Solutions to selected problems in

Brockwell and Davis

Anna Carlsund Henrik Hult
Spring 2003
This document contains solutions to selected problems in

Peter J. Brockwell and Richard A. Davis, Introduction to Time Series and Fore-
casting, 2nd Edition, Springer New York, 2002.

We provide solutions to most of the problems in the book that are not computer
exercises. That is, you will not need a computer to solve these problems. We en-
courage students to come up with suggestions to improve the solutions and to report
any misprints that may be found.

Contents
Chapter 1 1.1, 1.4, 1.5, 1.8, 1.11, 1.15 3

Chapter 2 2.1, 2.4, 2.8, 2.11, 2.15 8

Chapter 3 3.1, 3.4, 3.6, 3.7, 3.11 11

Chapter 4 4.4, 4.5, 4.6, 4.9, 4.10 14

Chapter 5 5.1, 5.3, 5.4, 5.11 19

Chapter 6 6.5, 6.6 23

Chapter 7 7.1, 7.5 25

Chapter 8 8.7, 8.9, 8.13, 8.14, 8.15 28

Chapter 10 10.5 31

Notation: We will use the following notation.

• The indicator function
Ω
1 if h 2 A,
1A (h) =
0 if h 2
/ A.

• Dirac’s delta function

Ω Z 1
+1 if t = 0,
±(t) = and f (t)±(t)dt = f (0).
0 if t =
6 0, °1
Chapter 1
Problem 1.1. a) First note that
E[(Y ° c)2 ] = E[Y 2 ° 2Y c + c2 ] = E[Y 2 ] ° 2cE[Y ] + c2
= E[Y 2 ] ° 2cµ + c2 .
Find the extreme point by differentiating,
d
(E[Y 2 ] ° 2cµ + c2 ) = °2µ + 2c = 0 ) c = µ.
dc
2
Since, dc
d
2 (E[Y ] ° 2cµ + c ) = 2 > 0 this is a min-point.
2 2

b) We have
E[(Y ° f (X))2 | X] = E[Y 2 ° 2Y f (X) + f 2 (X) | X]
= E[Y 2 | X] ° 2f (X)E[Y | X] + f 2 (X),
which is minimized by f (X) = E[Y | X] (take c = f (X) and µ = E[Y | X] in a).
c) We have £ §
E[(Y ° f (X))2 ] = E E[(Y ° f (X))2 | X] ,
so the result follows from b).
Problem 1.4. a) For the mean we have
µX (t) = E[a + bZt + cZt°2 ] = a,
and for the autocovariance
∞X (t + h, t) = Cov(Xt+h , Xt ) = Cov(a + bZt+h + cZt+h°2 , a + bZt + cZt°2 )
= b2 Cov(Zt+h , Zt ) + bc Cov(Zt+h , Zt°2 )
+ cb Cov(Zt+h°2 , Zt ) + c2 Cov(Zt+h°2 , Zt°2 )
= æ 2 b2 1{0} (h) + æ 2 bc1{°2} (h) + æ 2 cb1{2} (h) + æ 2 c2 1{0} (h)
8 2
< (b + c2 )æ 2 if h = 0,
= bcæ 2 if |h| = 2,
:
0 otherwise.
Since µX (t) and ∞X (t + h, t) do not depend on t, {Xt : t 2 Z} is (weakly) stationary.
b) For the mean we have
µX (t) = E[Z1 ] cos(ct) + E[Z2 ] sin(ct) = 0,
and for the autocovariance
∞X (t + h, t) = Cov(Xt+h , Xt )
= Cov(Z1 cos(c(t + h)) + Z2 sin(c(t + h)), Z1 cos(ct) + Z2 sin(ct))
= cos(c(t + h)) cos(ct) Cov(Z1 , Z1 ) + cos(c(t + h)) sin(ct) Cov(Z1 , Z2 )
+ sin(c(t + h)) cos(ct) Cov(Z1 , Z2 ) + sin(c(t + h)) sin(ct) Cov(Z2 , Z2 )
= æ 2 (cos(c(t + h)) cos(ct) + sin(c(t + h)) sin(ct))
= æ 2 cos(ch)
where the last equality follows since cos(Æ ° Ø) = cos Æ cos Ø + sin Æ sin Ø. Since
µX (t) and ∞X (t + h, t) do not depend on t, {Xt : t 2 Z} is (weakly) stationary.
c) For the mean we have
µX (t) = E[Zt ] cos(ct) + E[Zt°1 ] sin(ct) = 0,

3
and for the autocovariance
∞X (t + h, t) = Cov(Xt+h , Xt )
= Cov(Zt+h cos(c(t + h)) + Zt+h°1 sin(c(t + h)), Zt cos(ct) + Zt°1 sin(ct))
= cos(c(t + h)) cos(ct) Cov(Zt+h , Zt ) + cos(c(t + h)) sin(ct) Cov(Zt+h , Zt°1 )
+ sin(c(t + h)) cos(ct) Cov(Zt+h°1 , Zt )
+ sin(c(t + h)) sin(ct) Cov(Zt+h°1 , Zt°1 )
= æ 2 cos2 (ct)1{0} (h) + æ 2 cos(c(t ° 1)) sin(ct)1{°1} (h)
+ æ 2 sin(c(t + 1)) cos(ct)1{1} (h) + æ 2 sin2 (ct)1{0} (h)
8 2
< æ cos2 (ct) + æ 2 sin2 (ct) = æ 2 if h = 0,
= æ 2 cos(c(t ° 1)) sin(ct) if h = °1,
: 2
æ cos(ct) sin(c(t + 1)) if h = 1,
We have that {Xt : t 2 Z} is (weakly) stationary for c = ±kº, k 2 Z, since then
∞X (t + h, t) = æ 2 1{0} (h). For c 6= ±kº, k 2 Z, {Xt : t 2 Z} is not (weakly)
stationary since ∞X (t + h, t) depends on t.
d) For the mean we have
µX (t) = E[a + bZ0 ] = a,
and for the autocovariance
∞X (t + h, t) = Cov(Xt+h , Xt ) = Cov(a + bZ0 , a + bZ0 ) = b2 Cov(Z0 , Z0 ) = æ 2 b2 .
Since µX (t) and ∞X (t + h, t) do not depend on t, {Xt : t 2 Z} is (weakly) stationary.
e) If c = kº, k 2 Z then Xt = (°1)kt Z0 which implies that Xt is weakly stationary
when c = kº. For c 6= kº we have
µX (t) = E[Z0 ] cos(ct) = 0,
and for the autocovariance
∞X (t + h, t) = Cov(Xt+h , Xt ) = Cov(Z0 cos(c(t + h)), Z0 cos(ct))
= cos(c(t + h)) cos(ct) Cov(Z0 , Z0 ) = cos(c(t + h)) cos(ct)æ 2 .
The process {Xt : t 2 Z} is (weakly) stationary when c = ±kº, k 2 Z and not
(weakly) stationary when c 6= ±kº, k 2 Z, see 1.4. c).
f) For the mean we have
µX (t) = E[Zt Zt°1 ] = 0,
and
∞X (t + h, t) = Cov(Xt+h , Xt ) = Cov(Zt+h Zt+h°1 , Zt Zt°1 )
Ω 4
æ if h = 0,
= E[Zt+h Zt+h°1 Zt Zt°1 ] =
0 otherwise.
Since µX (t) and ∞X (t + h, t) do not depend on t, {Xt : t 2 Z} is (weakly) stationary.
Problem 1.5. a) We have
∞X (t + h, t) = Cov(Xt+h , Xt ) = Cov(Zt+h + µZt+h°2 , Zt + µZt°2 )
= Cov(Zt+h , Zt ) + µ Cov(Zt+h , Zt°2 ) + µ Cov(Zt+h°2 , Zt )
+ µ2 Cov(Zt+h°2 , Zt°2 )
= 1{0} (h) + µ1{°2} (h) + µ1{2} (h) + µ2 1{0} (h)
Ω Ω
1 + µ2 if h = 0, 1.64 if h = 0,
= =
µ if |h| = 2. 0.8 if |h| = 2.

4
Hence the ACVF depends only on h and we write ∞X (h) = ∞X (t + h, h). The ACF
is then
Ω
∞X (h) 1 if h = 0,
Ω(h) = =
∞X (0) 0.8/1.64 º 0.49 if |h| = 2.
b) We have
µ ∂
1 1
Var (X1 + X2 + X3 + X4 ) = Var(X1 + X2 + X3 + X4 )
4 16
1 ≥
= Var(X1 ) + Var(X2 ) + Var(X3 ) + Var(X4 ) + 2 Cov(X1 , X3 )
16 ¥
+ 2 Cov(X2 , X4 )
1° ¢ 1° ¢ 1.64 + 0.8
= 4∞X (0) + 4∞X (2) = ∞X (0) + ∞X (2) = = 0.61.
16 4 4
c) µ = °0.8 implies ∞X (h) = °0.8 for |h| = 2 so
µ ∂
1 1.64 ° 0.8
Var (X1 + X2 + X3 + X4 ) = = 0.21.
4 4
Because of the negative covariance at lag 2 the variance in c) is considerably smaller.
Problem 1.8. a) First we show that {Xt : t 2 Z} is WN (0, 1). For t even we have
E[Xt ] = E[Zt ] = 0 and for t odd
∑ 2 ∏
Zt°1 ° 1 1
E[Xt ] = E p = p E[Zt°1
2
° 1] = 0.
2 2
Next we compute the ACVF. If t is even we have ∞X (t, t) = E[Zt2 ] = 1 and if t is
odd
"µ ∂2 #
2
Zt°1 °1 1 1
∞X (t, t) = E p = E[Zt°1
4
° 2Zt°1
2
+ 1] = (3 ° 2 + 1) = 1.
2 2 2

If t is even we have
∑ ∏
Zt2 ° 1 1
∞X (t + 1, t) = E p Zt = p E[Zt3 ° Zt ] = 0,
2 2
and if t is odd
∑ ∏ ∑ 2 ∏
Z2 ° 1 Z °1
∞X (t + 1, t) = E Zt+1 t°1
p = E[Zt+1 ]E t°1p = 0.
2 2
Clearly ∞X (t + h, t) = 0 for |h| ∏ 2. Hence
Ω
1 if h = 0,
∞X (t + h, h) =
0 otherwise.
Thus {Xt : t 2 Z} is WN (0, 1). If t is odd Xt and Xt°1 is obviously dependent so
{Xt : t 2 Z} is not IID (0, 1).
b) If n is odd
E[Xn+1 | X1 , . . . , Xn ] = E[Zn+1 | Z0 , Z2 , Z4 . . . , Zn°1 ] = E[Zn+1 ] = 0.
If n is even
∑ ∏
Zn2 ° 1 Z2 ° 1 X2 ° 1
E[Xn+1 | X1 , . . . , Xn ] = E p | Z0 , Z2 , Z4 , . . . , Zn = np = np .
2 2 2
This again shows that {Xt : t 2 Z} is not IID (0, 1).

