ARMA Autocorrelation Functions

For a moving average process, MA(q):

j=0

j
w
t+hj
_
_
_
_
q

k=0

k
w
tk
_
_
_
_
=
_

2
w
qh

j=0

j+h
, 0 h q
0 h > q.
1
So the ACF is
(h) =
_

_
qh

j=0

j+h
q

j=0

2
j
, 0 h q
0 h > q.
Notes:
In these expressions,
0
= 1 for convenience.
(q) = 0 but (h) = 0 for h > q. This characterizes
MA(q).
2
For an autoregressive process, AR(p):
x
t
=
1
x
t1
+
2
x
t2
+ +
p
x
tp
+w
t
.
So
(h) = cov
_
x
t+h
, x
t
_
= E
_
_
_
_
p

j=1

j
x
t+hj
+w
t+h
_
_
x
t
_
_
=
p

j=1

j
(h j) +cov
_
w
t+h
, x
t
_
.
3
Because x
t
is causal, x
t
is w
t
+ a linear combination of w
t1
, w
t2
, . . . .
So
cov
_
w
t+h
, x
t
_
=
_
_
_

2
w
h = 0
0 h > 0.
Hence
(h) =
p

j=1

j
(h j), h > 0
and
(0) =
p

j=1

j
(j) +
2
w
.
4
If we know the parameters
1
,
2
, . . . ,
p
and
2
w
, these equa-
tions for h = 0 and h = 1, 2, . . . , p form p +1 linear equations
in the p +1 unknowns (0), (1), . . . , (p).
The other autocovariances can then be found recursively
from the equation for h > p.
Alternatively, if we know (or have estimated) (0), (1), . . . , (p),
they form p + 1 linear equations in the p + 1 parameters

1
,
2
, . . . ,
p
and
2
w
.
These are the Yule-Walker equations.
5
For the ARMA(p, q) model with p > 0 and q > 0:
x
t
=
1
x
t1
+
2
x
t2
+ +
p
x
tp
+w
t
+
1
w
t1
+
2
w
t2
+ +
q
w
tq
,
a generalized set of Yule-Walker equations must be used.
The moving average models ARMA(0, q) = MA(q) are the
only ones with a closed form expression for (h).
For AR(p) and ARMA(p, q) with p > 0, the recursive equation
means that for h > max(p, q + 1), (h) is a sum of geomet-
rically decaying terms, possibly damped oscillations.
6
The recursive equation is
(h) =
p

j=1

j
(h j), h > q.
What kinds of sequences satisfy an equation like this?
Try (h) = z
h
for some constant z.
The equation becomes
0 = z
h

j=1

j
z
(hj)
= z
h
_
_
1
p

j=1

j
z
j
_
_
= z
h
(z).
7
So if (z) = 0, then (h) = z
h
satises the equation.
Since (z) is a polynomial of degree p, there are p solutions,
say z
1
, z
2
, . . . , z
p
.
So a more general solution is
(h) =
p

l=1
c
l
z
h
l
,
for any constants c
1
, c
2
, . . . , c
p
.
If z
1
, z
2
, . . . , z
p
are distinct, this is the most general solution;
if some roots are repeated, the general form is a little more
complicated.
8
If all z
1
, z
2
, . . . , z
p
are real, this is a sum of geometrically
decaying terms.
If any root is complex, its complex conjugate must also be a
root, and these two terms may be combined into geometri-
cally decaying sine-cosine terms.
The constants c
1
, c
2
, . . . , c
p
are determined by initial condi-
tions; in the ARMA case, these are the Yule-Walker equa-
tions.
Note that the various rates of decay are the zeros of (z),
the autoregressive operator, and do not depend on (z), the
moving average operator.
9
Example: ARMA(1, 1)
x
t
= x
t1
+w
t1
+w
t
.
The recursion is
(h) = (h 1), h = 2, 3, . . .
So (h) = c
h
for h = 1, 2, . . . , but c = 1.
Graphically, the ACF decays geometrically, but with a dier-
ent value at h = 0.
10
5 10 15 20 25
0
.
2
0
.
4
0
.
6
0
.
8
1
.
0
Index
A
R
M
A
a
c
f
(
a
r

=

0
.
9
,

m
a

=

0
.
5
,

2
4
)
11
The Partial Autocorrelation Function
An MA(q) can be identied from its ACF: non-zero to lag q,
and zero afterwards.
We need a similar tool for AR(p).
The partial autocorrelation function (PACF) lls that role.
12
Recall: for multivariate random variables X, Y, Z, the partial
correlations of X and Y given Z are the correlations of:
the residuals of X from its regression on Z; and
the residuals of Y from its regression on Z.
Here regression means conditional expectation, or best lin-
ear prediction, based on population distributions, not a sam-
ple calculation.
In a time series, the partial autocorrelations are dened as

h,h
= partial correlation of x
t+h
and x
t
given x
t+h1
, x
t+h2
, . . . , x
t+1
.
13
For an autoregressive process, AR(p):
x
t
=
1
x
t1
+
2
x
t2
+ +
p
x
tp
+w
t
,
If h > p, the regression of x
t+h
on x
t+h1
, x
t+h2
, . . . , x
t+1
is

1
x
t+h1
+
2
x
t+h2
+ +
p
x
t+hp
So the residual is just w
t+h
, which is uncorrelated with
x
t+h1
, x
t+h2
, . . . , x
t+1
and x
t
.
14
So the partial autocorrelation is zero for h > p:

h,h
= 0, h > p.
We can also show that
p,p
=
p
, which is non-zero by as-
sumption.
So
p,p
= 0 but
h,h
= 0 for h > p. This characterizes AR(p).
15
The Inverse Autocorrelation Function
SASs proc arima also shows the Inverse Autocorrelation Func-
tion (IACF).
The IACF of the ARMA(p, q) model
(B)x
t
= (B)w
t
is dened to be the ACF of the inverse (or dual) process
(B)x
(inverse)
t
= (B)w
t
.
The IACF has the same property as the PACF: AR(p) is
characterized by an IACF that is nonzero at lag p but zero
for larger lags.
16
Summary: Identication of ARMA processes
AR(p) is characterized by a PACF or IACF that is:
nonzero at lag p;
zero for lags larger than p.
MA(q) is characterized by an ACF that is:
nonzero at lag q;
zero for lags larger than q.
For anything else, try ARMA(p, q) with p > 0 and q > 0.
17
For p > 0 and q > 0:
AR(p) MA(q) ARMA(p, q)
ACF Tails o Cuts o after lag q Tails o
PACF Cuts o after lag p Tails o Tails o
IACF Cuts o after lag p Tails o Tails o
Note: these characteristics are used to guide the initial choice
of a model; estimation and model-checking will often lead to
a dierent model.
18
Other ARMA Identication Techniques
SASs proc arima oers the MINIC option on the identify
statement, which produces a table of SBC criteria for various
values of p and q.
The identify statement has two other options: ESACF and
SCAN.
Both produce tables in which the pattern of zero and non-
zero values characterize p and q.
See Section 3.4.10 in Brocklebank and Dickey.
19
options linesize = 80;
ods html file = varve3.html;
data varve;
infile ../data/varve.dat;
input varve;
lv = log(varve);
run;
proc arima data = varve;
title Use identify options to identify a good model;
identify var = lv(1) minic esacf scan;
estimate q = 1 method = ml;
estimate q = 2 method = ml;
estimate p = 1 q = 1 method = ml;
run;
proc arima output