Box-Jenkins Analysis of ARMA(p,q) Models
Eric ZivotApril 7, 2011
Box-Jenkins Modeling Strategy for Fitting ARMA(
p,q
)
Models1. Transform the data, if necessary, so that the assumption of covariancestationarity is a reasonable one2. Make an initial guess for the values of
p
and
q
3. Estimate the parameters of the proposed ARMA(
p,q
)
model4. Perform diagnostic analysis to con
fi
rm that the proposed model adequatelydescribes the data (e.g. examine residuals from
fi
tted model)
Identi
fi
cation of Stationary ARMA(p,q) ProcessesIntuition: The mean, variance, and autocorrelations de
fi
ne the properties of an ARMA(p,q) model. A natural way to identify an ARMA model is to matchthe pattern of the observed (sample) autocorrelations with the patterns of thetheoretical autocorrelations of a particular ARMA(
p,q
)
model.Sample autocovariances/autocorrelations
ˆ
γ
j
= 1
T
T
X
t
=
j
+1
(
y
t
−
ˆ
μ
)(
y
t
−
j
−
ˆ
μ
)
,
ˆ
μ
= 1
T
T
X
t
=1
y
t
ˆ
ρ
j
=ˆ
γ
j
ˆ
γ
0
Sample autocorrelation function (SACF)/correlogramplot
ˆ
ρ
j
vs.
j
Result: If
Y
t
∼
WN
(0
,σ
2
)
then
ρ
j
= 0
for all
j
and
√
T
ˆ
ρ
jd
→
N
(0
,
1)
so that
avar
(ˆ
ρ
j
) = 1
T
Therefore, a simple
t
−
statistic for
H
0
:
ρ
j
= 0
is
√
T
ˆ
ρ
j
and we reject
H
0
:
ρ
j
= 0
if
|
ˆ
ρ
j
|
>
1
.
96
√
T
Remark:
ˆ
ρ
1
,...,
ˆ
ρ
k
are asymptotically independent:
√
T
ˆ
ρ
kd
→
N
(
0
,
I
k
)
where
ˆ
ρ
k
= (ˆ
ρ
1
,...,
ˆ
ρ
k
)
0
.