5
Problem 1.11. a) Since aj = (2q + 1)°1 , °q ∑ j ∑ q, we have
q
X Xq
1
aj mt°j = (c0 + c1 (t ° j))
j=°q
2q + 1 j=°q
0 1 0 1
Xq Xq
1 @ c1 @
= c0 (2q + 1) + c1 (t ° j)A = c0 + t (2q + 1) ° jA
2q + 1 j=°q
2q +1 j=°q
0 1
X q Xq
c1 @
= c0 + c1 t ° j+ °j A
2q + 1 j=1 j=1

= c0 + c1 t = mt

b) We have
2 3
q
X q
X
E [At ] = E 4 aj Zt°j 5 = aj E [Zt°j ] = 0 and
j=°q j=°q
0 1
q
X q
X q
X
1 æ2
Var (At ) = Var @ aj Zt°j A = a2j Var (Zt°j ) = æ2 =
j=°q j=°q (2q + 1)
2
j=°q
2q + 1

We see that the variance Var(At ) is small for large q. Hence, the process At will be
close to its mean (which is zero) for large q.
Problem 1.15. a) Put

Zt = rr12 Xt = (1 ° B)(1 ° B 12 )Xt = (1 ° B)(Xt ° Xt°12 )

= Xt ° Xt°12 ° Xt°1 + Xt°13
= a + bt + st + Yt ° a ° b(t ° 12) ° st°12 ° Yt°12 ° a ° b(t ° 1) ° st°1 ° Yt°1
+ a + b(t ° 13) + st°13 + Yt°13
= Yt ° Yt°1 ° Yt°12 + Yt°13 .

We have µZ (t) = E[Zt ] = 0 and

∞Z (t + h, t) = Cov (Zt+h , Zt )
= Cov (Yt+h ° Yt+h°1 ° Yt+h°12 + Yt+h°13 , Yt ° Yt°1 ° Yt°12 + Yt°13 )
= ∞Y (h) ° ∞Y (h + 1) ° ∞Y (h + 12) + ∞Y (h + 13) ° ∞Y (h ° 1) + ∞Y (h)
+ ∞Y (h + 11) ° ∞Y (h + 12) ° ∞Y (h ° 12) + ∞Y (h ° 11)
+ ∞Y (h) ° ∞Y (h + 1) + ∞Y (h ° 13) ° ∞Y (h ° 12) ° ∞Y (h ° 1) + ∞Y (h)
= 4∞Y (h) ° 2∞Y (h + 1) ° 2∞Y (h ° 1) + ∞Y (h + 11) + ∞Y (h ° 11)
° 2∞Y (h + 12) ° 2∞Y (h ° 12) + ∞Y (h + 13) + ∞Y (h ° 13).

Since µZ (t) and ∞Z (t + h, t) do not depend on t, {Zt : t 2 Z} is (weakly) stationary.

b) We have Xt = (a + bt)st + Yt . Hence,

Zt = r212 Xt = (1 ° B 12 )(1 ° B 12 )Xt = (1 ° B 12 )(Xt ° Xt°12 )

= Xt ° Xt°12 ° Xt°12 + Xt°24 = Xt ° 2Xt°12 + Xt°24
= (a + bt)st + Yt ° 2(a + b(t ° 12)st°12 + Yt°12 ) + (a + b(t ° 24))st°24 + Yt°24
= a(st ° 2st°12 + st°24 ) + b(tst ° 2(t ° 12)st°12 + (t ° 24)st°24 )
+ Yt ° 2Yt°12 + Yt°24
= Yt ° 2Yt°12 + Yt°24 .

6
Now we have µZ (t) = E[Zt ] = 0 and

∞Z (t + h, t) = Cov (Zt+h , Zt )
= Cov (Yt+h ° 2Yt+h°12 + Yt+h°24 , Yt ° 2Yt°12 + Yt°24 )
= ∞Y (h) ° 2∞Y (h + 12) + ∞Y (h + 24) ° 2∞Y (h ° 12) + 4∞Y (h)
° 2∞Y (h + 12) + ∞Y (h ° 24) ° 2∞Y (h ° 12) + ∞Y (h)
= 6∞Y (h) ° 4∞Y (h + 12) ° 4∞Y (h ° 12) + ∞Y (h + 24) + ∞Y (h ° 24).

Since µZ (t) and ∞Z (t + h, t) do not depend on t, {Zt : t 2 Z} is (weakly) stationary.

7
Chapter 2
Problem 2.1. We find the best linear predictor X̂n+h = aXn + b of Xn+h by
finding a and b such that E[Xn+h ° X̂n+h ] = 0 and E[(Xn+h ° X̂n+h )Xn ] = 0. We
have

E[Xn+h ° X̂n+h ] = E[Xn+h ° aXn ° b] = E[Xn+h ] ° aE[Xn ] ° b = µ (1 ° a) ° b

and

E[(Xn+h ° X̂n+h )Xn ] = E[(Xn+h ° aXn ° b)Xn ]

= E[Xn+h Xn ] ° aE[Xn2 ] ° bE[Xn ]
= E[Xn+h Xn ] ° E[Xn+h ]E[Xn ] + E[Xn+h ]E[Xn ]
° ¢
° a E[Xn2 ] ° E[Xn ]2 + E[Xn ]2 ° bE[Xn ]
° ¢
= Cov(Xn+h , Xn ) + µ2 ° a Cov(Xn , Xn ) + µ2 ° bµ
° ¢
= ∞(h) + µ2 ° a ∞(0) + µ2 ° bµ,

which implies that

∞(h) + µ2 ° bµ
b = µ (1 ° a) , a= .
∞(0) + µ2

Solving this system of equations we get a = ∞(h)/∞(0) = Ω(h) and b = µ(1 ° Ω(h))
i.e. X̂n+h = Ω(h)Xn + µ(1 ° Ω(h)).

Problem 2.4. a) Put Xt = (°1)t Z where Z is random variable with E[Z] = 0 and
Var(Z) = 1. Then

∞X (t + h, t) = Cov((°1)t+h Z, (°1)t Z) = (°1)2t+h Cov(Z, Z) = (°1)h = cos(ºh).

b) Recall problem 1.4 b) where Xt = Z1 cos(ct) + Z2 sin(ct) implies that ∞X (h) =

cos(ch). If we let Z1 , Z2 , Z3 , Z4 , W be independent random variables with zero mean
and unit variance and put
≥º ¥ ≥º ¥ ≥º ¥ ≥º ¥
Xt = Z1 cos t + Z2 sin t + Z3 cos t + Z4 sin t + W.
2 2 4 4
Then we see that ∞X (h) = °∑(h). ¢
c) Let {Zt : t 2 Z} be WN 0, æ 2 and put Xt = Zt + µZt°1 . Then E[Xt ] = 0 and

∞X (t + h, t) = Cov(Zt+h + µZt+h°1 , Zt + µZt°1 )

= Cov(Zt+h , Zt ) + µ Cov(Zt+h , Zt°1 ) + µ Cov(Zt+h°1 , Zt )
+ µ2 Cov(Zt+h°1 , Zt°1 )
8 2
< æ (1 + µ2 ) if h = 0,
= æ2 µ if |h| = 1,
:
0 otherwise.

If we let æ 2 = 1/(1+µ2 ) and choose µ such that æ 2 µ = 0.4, then we get ∞X (h) = ∑(h).
Hence, we choose µ so that µ/(1 + µ2 ) = 0.4, which implies that µ = 1/2 or µ = 2.

Problem 2.8. Assume that there exists a stationary solution {Xt : t 2 Z} to

Xt = ¡Xt°1 + Zt , t = 0, ±1, . . .

