EViews.com

Hello,
I am writing a dissertation about Capital structure and firm performance using a sample of 75 companies and 5 years.
ROA,ROE= βo+β_1 leverage+β_2 INSIDER shareholdings+β_3 tangibility+β_4 Cash ratio+β_5 EPS(lagged one period)
When I run my regression on ROA I get all the expected signs and significance levels, but when ROE is a dependent variable, results are absolutely insignificant(1 out of 5 is sign.). I have done only one semester of econometrics but I want to include something more than just a simple OLS.
I found something interesting: When I choose Cross-section: None, Period: None, GLS weights: cross-section weights, coef cov method: Ordinary my results become more significant both for ROA and ROE and my R-squared increases from 12% to 60%.
Can someone please explain to me if I should include results from OLS or the results with the weights and what is the meanings of these weights?

NB: I was told by my supervisor to do only simple OLS but I really want to understand the meaning of my results so I would be glad if someone helps me as I have no one to ask.

Thank you,
Natalia

Can someone please help me...
My coefficients are not significant and compared with previous studies they are quite different.
I did Cross-Section dependence test and I found that my variables are correlated.
Unfortunately, I have recently started using Eviews and I don't know exactly what options to choose.
I spoke with one of my lecturers who suggested that I do:
Least squares method
GLS weights: Cross-section weights
Coef cov method: White-cross section
(without choosing "no df correction")

Can someone please tell me if this suggestion is correct and what is the meaning of the options chosen?
Shall I say that I have chosen this method and options because of correlated variables?

PS: 75 companies and 5 years

Thank you