EViews.com

Hello,

can anyone help me with one thing, I'm not that familiar with VAR models and my professor told me I could test this thing using VAR:

I want to compare two different time periods (2005-2007 and 2007-2009) and check if for example 5 different stock indexes are more correlated during one of the periods. I don't know if I actually need correlation figures for this, if I understood correctly this could be analyzed from the VAR estimates.

First i wold make a VAR regression of all the indexes, then i would take the residuals from that and use them to a new VAR regression (of the residuals). Before I do that i would change the residuals, i would make them in this format, log(residuals^2).

What I would end up with is one VAR mean equation and one VAR residuals equation. Is it possible to make any significant analyses from those two equations wether the stock indexes are more integrated during on period than the other, the idea is two compare the two periods.

Hopefully someone understands what I mean and can help me :)

If I'm not completely lost I need to use a Granger Causality test to see the difference between the two periods. Then I have seen some research that study the correlations between the residual returns, what does that actually describe in simple English? Is there any good reason why I have to transform the residuals in the form of log(residuals^2) and make a new VAR estimate of them like my professor suggested, is this to make the data more significant?