8
 ° ¢
where {Zt : t 2 Z} ª WN 0, æ 2 and |¡1 | = 1. Use the recursions
n
X
Xt = ¡Xt°1 + Zt = ¡2 Xt°2 + ¡Zt°1 + Zt = . . . = ¡n+1 Xt°(n+1) + ¡i Zt°i ,
i=0

which yields that

n
X
Xt ° ¡ n+1
Xt°(n+1) = ¡i Zt°i .
i=0

We have that
√ n ! n n
X X X
Var ¡i Zt°i = ¡2i Var (Zt°i ) = æ 2 = (n + 1) æ 2 .
i=0 i=0 i=0

On the other side we have that

° ¢
Var Xt ° ¡n+1 Xt°(n+1) = 2∞(0) ° 2¡n+1 ∞(n + 1) ∑ 2∞(0) + 2∞(n + 1) ∑ 4∞(0).
This mean that (n + 1) æ 2 ∑ 4∞(0), 8n. Letting n ! 1 implies that ∞(0) = 1,
which is a contradiction, i.e. there exists no stationary solution.
Problem 2.11. We have that {Xt : t 2 Z} is an AR(1) process with mean µ so
{Xt : t 2 Z} satisfies
° ¢
Xt ° µ = ¡(Xt°1 ° µ) + Zt , {Zt : t 2 Z} ª WN 0, æ 2 ,
|h| 2
with ¡ = 0.6 and æ 2 = 2. Since {Xt : t 2 Z} is AR(1) we have that ∞X (h) = ¡1°¡æ2 .
Pn
We estimate µ by X n = n1 k=1 Xk . For large valuesP of n X n is approximately
normally distributed with mean µ and variance n1 |h|<1 ∞(h) (see Section 2.4 in
Brockwell and Davis). In our case the variance is
√ 1
! µ µ ∂∂
1 X æ2 1 1 æ2
1+2 ¡ h
= 1 + 2 ° 1
n 1 ° ¡2 n 1°¡ 1 ° ¡2
h=1
µ ∂ µ ∂
1 2 æ2 1 1+¡ æ2 æ2
= °1 = = .
n 1°¡ 1 ° ¡2 n 1 ° ¡ 1 ° ¡2 n(1 ° ¡)2
2
Hence, X n is approximately N (µ, n(1°¡)
æ
2 ). A 95% confidence interval is given by

I = (xn ° ∏0.025 pn(1°¡) , xn + ∏0.025 pn(1°¡)

æ æ
). Putting in the numeric values gives
I = 0.271 ± 0.69. Since 0 2 I the hypothesis that µ = 0 can not be rejected.
Problem 2.15. Let X̂n+1 = Pn Xn+1 = a0 + a1 Xn + · · · + an X1 . We may assume
that µX (t) = 0. Otherwise we can consider Yt = Xt ° µ. Let S(a0 , a1 , . . . , an ) =
E[(Xn+1 ° X̂n+1 )2 ] and minimize this w.r.t. a0 , a1 , . . . , an .
S(a0 , a1 , . . . , an ) = E[(Xn+1 ° X̂n+1 )2 ]
= E[(Xn+1 ° a0 ° a1 Xn ° · · · ° an X1 )2 ]
= a20 ° 2a0 E[Xn+1 ° a1 Xn ° · · · ° an X1 ]
+ E[(Xn+1 ° a1 Xn ° · · · ° an X1 )2 ]
= a20 + E[(Xn+1 ° a1 Xn ° · · · ° an X1 )2 ].
Differentiation with respect to ai gives
@S
= 2a0 ,
@a0
@S
= °2E[((Xn+1 ° a1 Xn ° · · · ° an X1 )Xn+1°i ], i = 1, . . . , n.
@ai

9
Putting the partial derivatives equal to zero we get that S(a0 , a1 , . . . , an ) is mini-
mized if

a0 = 0
E[(Xn+1 ° X̂n+1 )Xk ] = 0, for each k = 1, . . . , n.

Plugging in the expression for Xn+1 we get that for k = 1, . . . , n.

0 = E[(Xn+1 ° X̂n+1 )Xk ]

= E[(¡1 Xn + · · · + ¡p Xn°p+1 + Zn+1 ° a1 Xn ° · · · ° an X1 )Xk ].

This is clearly satisfied if we let

Ω
ai = ¡i , if 1 ∑ i ∑ p
ai = 0, if i > p

Since there is best linear predictor is unique this is the one. The mean square error
is

E[(Xn+1 ° X̂n+1 )2 ] = E[Zn+1

2
] = æ2 .

10
Chapter 3
Problem 3.1. We write the ARMA processes as ¡(B)Xt = µ(B)Zt . The process
{Xt : t 2 Z} is causal if and only if ¡(z) 6= 0 for each |z| ∑ 1 and invertible if and
only if µ(z) 6= 0 for each |z| ∑ 1.
a) ¡(z) = 1 + 0.2z ° 0.48z 2 = 0 is solved by z1 = 5/3 and z2 = °5/4.
Hence {Xt : t 2 Z} is causal.
µ(z) = 1. Hence {Xt : t 2 Z} is invertible.
b) ¡(z) = 1 + 1.9z + 0.88z 2 = 0 is solved by z1 = °10/11 and z2 = °5/4.
Hence {Xt : t 2 Z} is not causal.
p
µ(z) = 1 + 0.2z + 0.7z 2 = 0 is solved by z1 = °(1 ° i 69)/7
p p
and z2 = °(1 + i 69)/7. Since |z1 | = |z2 | = 70/7 > 1, {Xt : t 2 Z}
is invertible.
c) ¡(z) = 1 + 0.6z = 0 is solved by z = °5/3. Hence {Xt : t 2 Z} is causal.
µ(z) = 1 + 1.2z = 0 is solved by z = °5/6. Hence {Xt : t 2 Z} is not invertible.
d) ¡(z) = 1 + 1.8z + 0.81z 2 = 0 is solved by z1 = z2 = °10/9.
Hence {Xt : t 2 Z} is causal.
µ(z) = 1. Hence {Xt : t 2 Z} is invertible.
e) ¡(z) = 1 + 1.6z = 0 is solved by z = °5/8. Hence {Xt : t 2 Z} is not causal.
µ(z) = 1 ° 0.4z + 0.04z 2 = 0 is solved by z1 = z2 = 5.
Hence {Xt : t 2 Z} is invertible.
° ¢
Problem 3.4. We have Xt = 0.8Xt°2 + Zt , where {Zt : t 2 Z} ª WN 0, æ 2 . To
obtain the Yule-Walker equations we multiply each side by Xt°k and take expected
value. Then we get
E[Xt Xt°k ] = 0.8E[Xt°2 Xt°k ] + E[Zt Xt°k ],
which gives us
∞(0) = 0.8∞(2) + æ 2
∞(k) = 0.8∞(k ° 2), k ∏ 1.
We use that ∞(k) = ∞(°k). Thus, we need to solve
∞(0) ° 0.8∞(2) = æ 2
∞(1) ° 0.8∞(1) = 0
∞(2) ° 0.8∞(0) = 0
First we see that ∞(1) = 0 and therefore ∞(h) = 0 if h is odd. Next we solve for
∞(0) and we get ∞(0) = æ 2 (1 ° 0.82 )°1 . It follows that ∞(2k) = ∞(0)0.8k and hence
the ACF is
8
< 1 h = 0,
Ω(h) = 0.8h , h = 2k, k = ±1, ±2, . . .
:
0 otherwise.
The PACF can be computed as Æ(0) = 1, Æ(h) = ¡hh where ¡hh comes from that
the best linear predictor of Xh+1 has the form
h
X
X̂h+1 = ¡hi Xh+1°i .
i=1

11
For an AR(2) process we have X̂h+1 = ¡1 Xh + ¡2 Xh°1 where we can identify
Æ(0) = 1, Æ(1) = 0, Æ(2) = 0.8 and Æ(h) = 0 for h ∏ 3.
Problem 3.6. The ACVF for {Xt : t 2 Z} is
∞X (t + h, t) = Cov(Xt+h , Xt ) = Cov(Zt+h + µZt+h°1 , Zt + µZt°1 )
= ∞Z (h) + µ∞Z (h + 1) + µ∞Z (h ° 1) + µ2 ∞Z (h)
Ω 2
æ (1 + µ2 ), h = 0
=
æ 2 µ, |h| = 1.
On the other hand, the ACVF for {Yt : t 2 Z} is
∞Y (t + h, t) = Cov(Yt+h , Yt ) = Cov(Z̃t+h + µ°1 Z̃t+h°1 , Z̃t + µ°1 Z̃t°1 )
= ∞Z̃ (h) + µ°1 ∞Z̃ (h + 1) + µ°1 ∞Z̃ (h ° 1) + µ°2 ∞Z̃ (h)
Ω 2 2
æ µ (1 + µ°2 ) = æ 2 (1 + µ2 ), h = 0
=
æ 2 µ2 µ°1 = æ 2 µ, |h| = 1.
Hence they are equal.
° ¢
Problem 3.7. First we show that {Wt : t 2 Z} is WN 0, æw2
.
2 3
X1 1
X
E[Wt ] = E 4 (°µ)°j Xt°j 5 = (°µ)°j E[Xt°j ] = 0,
j=0 j=0

since E[Xt°j ] = 0 for each j. Next we compute the ACVF of {Wt : t 2 Z} for
h ∏ 0.
2 3
X1 1
X
∞W (t + h, t) = E[Wt+h Wt ] = E 4 (°µ)°j Xt+h°j (°µ)°k Xt°k 5
j=0 k=0
1 X
X 1 1 X
X 1
= (°µ)°j (°µ)°k E[Xt+h°j Xt°k ] = (°µ)°j (°µ)°k ∞X (h ° j + k)
j=0 k=0 j=0 k=0
© ™
= ∞X (r) = æ (1 + µ )1{0} (r) + æ µ1{1} (|r|)
2 2 2

