D2 D3 fcf fl fs fg fsmean flmean fgmean fcfmean, re

which embeds the Mundlak's correction.

I fail to get why considering LSDV.

[hiba naeem]

Carlo,

To deal with Dummy variable trap in Fixed effect, LSDV can be used (this is my opinion, I read this). One of limitation of LSDV is not capturing time-invariant variable and in my case I have not time in-variant variable. That's why LSDV can be used in place of Fixed effects. If I am correct.

[Carlo Lazzaro]

Hiba:
yes, but until your last post, your approach was Mundlak (ie, a relaxed version of -re- specification).
That's why I was surprised when you mentioned LSDV.

[hiba naeem]

Carlo,

Actually I was stuck with these approaches; at this forum while searching for my case I came to know LSDV approach but till then we have talked about your suggested approach i.e. Mundlak.

Now kindly suggest me I can go with LSDV? I have tried this it worked but correct me I have used following command:
xtset companyid year
gen D1=( sub==1)
gen D2=( sub==2)
gen D3=( sub==3)
xtreg roa D1 D2 fcf fl fs fg

is it OK, or I should go with noconstant option

thanking in anticipation.

[Carlo Lazzaro]

Hiba:
yes, you can go LSDV.
It may happen that some reviewer/teacher/colleague will ask you to justify this approach vs -xtreg, fe-.
I proposed Mundlak's approach and correction as options to verify that -fe- specification was actually the way to go (Mundlak' approach as reported in https://blog.stata.com/2015/10/29/fi...dlak-approach/) and a possible way to estimate a coefficient for -sector- by relaxing -re- assumptions (Mundlak' correction, as reported in -search mundlak-) as you seemed very interested in it. As repeatedly discussed, this coefficient cannot be estimated vi -fe-, as this specification wipes out time-invariant predictors.

[hiba naeem]

Carlo, Bundle of thanks for clearing my confusions.

lastly one more thing. Using LSDV for panel data will use following command

xtreg or we will simply use reg? I am confused because nocostant option is only available with reg command

it means even if we have panel data we will use reg instead of xtreg to get value of coefficients of all dummies without intercept.

[Carlo Lazzaro]

Hiba:
go -regress- to perform LSDV.

[hiba naeem]

Thanks a lot Carlo,
You really helped me to clear my confusions.

Now I am going with LSDV and regarding this one more questions.

is there any need to test hetro, auto and multi?