Fixed effects Panel data with 3 sector dummies

[hiba naeem]

Hi, I am working on a data set with 70 companies and 6 years. Companies are from textile industry (having 3 sub sectors i.e. composite, spinning and weaving sector) and I have to analyze relationship between performance and cash flows sector wise for which I have to use sector dummies. The hausman test has confirmed that fixed effect will be suitable for my panel data set; then how I will use dummies in this? And how I will run regression? how panel data Heteroskedasticity, autocorrelation and multicollinerity problems will be checked in this case.

hope to hear soon.

[Carlo Lazzaro]

Hiba:
despite not being your first post, you do not share with interested listers anything relevant for a positive reply.
By the way, if your concerns focus on the role of -i.subsector- categorical variable, which regression codes did you test via .-hausman-?

[hiba naeem]

thanks for your response an apologies for incomplete information. Fixed effects are confirmed for testing relationship for whole industry (without adding sector dummies). I have generated 2 sector dummies. Actually the confusion is that should I include these 2 dummies in regression and then to test for appropriate model (fixed or random)?

[Carlo Lazzaro]

Hiba:
can't you simply include -i.sector- categorical variable as a predictor from the start and test whether -fe- fits your data better than -re- specification (or the other way round?).

[hiba naeem]

Correct me if I am wrong

I have used following STATA commands for sub sector (Composite, spinning, weaving) analysis:
xtset companyid year
xtreg roa fcf fs fg if composite,fe
xtreg roa fcf fs fg if spinning,fe
xtreg roa fcf f fs fg if weaving,re

[Carlo Lazzaro]

Hiba:
instead of running three different regression models, you should consider an unique regression model, where you have an unique catgorical variable for subsectors (see -help fvvarlist-).
Hence, your code should become:

xtreg roa fcf f fs fg i.sub_sector,fe

[hiba naeem]

thanks Carlo,
I will ask for help again if still have any issue.

[hiba naeem]

When I used following command

. xi: xtreg roa fcf fl fs fg i.sector, fe

the results given below are showing 0 coefficient for 2 sectors what does it mean? kindly guide what should I do? when I used this command for individual sector the results were different.

[Carlo Lazzaro]

Hiba:
- -xi- is redundant with -xtreg- (and many other commands provided by Stata recent releases); -fvvralist- notation is enough;
- two levels of the categorical variable -Sector- were omitted due to collinearity (as Stata reported): this is the way the -fe- machinery works; there's nothing you can do about that, but changing your model specification (that is, switching to -re-) if the -hausman- test point you to -re-;
- for the future, as you should already aware of, please do not post screenshots, buit share what you typed and what Stata gave you back via CODE delimiters. Thanks.

[hiba naeem]

thanks Carlo,

When I used i.sector command in STATA I got this message:
. xtreg roa fcf fl fs fg i.sector

sector: string variables may not be used as factor variables

data in "sector" variable is in the form of
Composite
Spinning
Weaving

kindly guide

[Carlo Lazzaro]

Hiba:
-string- format is not suitable as -panelid-.
Try instead:

egen sector_num=group(sector)
label defibe sector 1 "composite" 2 "spinning" 3 "weaving"
label val sector_num sector_num
xtreg roa fcf fl fs fg i.sector_num

[hiba naeem]

it worked and when I tested for fixed or random effect, hausman test is giving fixed effects but fixed effect is again showing :

xtreg roa fcf fl fs fg i.sector_num, fe
note: 2.sector_num omitted because of collinearity
note: 3.sector_num omitted because of collinearity

now when hausman test is giving fixed effects how I can change model?

[Carlo Lazzaro]

Hiba:
Stata reports exactly the way -fe- machinery works: time-invariant predictors are wiped out.
There's nothing you can do about that (and -re- specification would give back biased results)..

[hiba naeem]

Carlo, then can I use If command to get results of each sector separately? by using following commands I can get results:
xtset companyid year
xtreg roa fcf fs fg if composite,fe
xtreg roa fcf fs fg if spinning,fe
xtreg roa fcf f fs fg if weaving,re

is it OK?

[Carlo Lazzaro]

Hiba:
I do not sponsor your approach that misses the contribution of -sector- in explaining the variation of the conditional mean of the dependent variable.