1 X
X 1
° ¢
= (°µ)°(j+k) æ 2 (1 + µ2 )1{j°k} (h) + æ 2 µ1{j°k+1} (h) + æ 2 µ1{j°k°1} (h)
j=0 k=0
X1 1
X
= (°µ)°(j+j°h) æ 2 (1 + µ2 ) + (°µ)°(j+j°h+1) æ 2 µ
j=h j=h°1,j∏0
1
X
+ (°µ)°(j+j°h°1) æ 2 µ
j=h+1
1
X 1
X
= æ (1 + µ )(°µ)
2 2 °h
(°µ) °2(j°h)
+ æ µ(°µ)
2 °(h°1)
(°µ)°2(j°(h°1))
j=h j=h°1,j∏0
X1
+ æ 2 µ(°µ) °(h+1)
(°µ)°2(j°(h+1))
j=h+1

µ2 °(h°1) µ
2
= æ 2 (1 + µ2 )(°µ)°h + æ 2
µ(°µ) + æ 2 µ2 1{0} (h)
µ2 ° 1 µ2 ° 1
µ2
+ æ 2 µ(°µ)°(h+1) 2
µ °1
µ 2 ° ¢
= æ 2 (°µ)°h 2 1 + µ2 ° µ2 ° 1 + æ 2 µ2 1{0} (h)
µ °1
= æ 2 µ2 1{0} (h)

12
 ° ¢
Hence, {Wt : t 2 Z} is WN 0, æw
2
with æw
2
= æ 2 µ2 . To continue we have that
1
X 1
X
Wt = (°µ)°j Xt°j = ºj Xt°j ,
j=0 j=0
P1 P1
with ºj = (°µ)°j and j=0 |ºj | = j=0 µ°j < 1 so {Xt : t 2 Z} is invertible and
P1
solves ¡(B)Xt = µ(B)Wt with º(z) = j=0 ºj z j = ¡(z)/µ(z). This implies that
we must have
1 ≥
z ¥j
1
X X 1 ¡(z)
ºj z =
j
° = = .
j=0 j=0
µ 1 + z/µ µ(z)

Hence, ¡(z) = 1 and µ(z) = 1 + z/µ, i.e. {Xt : t 2 Z} satisfies Xt = Wt + µ°1 Wt°1 .
Problem 3.11. The PACF can be computed as Æ(0) = 1, Æ(h) = ¡hh where ¡hh
comes from that the best linear predictor of Xh+1 has the form
h
X
X̂h+1 = ¡hi Xh+1°i .
i=1

In particular Æ(2) = ¡22 in the expression

X̂3 = ¡21 X2 + ¡22 X1 .

The best linear predictor satisfies

Cov(X3 ° X̂3 , Xi ) = 0, i = 1, 2.

This gives us

Cov(X3 ° X̂3 , X1 ) = Cov(X3 ° ¡21 X2 ° ¡22 X1 , X1 )

= Cov(X3 , X1 ) ° ¡21 Cov(X2 , X1 ) ° ¡22 Cov(X1 , X1 )
= ∞(2) ° ¡21 ∞(1) ° ¡22 ∞(0) = 0

and

Cov(X3 ° X̂3 , X2 ) = Cov(X3 ° ¡21 X2 ° ¡22 X1 , X2 )

= ∞(1) ° ¡21 ∞(0) ° ¡22 ∞(1) = 0.

Since we have an MA(1) process it has ACVF

8 2
< æ (1 + µ2 ), h = 0,
∞(h) = æ 2 µ, |h| = 1,
:
0, otherwise.

Thus, we have to solve the equations

¡21 ∞(1) + ¡22 ∞(0) = 0

(1 ° ¡22 )∞(1) ° ¡21 ∞(0) = 0.

Solving this system of equations we find

µ2
¡22 = ° .
µ4 + µ2 + 1

13
Chapter 4
P
Problem 4.4. By Corollary 4.1.1 we know that a function ∞(h) with |h|<1 |∞(h)|
is ACVF for some stationary process if and only if it is an even function and
1
1 X °ih∏
f (∏) = e ∞(h) ∏ 0, for ∏ 2 (°º, º].
2º
h=°1

We have that ∞(h) is even, ∞(h) = ∞(°h) and

3
1 X °ih∏
f (∏) = e ∞(h)
2º
h=°3
1 ° ¢
= °0.25ei3∏ ° 0.5ei2∏ + 1 ° 0.5e°i2∏ ° 0.25e°i3∏
2º
1 ° ¢
= 1 ° 0.25(ei3∏ + e°i3∏ ) ° 0.5(ei2∏ + e°i2∏ )
2º
1
= (1 ° 0.5 cos(3∏) ° cos(2∏)) .
2º
Do we have f (∏) ∏ 0 on ∏ 2 (°º, º]? The answer is NO, for instance f (0) =
°1/(4º). Hence, ∞(h) is NOT an ACVF for a stationary time series.
Problem 4.5. Let Zt = Xt + Yt . First we show that ∞Z (h) = ∞X (h) + ∞Y (h).
∞Z (t + h, t) = Cov(Zt+h , Zt ) = Cov(Xt+h + Yt+h , Xt + Yt )
= Cov(Xt+h , Xt ) + Cov(Xt+h , Yt ) + Cov(Yt+h , Xt ) + Cov(Yt+h , Yt )
= Cov(Xt+h , Xt ) + Cov(Yt+h , Yt )
= ∞X (t + h, t) + ∞Y (t + h, t).
We have that
Z
∞Z (h) = eih∏ dFZ (∏)
(°º,º]

but we also know that

Z Z
∞Z (h) = ∞X (h) + ∞Y (h) = e dFX (∏) +
ih∏
eih∏ dFY (∏)
(°º,º] (°º,º]
Z
= eih∏ (dFX (∏) + dFY (∏))
(°º,º]

Hence we have that dFZ (∏) = dFX (∏) + dFY (∏), which implies that
Z Z
FZ (∏) = dFZ (∫) = (dFX (∫) + dFY (∫)) = FX (∏) + FY (∏).
(°º,∏] (°º,∏]

Problem 4.6. Since {Yt : t 2 Z} is MA(1)-process we have

8 2
< æ (1 + µ2 ), h = 0,
∞Y (h) = æ 2 µ, |h| = 1,
:
0, otherwise.
By Problem 2.2 the process St = A cos(ºt/3) + B sin(ºt/3) has ACVF ∞S (h) =
∫ 2 cos(ºh/3). Since the processes are uncorrelated, Problem 4.5 gives that ∞X (h) =
∞S (h) + ∞Y (h). Moreover,
Z º
∫ 2 iºh/3
∫ 2 cos(ºh/3) = (e + e°iºh/3 ) = ei∏h dFS (∏),
2 °º

14
where

∫2 ∫2
dFS (∏) = ± (∏ ° º/3) d∏ + ± (∏ + º/3) d∏
2 2
This implies
8
< 0, ∏ < °º/3,
FS (∏) = ∫ 2 /2, °º/3 ∑ ∏ < º/3,
: 2
∫ , ∏ ∏ º/3.

Furthermore we have that

1
1 X °ih∏ 1 ° i∏ ¢
fY (∏) = e ∞Y (h) = e ∞Y (°1) + ∞Y (0) + e°i∏ ∞Y (1)
2º 2º
h=°1
1 ° 2° ¢ ° ¢¢ æ 2
= æ 1 + 2.52 + 2.5æ 2 ei∏ + e°i∏ = (7.25 + 5 cos(∏)).
2º 2º
This implies that
Z ∏ Z ∏
æ2 æ2 £ §∏
FY (∏) = fY (ª)dª = (7.25 + 5 cos(ª))dª = 7.25ª + 5 sin(ª) °º
°º °º 2º 2º
2
æ
= (7.25(∏ + º) + 5 sin(∏)).
2º
Finally we have FX (∏) = FS (∏) + FY (∏).

Problem 4.9. a) We start with ∞X (0),

Z º Z °º
6 +0.01
Z º
6 +0.01
∞X (0) = ei0∏ fX (∏)d∏ = 100 d∏ + 100 d∏ = 100 · 0.04 = 4.
°º °º
6 °0.01
º
6 °0.01

For ∞X (1) we have,

Z º
∞X (1) = ei∏ fX (∏)d∏
°º
Z ° º6 +0.01 Z º
6 +0.01
= 100 e d∏ + 100
i∏
ei∏ d∏
°º6 °0.01
º
6 °0.01
∑∏ º
i∏ ° 6 +0.01
∑ ∏ º
i∏ 6 +0.01
e e
= 100 + 100
i ° º °0.01 i º °0.01
6 6
≥
100 i(° º +0.01) º º º
¥
= e 6 °e 6°i( +0.01)
+ ei( 6 +0.01) ° e°i(° 6 +0.01)
i ≥ ≥ ¥ ≥º ¥¥
º
= 200 sin ° + 0.01 + sin + 0.01
p 6 6
= 200 3 sin(0.01) º 3.46.

