# 1

For the process (1 ) show that for long series the variance-covariance
ˆ
ˆ matrix of the maximum likelihood estimates , is approximately.
t t
w B a µ u
µ u
·
7.4.
2 2
-1
2
(1 ) 0
0 1
a
B
n
u o
u
]
]

]
2 2 1 2 2
1 1
1 2
ˆ ,
ˆ var( ) var( ) var((1 ) ) (1 ) .
ˆ
By (7.3.9) var( ) (1 )
ˆˆˆ
ˆˆ And cov( , ) ( ) ( ) 0
done!
n n
t t a
t t
w
n w n B a n B
n
E E
µ
µ u u o
u u
µ u µu µu

· ·

·
· · ·

·
¿ ¿
Sol :
;
;
1 1 2 2
(a) Show that the second-order autoregressive model
may be written in orthogonal form as
t t t t
z z z a o o

· + +
7.7.
% % %
1 1
1 2 2 1
2 2
1 1
t t t t t
z z z z a
o o
o
o o

| 
· + +
÷

. ,
% % % %
suggesting that the approximate estimates
2
1 2 1
1 2 2 2
2 1
ˆ
of and of
1 1
r r
r
r
o
o o
o

·

are uncorrelated for long series.
1 2
1 2
ˆˆ
(b) Starting from the variance-covariance matrix of and or otherwise, show that
ˆ
the variance-covariance matrix of and for long series is given approximately by r
o o
o
2 2
-1 2 1
2
2
(1 )(1 ) 0
0 1
n
o p
o
]
]

]
a)
2

( )
1 1
1 2 2 1
2 2
1 1 2 2
3 2
1 2 1 2 1
1 2 1 2 2
1 1
3 2
1 2 1 2 1
1 2 2
1 1
1 1
.
ˆ
Cov( , )
1 1
0.
1 1
t t t t t
t t t
p
z z z z a
z z a
r r r r r
r E E r E
r r
o o
o
o o
o o
o
p p p p p
p
p p

|
· + +
÷

. ,
· + +
|  |
·
÷ ÷

. , . ,

÷ × ·

Sol :
% % % %
% %
b)
2 2
2
2
2 2 2
1 2 1
2
2 2
1 2
1 (1 )(1 )
By Bartlett formula, var( ) 2 , where = , 1 1.
1
1 1
Thus var( ) 1 (1 )(1 ).
ˆ
By (7.3.7) var( ) 1
ˆ
By (a) we know that cov( , ) 0
variance-covaria
k
k k
k k
r k
n
r
n n
r
o o
o p o o
o
o o p
o o
o
] +
= < <
]

]
] = ·
]
·
·
=
Sol :
2 2
-1 2 1
1 2
2
2
(1 )(1 ) 0
ˆ
nce matrix of and is
0 1
r n
o p
o
o
]
]

]
A long series containing 326 terms was split into two halves and a (1,1,0) model
(1 B) identified, fitted and checked for each half. If the estimates of the
parameter for thw two halves are
t t
N
z a o
·
V ·
8.4.
(1) (2)
ˆˆ
0.5 and 0.5, is there ant evidence that
parameter has changed?
o o
o
· ·
a)
1 2
2 (1)
2 (2)
(1) (2) 2 (1) 2 (2)
(1) (2)
ˆ
for (1,1,0) var( ) (1 )
0.75
ˆ
ˆ ( ) 0.0046
163
0.51
ˆ
ˆ ( ) 0.0031
162
ˆˆˆˆ
ˆˆ standard error of ( ) ( ) 0.0880.
ˆˆ
2.272, it is likely that a real change in
n
S
T
S
o o
o o
o o
o o o o o o
o o

= ·
· ·
· ·
· · + ·

· ·
Sol :
has occurred. o
3

2
1
2
(a) Show that the variance of the sample mean of observations form a stationary
AR(1) process (1 ) is given by var[ ] .
(1 )
(b) The yields form consecutive batches of a chemicak proc
t t
n
B z a z
n
o
o
o
·

8.5.
% ;
ess obtained under fairly
uniform conditions of process control were shown to follow a stationary AR(1)
process (1+0.5 ) . A technical innovation is made at a given point in time
t t
B z a · %
2
1 1
2
2 2
points with mean 41.0 and residual variance 0.1012
before the innovation is made and 60 data points with 43.5, 0.0895
after the innovation. Is there any evidence that the innovation has i
a
a
z s
z s
· ·
· ·
mproved the yield?
a)
( )
2 2 2 2
2 2 4 2 2
2 2
1 1
2
2 2 2 3
2
note:cov( , ) cov( (1 ...), (1 ...))
( ... )
var( ) var( ) cov( , ) , 1... .
( 2 3 4 ...) .
(1 )
t t j t t j
j j j t j
a
n n
t t t t j
t t
a
a
z z a B B a B B
z n z n z z j n
n n n n n
n
o o o o
o o o o o
o
o o o o
o

+ +

· ·

· + + + +
· + + +
· · · ·
· + + + ·

¿ ¿
b)
2 2 2
0 0
2(1) 2
2(2) 2
1 2
1 2
ˆ
ˆ (1 ) for AR(1) process (1 )
ˆ 0.1012(1 0.5 ) 0.0759
ˆ 0.0895(1 0.5 ) 0.0671
85 0.0759+60 0.0671
S.D. for =0.2643.
85 60
9.4585 ,it is likely that a real chang
a a
a
a
c
z z S
z z
T
S
o y p o o o
o
o
´ · = = ·
· ·
· ·
× ×
· ·
+

· · e for yield has improved.
4

R Code
#a
GDP=scan()
81.37 82.6 82.3 83 82.87 83.6 83.33 83.53 84.27 85.5 84.33 84.3 85.07 83.6 84.37
84.5 85.20 87.07 88.4 90.03 92.30 92.13 93.17 93.5 94.77 95.37 95.03 95.23 95.07
96.4 96.97 96.5 96.16 99.79 101.14 102.95 103.96 105.28 105.81 107.14 108.07
107.64 108.87 109.75 110.2 110.2 110.9 110.4 111 112.1 112.5 113 114.3 115.1
116.4 117.8 116.8 117.8 119 119.6
par(mfrow=c(2,2))
acf(GDP,ylim=c(-1,1),lag.max=10)
acf(GDP,ylim=c(-1,1),lag.max=length(GDP))
pacf(GDP,ylim=c(-1,1),lag.max=10)
pacf(GDP,ylim=c(-1,1),lag.max=length(GDP))
n=length(GDP)
GDP.d=GDP[2:n]-GDP[1:n-1]
acf(GDP.d,ylim=c(-1,1),lag.max=10)
acf(GDP.d,ylim=c(-1,1),lag.max=length(GDP.d))
pacf(GDP.d,ylim=c(-1,1),lag.max=10)
pacf(GDP.d,ylim=c(-1,1),lag.max=length(GDP.d))
#c
phi.z=acf(GDP)\$acf[2]
mu.z=mean(GDP)
ssa.z=var(GDP)*(1-phi.z^2)
mu.w=mean(GDP.d)
ssa.w=var(GDP.d)