testing for the I in IGARCH

[uli]

Hi all.

I find hardly any literature on tests for integration in GARCH models. Specifically, I want to test if the assumption (a+b)=1 can be rejected in the GARCH(1,1) model

h(t) = c + a * u(t-1)**2 + b * h(t-1).

What I currently do is a simple LR test, i.e. -2.0*(%logl_restricted - %logl_unrestricted), which I assume to be asymtotically Chi**2(1) distributed under the null hypothesis. Has anybody any idea on some other test? Or on the validity and power of of my test?

Many thanks in advance for any comments.

Uli

[TomDoan]

For a standard GARCH model, so far as I know you can use a standard LR test for I-ness. The unit root boundary has completely different implications in the variance than it does in the mean for an AR model.