The RATS Software Forum

Hi,

I am searching for a procedure that tests the properties of VAR residuals; autocorrelation, normality, cross equations correlation.

I would like to use these properties to detremine the number of lags of the VAR model as I have a limited number of data.

Thank you

AhmedSahlool wrote:Hi,

I am searching for a procedure that tests the properties of VAR residuals; autocorrelation, normality, cross equations correlation.

I would like to use these properties to detremine the number of lags of the VAR model as I have a limited number of data.

Thank you

@MVQSTAT does the multivariate Q test for autocorrelation (which includes both self and cross)

@MVJB does a multivariate Jarque-Bera test. Though I wouldn't recommend basing decisions about lag length on the J-B statistic, since adding lags does nothing directly to affect the normality of residuals.

The VAR that I estimate has non constant variances.

I read that the MVQSTAT is not consistent in this case, are there other alternatives?

Thank you

AhmedSahlool wrote:The VAR that I estimate has non constant variances.

I read that the MVQSTAT is not consistent in this case, are there other alternatives?

Thank you

The West-Cho test will handle the autocorrelations. If you don't have much data, a generalization of that to a multivariate process would probably have rather bad properties.