Hi guys. I'm doing an empirical analysis at the moment and wanted to ask you few things. My data is a cross-section data with 1000 observations. In my model I'm using 12 independent variables -> there I have many dummies, one square term and 2 interaction terms. So I transformed the linear model into a semilog model (my dependent variable is log, the rest not).
Because of multicollinearity problem, I had to do transform two variables (one interaction term and the square one). So then was no multicollinearity problem, ramsey RESET test was fine too, but the problem is with normality. There is no normal distribution (did the Jarque-Bera test). This was a problem when I wanted to do White Test. I tried in gretl and in Eviews and in the 1st one, the software did the text, but I got a warning msg at the end, that " data matrix close to singularity". I tried in Eviews, but the software won't do the test at all, telling me I have insufficient number of observations... Which is ridiculous. So I looked for some informations on the internet and then I realised, White Test is working only under hypothesis of normal distribution. Then I found information, that with big samples (like more than 100) the normality is not that important anymore, it becomes asymptotic and that's it. So my question is, can I just use Breusch-Pagan Test instead? Also, does this non-normality affect my final results, so I cannot interprete them or not?
Thank you guys in advance.