The RATS Software Forum

Hi,

I ran the model VAR-GARCH MODEL:

@varlagselect(crit=sbc,lags=13)
# inf gdp

system(model=vargarchm)
variables gdp inf
lags 1 to 12
det constant sqrth(1) sqrth(2)
end(system)

dec vect[series] sqrth(2)

I attach the output.

clear(zeros) sqrth

garch(model=vargarchm,p=1,q=1,mv=bekk,hmatrices=hh,$
hadjust=%pt(sqrth,t,%sqrt(%xdiag(hh(t)))), rvectors=rd,$
stdresids=rstd, rseries=rs,mvhseries=hhs)

This model gave me the hoped results. Also, the model passes the test on univariate standardized residuals and the multivariate test (mvqstat and mvarchtest). I doubt about the hadjust part, I mean, how I estimate the square root of the variance. Could you tell me if it's ok? What does mean %pt, %xdiag, etc....? The varlag procedure suggest 12 lags and I ran the model with 12 lags in mean.

The model is a VAR-GARCH-M with square of the variance. I don't use maximize.

That seems like a lot of lags, particularly if SBC is giving you that. You're doing log of GDP?

Yes, I'm using log GDP. the @varlagselect procedure suggest me 12 lags. Also, I tested for serial correlation and arch effects.