Rolling AR(5)-GARCH(1,1)

[ac_1]

Tom,

Am running a rolling AR(5)-GARCH(1,1) model with moving window as:

Code: Select all

*  AR(5)-GARCH(1,1)
*
clear h2
clear yeq1
*
compute gstart1=7,gend=1306
*
do j=0,2932
garch(p=1,q=1, method=bfgh, hseries=h2,resids=u2,print,regressors) gstart1+j gend+j y
# constant y{1 to 5}
set uu2 = u2**2
com ym0=%beta(1), ym1=%beta(2), ym2=%beta(3), ym3=%beta(4), ym4=%beta(5), ym5=%beta(6)
com vc=%beta(7), vb=%beta(8), va=%beta(9)
frml heq h2 = vc + vb*h2{1} + va*uu2{1}
frml ueq uu2 = h2
frml yeq yeq1 = ym0 + ym1*y{1} + ym2*y{2} + ym3*y{3} + ym4*y{4} + ym5*y{5}
group garchmod2 heq>>h2 ueq>>uu2 yeq>>yeq1
forecast(model=garchmod2, from=gend+1+j, to=gend+1+j)
end do j
*
smpl
prin / h2 yeq1
*

The code is similar as in Brooks(2008) and I think is okay.

I am interested in both mean forecasts and vol forecasts. But occasionally the parameters of the GARCH(1,1) 'blow-up' with some turning negative and not within certain bounds on various iterations. How do I deal with this ?

[TomDoan]

See section 11.2 of the (version User's Guide. We have suggestions there for dealing with rolling GARCH models.

[ac_1]

See section 11.2 of the (version User's Guide. We have suggestions there for dealing with rolling GARCH models.

Okay thanks for the prompt reply. The section refers to the rollgarch.rpf code that is with RATS.

So I need to use the INITIAL and HESSIAN options to feed the previous estimates in, but handle the first estimate individually.