[Carlo Lazzaro]

Jihad:
please note that, while creating a regression model with no omitted variable bias (no endogeneity) and no heteroskedasticity (quasi-extreme multicollinearity is, in general less relevant for predictors and almost immaterial for controls) means fulfilling regression requirements, searching for statistical significance as your ultimate research goal is not.
Everyone with an average knowledge of frequentist statistics (like me, for instance) does not care that much about p<0.05, but look to confidence intervals, sample size, missing values and the reperesentation of the data generating process provided by the regression model, instead.
That said:
- your first model, as already expressed, due to its sky-rocketing R-sq values looks like a data make-up;
- as far as your second model is concerned: why did you start with -regress- instead of -xtreg-?

[Jihad El Yaagoubi]

I tried xtreg fe, re and seems not appropriate (results of LM test p =1.000 and hausman test).
what are characteristics of a good model (not perfect one) just good to be reported in a thesis ?

[Carlo Lazzaro]

Jihad:
- go pooled OLS with ROS as dependent variable and clustered standard errors on panelid (as you already did);
- the requirements of a reliable OLS model are the ones reported in my previous reply (forget about perfect models: they do not exist in real life);
- your ROS model seems defendable, whereas I would be more concerned about the ROA one.

[Jihad El Yaagoubi]

So even if the sample size is only 24 firms(not 28 as the previous) the ROS model is defendable ?

[Carlo Lazzaro]

Jihad:
yes, I would say so.

[Jihad El Yaagoubi]

Thank you very much ...

[Jihad El Yaagoubi]

Hello again,
How can I interpret standard errors?

[Carlo Lazzaro]

Jihad:
admittedly, I fail to get your question.
Standard error is, as usual, the standard deviation of the sample distribution of the parameter of interest.
Please note that any decent textbook on regression/econometrics covers this issue.

[Jihad El Yaagoubi]

Dear Carlo,
im sorry for the late reply. I mean if I have SE=0.00127 and SE=0.30345, which one is better ?

[Jihad El Yaagoubi]

Do we use vce (cluster companyscode) to control for heteroskedasticity and autocorrelation? I found that cluster () control for this two assumptions and I don't know if it's correct.

[Carlo Lazzaro]

Jihad:
- under -xtreg-, -cluster- and -robust- options do the very same job and accounts for both heteroskedasticity and/or autocorrelation;
- under -regress-, -robust- option accounts for heteroskedasticity, whereas -cluster- option accounts for autocorrelation.
As an aside, I still fail to get your question about standard errors.

[Jihad El Yaagoubi]

Dear Carlo, thank you very much for your answer. Im so sorry if I was not clear in my question. When I started interpretation, I interpreted p values and their significance, R-squared and I don't know if there is other values to interpret in the outputs of the software results, so I ask about standard errors if they should be interpreted.
Im sorry again for the late reply but I have some troubles for this I couldn't connect before.

THANK YOU