component GARCH

[albeta456]

Tom,

Can you post the program codes of component GARCH of Engle, R. F., and G. G. J. Lee (1999): “A Long-Run and Short-Run
Component Model of Stock Return Volatility,” in Cointegration, Causality, and Forecasting, ed. by Engle, and White. Oxford University Press.

I tried the following:

frml hf = h1+alpha*(uu{1}-h1{1})+beta*(h{1}-h1{1})
frml h1f = omega+rho*h1{1}+phi*(uu{1}-h{1})
frml logl = h=hf,h1=h1f,u=r,uu=u^2,%logdensity(h,u)
maximize(parmset=meanparms+garchparms,pmethod=simplex,piters=15,method=bfgs,iters=300) logl 2 gend

where h1 is the permanent component.

But it didn't work.
Thanks.

[TomDoan]

I tried the following:

frml hf = h1+alpha*(uu{1}-h1{1})+beta*(h{1}-h1{1})
frml h1f = omega+rho*h1{1}+phi*(uu{1}-h{1})
frml logl = h=hf,h1=h1f,u=r,uu=u^2,%logdensity(h,u)
maximize(parmset=meanparms+garchparms,pmethod=simplex,piters=15,method=bfgs,iters=300) logl 2 gend

where h1 is the permanent component.

But it didn't work.
Thanks.

In FRML LOGL, you need to switch the order of H1=H1F and H=HF. H depends upon current H1, but H1 depends only upon lagged H. You also need to make sure that you have pre-sample (i.e. entry 1) values for UU, H and H1. It looks like the unconditional variance would be good for all of those.