GARCH with ARMA mean equation

[Farid]

Hi. Could you please look and see what is wrong in the following code? It gives an error in both estimation and rolling process:

Code: Select all

OPEN DATA "C:\Documents and Settings\admin\Desktop\FINAL - Daily.RAT"
CALENDAR(D) 2006:1:4
DATA(FORMAT=RATS) 2006:01:04 2010:11:26 RT

*** Estimating GARCH
garch(p=1,q=1,reg,resids=u,hseries=h) 2 506 rt
# constant rt{1} u{1}

***Rolling GARCH
compute width=505
dec vect[series] coeffs(6)

do gend=width,1278

   if gend==width
      garch(p=1,q=1,reg,resids=u,hseries=h,noprint) gend-width+2 gend rt
      # constant rt{1} u{1}

   else
      garch(p=1,q=1,reg,resids=u,hseries=h,initial=%beta,hessian=%xx,noprint) gend-width+2 gend rt
      # constant rt{1} u{1}

   compute %pt(coeffs,gend,%beta)

end do gend

In both cases the error is: "## SR10. Missing Values And/Or SMPL Options Leave No Usable Data Points". I can't understand why that happens.

Thanks in advance.

[TomDoan]

Do

SET U = 0.0

before the GARCH instruction. The GARCH defines U from entry 2, but you need some value for U{1} when T=2.

[Farid]

Thanks Tom. Is it a correct way to estimate and forecast ARMA-GARCH? The reason why I'm asking - because now I did what you said, but the program is too slow in implementing the instruction and it's even frozen.

[TomDoan]

You should be using %MVGAVGE{1} (not U{1}) in the regressor lists. You may be running into numerical problems by estimating an ARMA(1,1) on a series which might be closer to white noise. If it is WN, the ARMA parameters aren't identified. You should probably have either the AR or the MA but not both.