The RATS Software Forum

Dear TOM:

While I estimate 4 variate MVgarch,var(p)-MVgarch-BeKK, I use var as mean equation. However, @varlagslecet procedure gives different result, different var lags,eg, p, lead to different results of covariance - equation.
aic is 7, SBC is 1, HQC is 3.
How can I do?


haardmann.

With respect to VAR models in general, it's not uncommon to see conflicting results from the various lag-selection algorithms. In those cases, you really just need to consider the data carefully and use your best judgment. You may want to try different models, and do some tests on specific lag lengths (i.e. tests on omitting longer lags).

If you will ultimately be estimating the VAR as an MVGARCH model, the lag-selection statistics are of even less value. For the MVGARCH case, we would recommend that you choose a fairly small number of lags, and test the resulting standardized residuals for remaining autocorrelation.

Regards,
Tom Maycock
Estima