The RATS Software Forum

Dear Sir/Madam,

I got few questions and not sure whether could be given some kind lights from you, thanks.

1、In var-dcc-garch model, the coefficient DCC(B) of outcome is extremely close to 1 which is 0.99**, is that okay? Though DCC(A) + DCC(B) still less than 1 but that made me a little afraid wondering whether this indicate somewhere can be improved.

2、In var-bekk-garch model, though increase the iterations to a very large number, it still can not be converged; is it the only solution to change the model?

3、In var-bekk-garch model, the first and second variable belong one area and the rest of the variables belong to another area, I want to test whether there is variance spillover effect among area1 and area2, can I implement Wald test just within one step for one direction? (that might be taking all the related coefficients from area1 to area2 into the calculation of Wald test)

system(model=bekkvar)
variables rsh rsz rdji rixic rftse rn225
lags 1 to 2
det constant
end(system)
garch(p=1,q=1,ITERATIONS=1000000,SUBITERATIONS=1000000,model=bekkvar,mv=bekk,pmethod=simplex,piters=10,method=bhhh,stdresids=bekkresids)
@mvqstat(lags=20)
# bekkresids
@mvarchtest(lags=5)
# bekkresids

system(model=dccvar)
variables rsh rsz rdji rixic rftse rn225
lags 1 to 4
det constant
end(system)
garch(p=1,q=1,ITERATIONS=200,SUBITERATIONS=1000,model=dccvar,mv=dcc,pmethod=simplex,piters=10,method=bhhh,stdresids=dccresids,HMATRICES=hma)
@mvqstat(lags=20)
# dccresids
@mvarchtest(lags=5)
# dccresids

Best Regards
Bill

yangdashan wrote:Dear Sir/Madam,

I got few questions and not sure whether could be given some kind lights from you, thanks.

1、In var-dcc-garch model, the coefficient DCC(B) of outcome is extremely close to 1 which is 0.99**, is that okay? Though DCC(A) + DCC(B) still less than 1 but that made me a little afraid wondering whether this indicate somewhere can be improved.

That's not at all uncommon. Note that if the corrrelation is fairly constant, DCC(B) needs to be near one.

yangdashan wrote: 2、In var-bekk-garch model, though increase the iterations to a very large number, it still can not be converged; is it the only solution to change the model?

BEKK rarely works above 4 variables.

yangdashan wrote: 3、In var-bekk-garch model, the first and second variable belong one area and the rest of the variables belong to another area, I want to test whether there is variance spillover effect among area1 and area2, can I implement Wald test just within one step for one direction? (that might be taking all the related coefficients from area1 to area2 into the calculation of Wald test)

While that's testable by a block Wald test, you might want to read the article from the April 2019 newsletter:

https://estima.com/newslett/Apr2019RATS ... pdf#page=3

In BEKK models the "spillover" is sometimes just an artifact of how BEKK handles the covariances.

Dear Tom,

'BEKK rarely works above 4 variables'. Does that because over-parameterisation?

Thanks
Bill

yangdashan wrote: Dear Tom,

'BEKK rarely works above 4 variables'. Does that because over-parameterisation?

Thanks
Bill

Not so much the total number of parameters as that the connection between the parameters and what they're supposed to produce (the covariance matrix) gets increasingly weak as the number of series increases.