7. GARCH SPECIFICATION
Specification of mean equation – if correctly
specified, there should be no autocorrelation of
the standardized residuals. [REFER TO ARIMA
MODELING]
Variance equation – research has shown that
GARCH(1, 1) is an adequate specification.
However, an information criteria can be used to
choose (p, q) of GARCH(p, q).
In this case, the squared standardized
residuals should not be autocorrelated or the
standardized residuals do not exhibit additional
ARCH.
8. GARCH SPECIFICATION
BASIC GARCH(1, 1)
Squared residuals (ARCH term) – news about
volatility from the period period
GARCH term (h) – last period’s variance
tttY εµ +=
tttt
titt
uhh
hI
+++= −−
−
1
2
1
),0(~|
ϕφεα
ε
9. GARCH SPECIFICATION
BASIC GARCH(1, 1) – M
The mean equation has conditional variance as a regressor
(use: risk-return tradeoff, inflation-inflation uncertainty
tradeoff).
In certain case, the standard deviation is used instead.
Other variables can also be included in the mean equation.
tttt hY εθµ ++=
tttt
titt
uhh
hI
+++= −−
−
1
2
1
),0(~|
ϕφεα
ε
10. GARCH SPECIFICATION
TARCH (1, 1) – Threshold ARCH
This is asymmetric GARCH introduced by Zakoian (1990)
and Glosten et al. (1993) to capture the observation that
downward movements in the market are followed by higher
volatilities.
It capture volatility response to good and bad news.
If θ > 0, we say that the LEVERAGE EFFECT exists.
ttttY εµ +=
otherwise0and,0if1
),0(~|
11
2
1
2
1
<=
++++= −−−−
−
tt
tttttt
titt
d
uhdh
hI
ε
ϕθεφεα
ε
11. GARCH SPECIFICATION
EGARCH
This is asymmetric GARCH introduced proposed
by Nelson (1991)
If φ < 0, we say that the LEVERAGE EFFECT
exists.
tt
t
t
t
t
t uhh ++++= −
−
−
−
−
1
1
1
1
1
loglog ϕ
σ
ε
θ
σ
ε
φα
12. GARCH ESTIMATION-
EVALUATION - INFERENCE
Estimation of GARCH models is done using
Maximum Likelihood Estimation
Diagnostic tests – most important is
autocorrelation of standardized residuals, squared
standardized residuals, and ARCH tests. This is
done using VIEW, Residual Tests.
Once satisfied, the results can be interpreted and
measures of volatility can be obtained.
EXERCISE: OBTAIN A GARCH MODEL OF
A VARIABLE OF YOUR INTEREST