Professional Documents
Culture Documents
i=1
The estimated ARCH models (variance equations) for monthly and quarterly data are given
below as, ARCH(5,0):
GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*RESID(-2)^2 + C(5)*RESID(-3)^2 +
C(6)*RESID(-4)^2 + C(7)*RESID(-5)^2
GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*RESID(-2)^2 + C(5)*RESID(-3)^2 +
C(6)*RESID(-4)^2 + C(7)*RESID(-5)^2
The above variance equations contain five arch terms and zero garch term for each. The AIC and
SIC values are reported in the last section.
GARCH Model (Generalized ARCH)
A disadvantage that an ARCH model has is of assigning equal weights to all squared residuals.
This assumption of equal weights seems unattractive because one would suppose that most
recent events would be most relevant and should be given higher weights. Therefore, a useful
generalization of ARCH model is GARCH model which also assigns weights to past squared
residuals, but those are declining weights which never reach completely to zero. The most widely
GARCH specification, GARCH(1, 1), shows that the next period variance is best predicted by a
weighted average of long run average variance. GARCH (1, 1) model is given as follows,
σt2 = βVL + αiεt-12 + ρσt-12
We have estimated the following GARCH(1, 1) models as,
GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*GARCH(-1)
GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*GARCH(-1)
Component ARCH/GARCH
Long-term volatility dependencies are an important feature of variance which are unaccounted
for by the GARCH/ARCH models. Hence, component GARCH model is developed to accurately
measure the long-term volatility dependencies. The component GARCH model is written as,
(σt2 - σ2) = α (εt-12 - σ2) + β (σt-i2 - σ2)
Here the component ARCH (1, 1) is estimated including threshold term as:
Q = C(2) + C(3)*(Q(-1) - C(2)) + C(4)*(RESID(-1)^2 - GARCH(-1))
GARCH = Q + (C(5) + C(6)*(RESID(-1)<0))*(RESID(-1)^2 - Q(-1)) +C(7)*(GARCH(-1) - Q(-
1))
992.130
RESID(-1)^2 3 GARCH(-1) -991.1303