Questions about Panel Data - I am a bit confused

[Sinclair Alowa]

Hello everyone, so I am doing a panel data on fundraising determinants in private equity. It consists of 5 countries over the period (2010-2022).

These are the steps I have in mind according to my research:

1. Unit Root Tests (checking for stationarity)
2. Linearity
3. No edogeneity
4. No collinearity
5. Homoscedasticity
6. No autocorrelation.
7. Independence of obserations.
8. Normality of residuals.

My questions:

1. Do all the assumptions have to be validated? Because what i found online and even in the reports of other students: they focus solely on autocorrelation, Homoscedasticity and collinearity.
2. Do I need to address each assumption and only move on to the next step if it is validated?
3. When should I remove outliers? Because I have seen somewhere that it's better to keep them.
4. Which method is better to deal with The heteroscedasticity problem? Is it the robust command or gls?
5. Is it okay to run multiple iterations in the case of gls?
6. If i find that a gls model is appropriate, but then i find cross-sectional dependence issue and i moved to another model, is that correct?

Thank you!

[Carlo Lazzaro]

Sinclair:
welcome to this forum.
As far as your questions are concerned:
1) Yes, they should, but 8., which is a weak requirement for residual distribution;
2) All the asusmption should be evaluated together;
3) You should keep the (so called) outliers, unless they represent a blatant example of mistaken data entry;
4) if you have a N>T panel dataset, go -robust-. Otherwise, please be more specific in a follow-up post;
5) the default iteration are 100. Why do you want to increase them?
6) See -help xtscc-.