The spectral density fX (∏) is plotted in Figure 4.9(a).

b) Let
1
X
Yt = r12 Xt = Xt ° Xt°12 = √k Xt°k ,
k=°1

15
with √0 = 1, √12 = °1 and √j = 0 otherwise. Then we have the spectral density
fY (∏) = |√(e°i∏ )|2 fX (∏) where
1
X
√(e°i∏ ) = √k e°ik∏ = 1 ° e°i12∏ .
k=°1

Hence,
Ø Ø2
fY (∏) = Ø1 ° e°12i∏ Ø fX (∏) = (1 ° e°12i∏ )(1 ° e12i∏ )fX (∏)
= 2(1 ° cos(12∏))fX (∏).
The power transfer function |√(e°i∏ )|2 is plotted in Figure 4.9(b) and the resulting
spectral density fY (∏) is plotted in Figure 4.9(c).

c) The variance of Yt is ∞Y (0) which is computed by

Z º
∞Y (0) = fY (∏)d∏
°º
Z°º6 +0.01
Z º
6 +0.01
= 200 (1 ° cos(12∏))d∏ + 200 (1 ° cos(12∏))d∏
°º
6 °0.01
º
6 °0.01
µh ∂
sin(12∏) i° º6 +0.01 h sin(12∏) i º6 +0.01
= 200 ∏ ° + ∏°
12 °º 6 °0.01
12 º
6 °0.01
µ
sin(12(°º/6 + 0.01)) ° sin(12(°º/6 ° 0.01))
= 200 0.02 °
12
∂
sin(12(º/6 + 0.01)) ° sin(12(º/6 ° 0.01))
+0.02 °
12
µ ∂
sin(2º ° 0.12) ° sin(2º + 0.12)
= 200 0.04 +
6
µ ∂
1
= 200 0.04 ° sin(0.12) = 0.0192.
3

µ(B)
Problem 4.10. a) Let ¡(z) = 1 ° ¡z and µ(z) = 1 ° µz. Then Xt = ¡(B) Zt and
Ø Ø Ø Ø
Ø µ(e°i∏ ) Ø2 °i∏ Ø2 2
Ø fZ (∏) = Ø µ(e ) Ø æ .
Ø
fX (∏) = ØØ
¡(e°i∏ ) Ø Ø ¡(e°i∏ ) Ø 2º

For {Wt : t 2 Z} we get

Ø Ø2 Ø Ø Ø2 Ø Ø2
Ø Ø 1 °i∏ Ø Ø
Ø ¡(e
Ø ˜ °i∏ Ø Ø °i∏ Ø2 2
) Ø Ø µ(e ) Ø æ Ø1 ° ¡ e Ø 1 ° µe°i∏ Ø æ 2
fW (∏) = Ø Ø = Ø Ø .
Ø µ̃(e°i∏ ) Ø Ø ¡(e°i∏ ) Ø 2º Ø1 ° 1 e°i∏ Ø2 |1 ° ¡e°i∏ |2 2º
µ

Now note that we can write

Ø Ø2 Ø i∏ Ø2
Ø Ø Ø
Ø1 ° 1 e°i∏ Ø = 1 Ø¡ ° e°i∏ Ø2 = e Ø¡ ° e°i∏ Ø2 = 1 Ø¡ei∏ ° 1Ø2
Ø Ø Ø Ø Ø Ø Ø
Ø ¡ Ø ¡2 ¡ 2 ¡ 2

1 ØØ Ø 2 1 Ø Ø 2
= 2 1 ° ¡ei∏ Ø = 2 Ø1 ° ¡e°i∏ Ø .
¡ ¡
Inserting this and the corresponding expression with ¡ substituted by µ in the
computation above we get
1 Ø
Ø Ø Ø
°i∏ Ø2 Ø
Ø2
¡2 1 ° ¡e 1 ° µe°i∏ Ø æ 2 µ2 æ 2
fW (∏) = 1 =
°i∏ |2 |1 ° ¡e°i∏ |2 2º ¡2 2º
µ 2 |1 ° µe

16
100

80

60

40

20

0
0 π/6 π/3 π/2 2π/3 5π/6

(a) fX (∏)

0
0 π/6 π/3 π/2 2π/3 5π/6

2
(b) √ e°i∏

1.5

0.5

0
π/6−0.01 π/6 π/6+0.01

(c) fY (∏)

Figure 1: Exercise 4.9

17
which is constant.
b) Since {Wt : t 2 Z} has constant spectral density it is white noise and
Z º
µ2 æ2 µ2
æw2
= ∞W (0) = fW (∏)d∏ = 2 2º = 2 æ 2 .
°º ¡ 2º ¡

˜
c) From definition of {Wt : t 2 Z} we get that ¡(B)X t = µ̃(B)Wt which is a causal
and invertible representation.

18
Chapter 5
Problem 5.1. We begin by writing the Yule-Walker equations. {Yt : t 2 Z}
satisfies

Yt ° ¡1 Yt°1 ° ¡2 Yt°2 = Zt , {Zt : t 2 Z} ª WN(0, æ 2 ).

Multiplying this equation with Yt°k and take expectation gives

Ω 2
æ k = 0,
∞(k) ° ¡1 ∞(k ° 1) ° ¡2 ∞(k ° 2) =
0 k ∏ 1.

We rewrite the first three equations as

Ω
∞(k) k = 1, 2,
¡1 ∞(k ° 1) + ¡2 ∞(k ° 2) =
∞(0) ° æ 2 k = 0.

Introducing the notation

µ ∂ µ ∂ µ ∂
∞(0) ∞(1) ∞(1) ¡1
Γ2 = , ∞2 = , ¡=
∞(1) ∞(0) ∞(2) ¡2

ˆ 2 and
we have Γ2 ¡ = ∞ 2 and æ 2 ° ∞(0) ° ¡T ∞ 2 . We replace Γ2 by Γ̂2 and ∞ 2 by ∞
ˆ
solve to get an estimate ¡ for ¡. That is, we solve

ˆ =∞
Γ̂2 ¡ ˆ2 ˆT ∞
æ̂ 2 = ∞ˆ (0) ° ¡ ˆ 2.

Hence
µ ∂µ ∂
ˆ = Γ̂°1 ∞ 1 ∞ˆ (0) °ˆ ∞ (1) ∞ˆ (1)
¡ 2 ˆ2 = ∞ (1) ∞ˆ (0) ∞ˆ (2)
∞ˆ (0)2 ° ∞ˆ (1)2 °ˆ
µ ∂
1 ∞ˆ (0)ˆ
∞ (1) °ˆ ∞ (1)ˆ
∞ (2)
= .
∞ˆ (0)2 ° ∞ˆ (1)2 °ˆ∞ (1)2 ∞ˆ (0)ˆ
∞ (2)

We get that

(ˆ∞ (0) ° ∞ˆ (2))ˆ ∞ (1)

¡ˆ1 = = 1.32
∞ˆ (0)2 ° ∞ˆ (1)2
∞ˆ (0)ˆ ∞ (2) ° ∞ˆ (1)2
¡ˆ2 = = °0.634
∞ˆ (0)2 ° ∞ˆ (1)2
æ̂ 2 = ∞ˆ (0) ° ¡ˆ1 ∞ˆ (1) ° ¡ˆ2 ∞ˆ (2) = 289.18.

ˆ ª AN(¡, æ̂ 2 Γ̂°1 /n).

ˆ ª AN(¡, æ 2 Γ°1 /n) and approximately ¡
We also have that ¡ 2 2
Here
µ ∂ µ ∂
2 °1 289.18 0.0021 °0.0017 0.0060 °0.0048
æ̂ Γ̂2 /n = =
100 °0.0017 0.0021 °0.0048 0.0060

So we have approximately ¡ˆ1 ª N (¡1 , 0.0060) and ¡ˆ2 ª N (¡2 , 0.0060) and the
confidence intervals are
p
I¡1 = ¡ˆ1 ± ∏0.025 0.006 = 1.32 ± 0.15
p
I¡ = ¡ˆ2 ± ∏0.025 0.006 = °0.634 ± 0.15.
2

19
Problem 5.3. a) {Xt : t 2 Z} is causal if ¡(z) 6= 0 for |z| ∑ 1 so let us check for
which values of ¡ this can happen. ¡(z) = 1 ° ¡z ° ¡2 z 2 so putting this equal to
zero implies
p p
z 1 1° 5 1+ 5
z + ° 2 = 0 ) z1 = °
2
and z2 = °
¡ ¡ 2¡ 2¡
p p
Furthermore |z1 | > 1 if |¡| < ( 5 ° 1)/2 = 0.61 and |z2 | > 1 if |¡| < (1 + 5)/2 =
1.61. Hence, the process is causal if |¡| < 0.61.
b) The Yule-Walker equations are
Ω 2
æ k = 0,
∞(k) ° ¡∞(k ° 1) ° ¡ ∞(k ° 2) =
2
0 k ∏ 1.

Rewriting the first 3 equations and using ∞(k) = ∞(°k) gives

∞(0) ° ¡∞(1) ° ¡2 ∞(2) = æ 2

∞(1) ° ¡∞(0) ° ¡2 ∞(1) = 0
∞(2) ° ¡∞(1) ° ¡2 ∞(0) = 0.

Multiplying the third equation by ¡2 and adding the first gives

°¡3 ∞(1) ° ¡∞(1) ° ¡4 ∞(0) + ∞(0) = æ 2

∞(1) ° ¡∞(0) ° ¡2 ∞(1) = 0.

We solve the second equation to obtain

s
1 1
¡=° ± + 1.
2Ω(1) 4Ω(1)2

Inserting the estimated values of ∞ˆ (0) and ∞ˆ (1) = ∞ˆ (0)Ω̂(1) gives the solutions
¡ˆ = {0.509, °1.965} and we choose the causal solution ¡ˆ = 0.509. Inserting this
value in the expression for æ 2 we get

æ̂ 2 = °¡ˆ3 ∞ˆ (1) ° ¡ˆ
ˆ∞ (1) ° ¡ˆ4 ∞ˆ (0) + ∞ˆ (0) = 2.985.

Problem 5.4.° a) Let ¢ us construct a test to see if the assumption that {Xt ° µ :
t 2 Z}
° is WN
¢ 0, æ 2
is reasonable. To this end suppose that {Xt ° µ : t 2 Z} is
WN 0, æ 2 . Then, since Ω(k) = 0 for k ∏ 1 we have that Ω̂(k) ª AN(0, 1/n). A
p
95% confidence interval for Ω(k) is then IΩ(k) = Ω̂(k) ± ∏0.025 / 200. This gives us

IΩ(1) = 0.427 ± 0.139

IΩ(2) = 0.475 ± 0.139
IΩ(3) = 0.169 ± 0.139.

