EViews.com

Even if there is heteroskedasticity, the estimated equations should be unbiased. What would you differently if you found there is heteroskedasticity?

What would you recommend to do in this case, as long as I can t perform the test on that 6 variables?

I don t know if there is heteroskedasticity or not. I want to test it.

I would just proceed without the test.

And why do you want to test for heteroskedasticity anyway?

Because I read in an article that the VAR models have to meet the conditions regarding the quality of the residuals - the normal distribution (Normality test), the absence of heteroskedasticity (White heteroskedasticity test) and the lack of autocorrelation of the errors (Autocorrelation LM test).

The article is just wrong.

A VAR is just a collection of least squares regressions. Normality and heteroskedasticity matter for coefficient tests, but usually no one is interested in doing those for a VAR. Autocorrelation does matter-it's usually handled by adding lags to the VAR.

Here is my autocorrelation test. Could I choose the first lag?

This probably suggests adding one more lag to the VAR. But you have very few observations, so it's not entirely obvious whether you should.

Could I run this model with just one lag?

Perhaps run with one lag and run with two and see if it makes any difference. The real issue is that only having 20 observations limits what you can do.

Thank you so much for all you advices!

Sorry, I forgot. Should I eliminate high correlation between variables before I create the VAR?

No, the VAR is fine with highly correlated variables.

How can I solve the problem if the normality test is p 0.000, in a VAR model?

What do you think the problem is?