Panel Data: Jarque Bera and Heteroskedasticity

[Rhadamantes]

Hi,

I am currently performing a panel regression using a least square method. Thereafter, I tested the normality of the residuals and unfortunately my Jarque Bera is around 283. I have read (notably here) that this was not a big concern except if the residuals were far away to be normal. So if this is true, can somebody give me a reference (I will probably have to justify it) or is it due to the Central Limit Theorem? Moreover, how do we know that the distribution is too far away to be normal? Is looking at the shape ok?

Finally, I would also know how to determine which "coef covariance method" I should use between the 8 proposed (Ordinary, White cross-section, White period ...), I have seen that they have to be used in presence of heteroskedasticity but I do not find how to perform a related test. In case there is none, is there a best default option?

For information, I am using Eviews6, my database goes from 1954 to 2010 and has 8 cross-sections.