Clearly 0 2
/ IΩ(k) for any of the observed k = 1, 2, 3 and we conclude that it is not
reasonable to assume that {Xt ° µ : t 2 Z} is white noise.
b) We estimate the mean by µ̂ = x200 = 3.82. The Yule-Walker estimates is given
by
ˆ = R̂°1 Ωˆ , æ̂ 2 = ∞ˆ (0)(1 ° Ωˆ2 T R̂°1
¡ 2 2 2 Ωˆ2 ),

where
µ ∂ µ ∂ µ ∂
ˆ= ¡ˆ1 Ω̂(0) Ω̂(1) Ω̂(1)
¡ , R̂2 = , Ω̂2 = .
¡ˆ2 Ω̂(1) Ω̂(0) Ω̂(2)

20
Solving this system gives the estimates ¡ˆ1 = 0.2742, ¡ˆ2 = 0.3579 and æ̂ 2 = 0.8199.
c) We construct a 95% confidence interval for µ to test if we can reject the hypothesis
that µ = 0. We have that X 200 ª AN(µ, ∫/n) with
1
X
∫= ∞(h) º ∞ˆ (°3) + ∞ˆ (°2) + ∞ˆ (°1) + ∞ˆ (0) + ∞ˆ (1) + ∞ˆ (2) + ∞ˆ (3) = 3.61.
h=°1

An approximate 95% confidence interval for µ is then

p p
I = xn ± ∏0.025 ∫/n = 3.82 ± 1.96 3.61/200 = 3.82 ± 0.263.

Since 0 2
/ I we reject the hypothesis that µ = 0.
d) We have that approximately ¡ˆ ª AN(¡, æ̂ 2 Γ̂°1 /n). Inserting the observed values
2
we get
°1 µ ∂
æ̂ 2 Γ̂2 0.0050 °0.0021
= ,
n °0.0021 0.0050

and hence ¡ˆ1 ª AN(¡1 , 0.0050) and ¡ˆ2 ª AN(¡2 , 0.0050). We get the 95% confi-
dence intervals
p
I¡1 = ¡ˆ1 ± ∏0.025 0.005 = 0.274 ± 0.139
p
I¡ = ¡ˆ2 ± ∏0.025 0.005 = 0.358 ± 0.139.
2

e) If the data were generated from an AR(2) process, then the PACF would be
Æ(0) = 1, Æ̂(1) = Ω̂(1) = 0.427, Æ̂(2) = ¡ˆ2 = 0.358 and Æ̂(h) = 0 for h ∏ 3.

Problem 5.11. To obtain the maximum likelihood estimator we compute as if the

process were Gaussian. Then the innovations

X1 ° X̂1 = X1 ª N (0, ∫0 ),
X2 ° X̂2 = X2 ° ¡X1 ª N (0, ∫1 ),

where ∫0 = æ 2 r0 = E[(X1 ° X̂1 )2 ], ∫1 = æ 2 r1 = E[(X2 ° X̂2 )2 ]. This implies

∫0 = E[X12 ] = ∞(0), r0 = 1/(1°¡2 ) and ∫1 = E[(X2 ° X̂2 )2 ] = ∞(0)°2¡∞(1)+¡2 ∞(0)
and hence

∞(0)(1 + ¡2 ) ° 2¡∞(1) 1 + ¡2 ° 2¡2

r1 = = = 1.
æ 2 1 ° ¡2

Here we have used that ∞(1) = æ 2 ¡/(1 ° ¡2 ). Since the distribution of the innova-
tions is normal the density for Xj ° X̂j is
µ ∂
1 x2
fXj °X̂j =p exp ° 2
2ºæ 2 rj°1 2æ rj°1

and the likelihood function is

2
Y µΩ ∂æ
1 1 (x1 ° x̂1 )2 (x2 ° x̂2 )2
L(¡, æ ) =
2
fXj °X̂j = p exp ° 2 +
j=1
(2ºæ 2 )2 r0 r1 2æ r0 r1
Ω µ 2 ∂æ
1 1 x1 (x2 ° ¡x1 )2
=p exp ° 2 + .
(2ºæ 2 )2 r0 r1 2æ r0 r1

21
We maximize this by taking logarithm and then differentiate:

1 1 ° x2 (x2 ° ¡x1 )2 ¢
log L(¡, æ 2 ) = ° log(4º 2 æ 4 r0 r1 ) ° 2 1 +
2 2æ r0 r1
1 1 ° ¢
= ° log(4º 2 æ 4 /(1 ° ¡2 )) ° 2 x21 (1 ° ¡2 ) + (x2 ° ¡x1 )2
2 2æ
1 1 ° ¢
= ° log(2º) ° log(æ 2 ) + log(1 ° ¡2 ) ° 2 x21 (1 ° ¡2 ) + (x2 ° ¡x1 )2 .
2 2æ
Differentiating yields

@l(¡, æ 2 ) 1 1 ° ¢
= ° 2 + 4 x21 (1 ° ¡2 ) + (x2 ° ¡x1 )2 ,
@æ 2 æ 2æ
@l(¡, æ 2 ) 1 °2¡ x1 x2
= · + 2 .
@¡ 2 1°¡ 2 æ
° ¢
Putting these expressions equal to zero gives æ 2 = 12 x21 (1 ° ¡2 ) + (x2 ° ¡x1 )2 and
then after some computations ¡ = 2x1 x2 /(x21 + x22 ). Inserting the expression for ¡
is the equation for æ gives the maximum likelihood estimators

(x21 ° x22 )2 2x1 x2

æ̂ 2 = and ¡ˆ = 2
2(x1 + x2 )
2 2 x1 + x22

22
Chapter 6
Problem 6.5. The best linear predictor of Yn+1 in terms of 1, X0 , Y1 , . . . , Yn i.e.

Ŷn+1 = a0 + cX0 + a1 Y1 + · · · + an Yn ,

must satisfy the orthogonality relations

Cov(Yn+1 ° Ŷn+1 , 1) = 0
Cov(Yn+1 ° Ŷn+1 , X0 ) = 0
Cov(Yn+1 ° Ŷn+1 , Yj ) = 0, j = 1, . . . , n.

The second equation can be written as

Cov(Yn+1 ° Ŷn+1 , X0 ) = E[(Yn+1 ° a0 + cX0 + a1 Y1 + · · · + an Yn )X0 ] = cE[X02 ] = 0

so we must have c = 0. This does not effect the other equations since E[Yj X0 ] = 0
for each j.
Problem 6.6. Put Yt = rXt . Then {Yt : t 2 Z} is an AR(2) process. We can
rewrite this as Xt+1 = Yt + Xt°1 . Putting t = n + h and using the linearity of the
projection operator Pn gives Pn Xn+h = Pn Yn+h + Pn Xn+h°1 . Since {Yt : t 2 Z} is
AR(2) process we have Pn Yn+1 = ¡1 Yn + ¡2 Yn°1 , Pn Yn+2 = ¡1 Pn Yn+1 + ¡2 Yn and
iterating we find Pn Yn+h = ¡1 Pn Yn+h°1 + ¡2 Pn Yn+h°2 . Let ¡§ (z) = (1 ° z)¡(z) =
1 ° ¡§1 z ° ¡§2 z 2 ° ¡§3 z 3 . Then

(1 ° z)¡(z) = 1 ° ¡1 z ° ¡2 z ° z + ¡1 z 2 + ¡2 z 3 ,

i.e. ¡§1 = ¡1 + 1, ¡§2 = ¡2 ° ¡1 and ¡§3 = °¡2 . Then

3
X
Pn Xn+h = ¡§j Xn+h°j .
j=1

This can be verified by first noting that

Pn Yn+h = ¡1 Pn Yn+h°1 + ¡2 Pn Yn+h°2

= ¡1 (Pn Xn+h°1 ° Pn Xn+h°2 ) + ¡2 (Pn Xn+h°2 ° Pn Xn+h°3 )
= ¡1 Pn Xn+h°1 + (¡2 ° ¡1 )Pn Xn+h°2 ° ¡2 Pn Xn+h°3 .

and then

Pn Xn+h = Pn Yn+h + Pn Xn+h°1

= (¡1 + 1)Pn Xn+h°1 + (¡2 ° ¡1 )Pn Xn+h°2 ° ¡2 Pn Xn+h°3
= ¡§1 Pn Xn+h°1 + ¡§2 Pn Xn+h°2 + ¡§3 Pn Xn+h°3 .

Hence, we have
Ω
¡§1 g(h ° 1) + ¡§2 g(h ° 2) + ¡§3 g(h ° 3), h ∏ 1,
g(h) =
Xn+h , h ∑ 0.

We may suggest a solution of the form g(h) = a+bª1°h +cª2°h , h > °3 where ª1 and
ª2 are the solutions to ¡(z) = 0 and g(°2) = Xn°2 , g(°1) = Xn°1 and g(0) = Xn .
Let us first find the roots ª1 and ª2 .
4 1 16
¡(z) = 1 ° 0.8z + 0.25z 2 = 1 ° z + z 2 = 0 ) z 2 ° z + 4 = 0.
5 4 5

23
 p
We get that z = 8/5 ± (8/5)2 ° 4 = (8 ± 6i)/5. Then ª1°1 = 5/(8 + 6i) = · · · =
0.4 ° 0.3i and ª2°1 = 0.4 + 0.3i. Next we find the constants a, b and c by solving

Xn°2 = g(°2) = a + bª1°2 + cª2°2 ,

Xn°1 = g(°1) = a + bª1°1 + cª2°1 ,
Xn = g(0) = a + b + c.

