The RATS Software Forum

Hello,

I am trying to interpret my multivariate BEKK model results in the context of the variance and covariance equations for 4 price return series. For example, the variance equation for one series and the covariance equation between series 1 and 2 can be represented by the attachment I have included in this post. My question is: How are the ARCH/GARCH/Constant estimates in the BEKK output presented? Are they squared? Not squared? I may need to manipulate some of these estimated parameters to get the "economic" interpretation and correct statistical significance.

Thank you for the help!


Curtis

The variance constants gets increasingly complicated as you get to the higher row numbers as it's a lower triangular matrix times its transpose. So the sigma22 will be c(2,1)^2+c(2,2)^2, sigma33 is c(3,1)^2+c(3,2)^2+c(3,3)^2, ...

However, it's been my general impression that these expansions of BEKK are a waste of paper.

Hello Tom,

Thanks for the feedback on the variance constants. For the ARCH and GARCH parameters, can the parameters themselves (ex. A(1,3)) be interpreted as spillovers? Or do I need to square them or perform an interaction with the other terms first?

Thanks,

Curtis

cmcknigh wrote:Hello Tom,

Thanks for the feedback on the variance constants. For the ARCH and GARCH parameters, can the parameters themselves (ex. A(1,3)) be interpreted as spillovers? Or do I need to square them or perform an interaction with the other terms first?

Thanks,

Curtis

That's debatable. See

https://estima.com/newslett/Apr2019RATS ... pdf#page=3

Particularly in a 4 variable model (as opposed to a 2 variable model) each of those coefficients is entering into so many different elements of the covariance matrix that it's effectively impossible to interpret any bidirectional effect in isolation.