Compare unrestricted and restricted VARX-DCC model

[mengqi]

Hi Tom,
I have bivariate VAR-DCC model with some exogenous variables in the conditional mean equations. How do I run a F-test, to see whether a restricted version of the model is preferable?
kind regards

[TomDoan]

Probably the simplest thing would be to do a likelihood ratio test. To do a Wald test, use the Statistics--Regression Tests, pick Exclusion Restrictions and select the exogenous variables in the scrolling list.

[Esteban]

Hi Tom,


One doubt about what you answered. The thing is that the GARCH instruction doesn't define the variable %logdet (Log determinant), so what should we do in order to calculate the loglikelihood ratio?. By first calculating the determinant of the H matrices?. Or we just simply use the %logdet of the VAR as a good estimation of the average variance matrix Σ.

Thanks for your help.

[TomDoan]

Use %LOGL. %LOGDET only has a 1-1 mapping with the log likelihood for VAR's and similar models (with a covariance matrix fixed over time)---it's not defined by GARCH because it has no meaning with a GARCH model.