Note that (0.4 ° 0.3i)2 = 0.07 ° 0.24i and (0.4 + 0.3i)2 = 0.07 + 0.24i so we get the
equations

Xn°2 = a + b(0.07 ° 0.24i) + c(0.07 + 0.24i),

Xn°1 = a + b(0.4 ° 0.3i) + c(0.4 + 0.3i),
Xn = a + b + c.

Let a = a1 + a2 i, b = b1 + b2 i and c = c1 + c2 i. Then we split the equations into a

real part and an imaginary part and get

Xn°2 = a1 + 0.07b1 + 0.24b2 + 0.07c1 ° 0.24c2 ,

Xn°1 = a1 + 0.4b1 + 0.3b2 + 0.4c1 ° 0.4c2 ,
Xn = a1 + b1 + c1 ,
0 = a2 + 0.07b2 ° 0.24b1 + 0.07c2 + 0.24c1 ,
0 = a2 + 0.4b2 ° 0.3b1 + 4c2 + 0.3c1 ,
0 = a2 + b2 + c2 .

We can write this as a matrix equation by

0 10 1 0 1
1 0 0.07 0.24 0.07 °0.24 a1 Xn°2
B 1 0 0.4 0.3 0.4 °0.3 CB a2 C B Xn°1 C
B CB C B C
B 1 0 1 0 1 0 CB b1 C B Xn C
B CB C=B C,
B 0 1 °0.24 0.07 0.24 0.07 CB b2 C B 0 C
B CB C B C
@ 0 1 °0.3 0.4 0.3 0.4 A@ c1 A @ 0 A
0 1 0 1 0 1 c2 0

which has the solution a = 2.22Xn ° 1.77Xn°1 + 0.55Xn°2 , b = c = °1.1Xn°2 +

0.88Xn°1 + 0.22Xn + (°2.22Xn°2 + 3.44Xn°1 ° 1.22Xn )i.

24
Chapter 7
Problem 7.1. The problem is not very well formulated; we replace the condition
ΩY (h) ! 0 as h ! 1 by the condition that ΩY (h) is strictly decreasing.

The process is stationary if µ̄t = E[(X1,t , X2,t )T ] = (µ1 , µ2 )T and Γ(t + h, t) does
not depend on t. We may assume that {Yt } has mean zero so that
E[X1,t ] = E[Yt ] = 0
E[X2,t ] = E[Yt°d ] = 0,
and the covariance function is
µ ∂
E[Yt+h Yt ] E[Yt+h Yt°d ]
Γ(t + h, t) = E[(X1,t+h , X2,t+h ) (X1,t , X2,t )] =
T
E[Yt+h°d Yt ] E[Yt+h°d Yt°d ]
µ ∂
∞Y (h) ∞Y (h + d)
= .
∞Y (h ° d) ∞Y (h)
Since neither µ̄t or Γ(t + h, t) depend on t, the process is stationary. We assume
that ΩY (h) ! 0 as h ! 1. Then we have that the cross-correlation
∞12 (h) ∞Y (h + d)
Ω12 (h) = p = = ΩY (h + d).
∞11 (0)∞22 (0) ∞Y (0)

In particular, Ω12 (0) = ΩY (d) < 1 whereas Ω12 (°d) = ΩY (0) = 1.

Problem 7.3. We want to estimate the cross-correlation
p
Ω12 (h) = ∞12 (h)/ ∞11 (0)∞22 (0).
We estimate
µ ∂
∞11 (h) ∞12 (h)
Γ(h) =
∞21 (h) ∞22 (h)
by
Ω Pn°h
1
(Xt+h ° X̄n )(Xt ° X̄n )T 0 ∑ h ∑ n ° 1
Γ̂(h) = n t=1
ΓT (°h) °n + 1 ∑ h < 0.
p
Then we get Ω̂12 (h) = ∞ˆ12 (h)/ ∞ˆ11 (0)ˆ∞22 (0). According to Theorem 7.3.1 in Brock-
well and Davis we have, for h 6= k, that
µ p ∂
pnΩ̂12 (h) ª approx. N (0, Λ)
nΩ̂21 (h)
where
1
X
Λ11 = Λ22 = Ω11 (j)Ω22 (j)
j=°1
X1
Λ12 = Λ21 = Ω11 (j)Ω22 (j + k ° h).
j=°1

Since {X1,t } and {X2,t } are MA(1) processes we know that their ACF’s are
Ω
1 h=0
ΩX1 (h) =
0.8/(1 + 0.82 ) h = ±1
Ω
1 h=0
ΩX2 (h) =
°0.6/(1 + 0.62 ) h = ±1

25
Hence
1
X
Ω11 (j)Ω22 (j) = Ω11 (°1)Ω22 (°1) + Ω11 (0)Ω22 (0) + Ω11 (1)Ω22 (1)
j=°1
0.8 °0.6 0.8 °0.6
= · +1+ · º 0.57.
1 + 0.8 1 + 0.6
2 2 1 + 0.8 1 + 0.62
2

For the covariance we see that Ω11 (j) 6= 0 if j = °1, 0, 1 and Ω22 (j + k ° h) 6= 0 if
j + k ° h = °1, 0, 1. Hence, the covariance is
1
X
Ω11 (j)Ω22 (j + k ° h) = Ω11 (°1)Ω22 (0) + Ω11 (0)Ω22 (1) º 0.0466, if k ° h = 1
j=°1
X1
Ω11 (j)Ω22 (j + k ° h) = Ω11 (0)Ω22 (°1) + Ω11 (1)Ω22 (0) º 0.0466, if k ° h = °1
j=°1
X1
Ω11 (j)Ω22 (j + k ° h) = Ω11 (°1)Ω22 (1) º °0.2152, if k ° h = 2
j=°1
X1
Ω11 (j)Ω22 (j + k ° h) = Ω11 (1)Ω22 (°1) º °0.2152, if k ° h = °2.
j=°1

Problem 7.5. We have {Xt : t 2 Z} is a causal process if det (Φ (z)) 6= 0 for all
|z| ∑ 1, due to Brockwell-Davis page 242. Further more we have that if {Xt : t 2 Z}
is a causal process, then
1
X
Xt = Ψj Zt°j ,
j=0

where
1
X
Ψj = Θj + Φk Ψj°k
k=1
Θ0 = I
Θj = 0 for j > q
Φj = 0 for j > p
Ψj = 0 for j < 0
and
1
X
Γ(h) = Ψh+j ΣΨTj , h = 0, ±1, ±2, . . .
j=0

(where in this case Σ = I2 ). We have to establish that {Xt : t 2 Z} is a causal

process and then derive Γ(h).
µ∑ ∏ ∑ ∏∂
1 0 z 1 1
det(Φ(z)) = det(I ° zΦ1 ) = det °
0 1 2 0 1
µ∑ ∏∂
1° 2z z 1 2
= det 2 = (2 ° z)
0 1 ° z2 4
Which implies that |z1 | = |z2 | = 2 > 1 and hence {Xt : t 2 Z} is a causal process.
We have that Ψj = Θj + Φ1 Ψj°1 and
Ψ0 = Θ0 + Φ1 Ψ°1 = Θ0 = I
Ψ1 = Θ1 + Φ1 Ψ0 = ΦT1 + Φ1
Ψn+1 = Φ1 Ψn for n ∏ 1.

26
From the last equation we get that Ψn+1 = Φn1 Ψ1 = Φn1 (ΦT1 + Φ1 ) and from the
definition of Φ1
∑ ∏ ≥ ¥2 ∑ ∏
1 1 n 1 5 4
Φn1 = n ΦT1 + Φ1 = .
2 0 1 4 4 5

Assume that h ∏ 0, then

1
X 1
X
Γ(h) = Ψh+j ΨTj = Ψh + Ψh+j ΨTj
j=0 j=1
1
X ≥ ¥≥ ≥ ¥¥T
= Ψh + Φh+j°1
1 ΦT1 + Φ1 Φj°1
1 ΦT
1 + Φ1
j=1
1
X ≥ ¥2 ≥ ¥T
= Ψh + Φh1 Φj1 ΦT1 + Φ1 Φj1
j=0
X1 ∏ ∑ ∑ ∏ ∑ ∏
1
1 j 1 5 4 1 1 0
= Ψh + Φh1
0 1 4 4 5 2j j 1
2j
j=0
X1 ∑ ∏
h1 1 5 + 8j + 5j 2 4 + 5j
= Ψh + Φ1
4 j=0 22j 4 + 5j 5
∑ ∏
94/27 17/9
= Ψh + Φh1 .
17/9 5/3

We have that
(
I, ≥ ¥ h=0
Ψh =
Φh°1
1 ΦT
1 + Φ1 , h>0

which gives that

∑ ∏ ∑ ∏ ∑ ∏
1 0 94/27 17/9 121/27 17/9
Γ (0) = + =
0 1 17/9 5/3 17/9 8/3

and for h > 0

≥ ¥ ∏ ∑
94/27 17/9
Γ (h) = Φh°1 + Φ1 +
ΦT1 Φh1
1 17/9 5/3
µ ∑ ∏ ∑ ∏∑ ∏∂
1 2 1 1 1 1 94/27 17/9
= Φ1 h°1
+
2 1 2 2 0 1 17/9 5/3
∑ ∏∑ ∏
1 1 h°1 199/27 41/9
= h .
2 0 1 26/9 11/3

