BEKK starting values

[mariachri]

Hi,

I want to estimate a VAR - BEKK model with two variables and I am confused on the starting values I should use. I see that eviews uses the starting values from an OLS estimation for the mean equation, although I cannot understand what values does it use for the variance equation.

I also tried to use as starting values the estimates from the two univariate GARCH models (for the two variables). Of course, in this case the variance may be nonstationary and the starting value for the BEKK be greater than 1. Is this possible?

I have also seen some other values like 0.01 or 0.5 in some codes.

Are there "correct starting values"? Because in some cases my results change. How can I be sure that I have the correct ones and that I have a global maximum?

Thank you in advance.

[trubador]

In multivariate modeling, using the results from univariate counterparts can be considered as an appropriate approach. Though it has no guarantee that the algorithm will converge and you'll end up with the global optima. You should try multistart, which is the fancy way of saying that you should run the model with different starting values repeatedly and see if you can arrive at a better local optima. And you can always check the parameter domain/constraints to make sure that results actually make sense...

[mariachri]

Thank you for your reply. I will try different values.