27
Chapter 8
Problem 8.7. First we would like to show that
∑ ∏∑ ∏
1 µ Zt+1
Xt+1 = (8.1)
µ 0 Zt

is a solution to
∑ ∏ ∑ ∏
0 1 1
Xt+1 = Xt + Zt+1 . (8.2)
0 0 µ

Let
∑ ∏ ∑ ∏
0 1 1
A= and B = ,
0 0 µ

and note that

∑ ∏
0 0
A =
2
.
0 0

Then equation (8.2) can be written as

Xt+1 = AXt + BZt+1 = A (AXt°1 + BZt ) + BZt+1 = A2 Xt°1 + ABZt + BZt+1

∑ ∏ ∑ ∏ ∑ ∏ ∑ ∏∑ ∏
µ 1 µZt + Zt+1 1 µ Zt+1
= Zt + Zt+1 = = ,
0 µ µZt+1 µ 0 Zt

and hence (8.1) is a solution to equation (8.2). Next we prove that (8.1) is a unique
solution to (8.2). Let X0t+1 be another solution to equation (8.2) and consider the
difference

Xt+1 ° X0t+1 = AXt + BZt+1 ° AX0t ° BZt+1 = A (Xt ° X0t )

° ¢ ° ¢
= A AXt°1 + BZt ° AX0t°1 ° BZt = A2 Xt°1 ° X0t°1 = 0,

since A2 = 0. This implies that Xt+1 = X0t+1 , i.e. (8.1) is a unique solution to
(8.2). Moreover, Xt is stationary since
∑ ∏∑ ∏ ∑ ∏
1 µ E[Zt ] 0
µX (t) = =
µ 0 E[Zt°1 ] 0

and
∑ ∏
∞11 (t + h, t) ∞12 (t + h, t)
ΓX (t + h, t) =
∞21 (t + h, t) ∞22 (t + h, t)
∑ ∏
Cov(Zt+h + µZt+h°1 , Zt + µZt°1 ) Cov(Zt+h + µZt+h°1 , µZt )
=
Cov(µZt+h , Zt + µZt°1 ) Cov(µZt+h , µZt )
∑ ° ¢ ∏
1 + µ 1{0} (h) + µ1{°1,1} (h) µ1{0} (h) + µ2 1{1} (h)
2
= æ2 ,
µ1{0} (h) + µ2 1{°1} (h) µ2 1{0} (h)

i.e. neither of them depend on t. Now we see that

∑ ∏∑ ∏ ∑ ∏
1 µ Zt Zt
Yt = [1 0]Xt = [1 0] = [1 µ] = Zt + µZt°1 ,
µ 0 Zt°1 Zt°1

which is the MA(1) process.

28
Problem 8.9. Let Yt consist of Yt,1 and Yt,2 , then we can write
∑ ∏ ∑ ∏ ∑ ∏ ∑ ∏
Yt,1 G1 Xt,1 + Wt,1 G1 Xt,1 Wt,1
Yt = = = +
Yt,1 G2 Xt,2 + Wt,2 G2 Xt,2 Wt,2
∑ ∏∑ ∏ ∑ ∏
G1 0 Xt,1 Wt,1
= + .
0 G2 Xt,2 Wt,2
Set
∑ ∏ ∑ ∏ ∑ ∏
G1 0 Xt,1 Wt,1
G= , Xt = and Wt =
0 G2 Xt,1 Wt,2
then we have Yt = GXt + Wt . Similarly we have that
∑ ∏ ∑ ∏ ∑ ∏ ∑ ∏
Xt+1,1 F1 Xt,1 + Vt,1 F1 Xt,1 Vt,1
Xt+1 = = = +
Xt+1,1 F2 Xt,2 + Vt,2 F2 Xt,2 Vt,2
∑ ∏∑ ∏ ∑ ∏
F1 0 Xt,1 Vt,1
= +
0 F2 Xt,2 Vt,2
and set
∑ ∏ ∑ ∏
F1 0 Vt,1
F = and Vt = .
0 F2 Vt,2
Finally we have the state-space representation
Yt = GXt + Wt
Xt+1 = F Xt + Vt .
Problem 8.13. We have to solve
Ω2
Ω + æv2 ° =Ω
Ω + æw
2

which is equivalent to
Ω2
° æv2 = 0.
Ω + æw
2

Multiplying with Ω + æw
2
we get
Ω2 ° Ωæv2 ° æw æv = 0,
2 2

which has the solutions

r p
1 2 æv4 æv2 ± æv4 + 4æw2 æ2
v
Ω = æv ± + æw æv =
2 2 .
2 4 2
Since Ω ∏ 0 we have the positive root which is the solution we wanted.
Problem 8.14. We have that
Ω2t
Ωt+1 = Ωt + æv2 °
Ωt + æw2

and since æv2 = Ω2 /(Ω + æw

2
) substracting Ω yields
Ω2 Ω2t
Ωt+1 ° Ω = Ωt + ° °Ω
Ω + æw 2 Ωt + æw2
° ¢ ° ¢
Ωt Ωt + æw 2
° Ω2t Ω Ω + æw 2
° Ω2
= °
Ωt + æw 2 Ω + æw2

Ωt æ w
2
Ωæw2
= °
Ωt + æw 2 Ω + æw 2
µ ∂
Ωt Ω
= æw2
° .
Ωt + æw 2 Ω + æw 2

29
This implies that
µ ∂
Ωt Ω
(Ωt+1 ° Ω)(Ωt ° Ω) = æw
2
° (Ωt ° Ω).
Ωt + æw
2 Ω + æw
2

Now, note that the function f (x) = x/(x + æw

2
) is increasing in x. Indeed, f 0 (x) =
æw /(x + æw ) > 0. Thus we get that for Ωt > Ω both terms are > 0 and for Ωt < Ω
2 2 2

both terms are < 0. Hence, (Ωt+1 ° Ω)(Ωt ° Ω) ∏ 0.

Problem 8.15. We have the equations for µ:

µæ 2 = °æw
2

æ 2 (1 + µ2 ) = 2æw
2
+ æv2 .

From the first equation we get that æ 2 = °æw

2
/µ and inserting this in the second
equation gives
2
æw
2æw
2
+ æv2 = ° (1 + µ2 ),
µ
and multiplying by µ gives the equation

(2æw
2
+ æv2 )µ + æw
2
+ æw µ = 0.
2 2

This can be rewritten as

2æw
2
+ æv2
µ2 + µ 2
+1=0
æw

which has the solution

s p
2æ 2 + æ 2 (2æw
2 + æ 2 )2 2æw
2
+ æv2 ± æv4 + 4æv2 æw
2
µ=° w 2 v ± v
°1=° .
2æw 4æw
4 2æw
2

To get an invertible representation we choose the solution

p
2æw2
+ æv2 ° æv4 + 4æv2 æw
2
µ=° .
2æw
2

æ2
To show that µ = ° æ2 +Ω
w
, recall the steady-state solution
w

p
æv2 + æv4 + 4æv2 æw
2
Ω= ,
2
which gives
p
2æw2
+ æv2 ° æv4 + 4æv2 æw
2
µ=°
2æw
2
≥ p ¥≥ p ¥
2æw + æv ° æv4 + 4æv2 æw
2 2 2 2æw
2
+ æv2 + æv4 + 4æv2 æw
2
=° ≥ p ¥
2æw2 2æ 2 + æ 2 +
w v æv4 + 4æv2 æw
2

4æw
4
+ 4æv2 æw2
+ æv4 ° æv4 ° 4æv2 æw
2
4æw 4 2
æw
=° = ° = ° .
2æw 2 (2æ 2 + 2Ω)
w 4æw
2 (æ 2 + Ω)
w
2 +Ω
æw

30
Chapter 10
Problem 10.5. First a remark on existence of such a process: We assume for
simplicity that p = 1. A necessary and sufficient condition for the existence of a
causal, stationary solution to the ARCH(1) equations with E[Zt4 ] < 1 is that Æ12 <
1/3. If p > 1 existence of a causal, stationary solution is much more complicated.
Let us now proceed with the solution to the problem.
We have
√ p
! √ p
! √ p
!
X X Z 2
e2 X e2 ht Z2
e2t 1 + Æi Yt°i = e2t 1 + Æi t°i = t Æ0 + 2
Æi Zt°i = t = t = Yt ,
i=1 i=1
Æ0 Æ0 i=1
Æ0 Æ0

hence Yt = Zt2 /Æ0 satisfies the given equation. Let us now compute its ACVF. We
assume h ∏ 1, then
" √ p
! #
X
E[Yt Yt°h ] = E et 1 +
2
Æi Yt°i Yt°h
i=1
" p
#
X
= E[e2t ]E Yt°h + Æi Yt°i Yt°h
i=1
p
X
= E[Yt°h ] + Æi E[Yt°i Yt°h ].
i=1

Since ∞Y (h) = Cov(Yt , Yt°h ) = E[Yt Yt°h ] ° µ2Y we get

p
X ° ¢
∞Y (h) + µ2Y = µY + Æi ∞Y (h ° i) + µ2Y
i=1

and then
p
√ p
!
X X
∞Y (h) ° Æi ∞Y (h ° i) = µY + µ2Y Æi ° 1 .
i=1 i=1

We can compute µY as
" √ p
!# p p
X X X
µY = E[Yt ] = E e2t 1+ Æi Yt°i =1+ Æi E[Yt ] = 1 + µY Æi .
i=1 i=1 i=1
Pp
From this expression we see that µY = 1/(1 ° i=1 Æi ). This means that we have
p
X Pp
1 Æ °1
∞Y (h) ° Æi ∞Y (h ° i) = Pp + Pp i
i=1
= 0.
i=1
1° i=1 Æi (1 ° i=1 Æi )
2

Dividing by ∞Y (0) we find that the ACF ΩY (h) satisfies

ΩY (0) = 1,
p
X
ΩY (h) ° Æi ΩY (h ° i) = 0, h ∏ 1,
i=1

which corresponds to the Yule-Walker equations for the ACF for an AR(p) process

Wt = Æ1 Wt°1 + · · · + Æp Wt°p + Zt .

31