Professional Documents
Culture Documents
Nonlinear Univariate
and
Linear Multivariate
Time Series
Seppo Pynn
onen and
Bernd Pape, 2005
University
8
10
4
-5
-10
2
1
3.1.1989
2.1.1990
31.12.1990
30.12.1991
28.12.1992
27.12.1993
23.12.1994
22.12.1995
20.12.1996
19.12.1997
21.12.1998
-15
21.12.1999
Day
Return
Log Index
Below are autocorrelations of the log-index. Obviously the persistence of autocorrelations indicate that
the series is integrated. The autocorrelations of the
return series suggest that the returns are stationary
with statistically signicant rst order autocorrelation.
Correlogram of LNSDQ
Correlogram of DNSDQ
Partial Correlation
AC
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
0.998
0.997
0.995
0.993
0.992
0.990
0.988
0.987
0.985
0.983
0.981
0.980
0.978
0.976
0.974
0.973
0.971
0.970
0.968
0.966
0.965
0.963
0.961
0.959
0.958
0.956
0.954
0.953
0.951
0.949
0.948
0.946
0.944
0.943
0.941
0.939
PAC
0.998
-0.010
0.016
0.001
0.013
0.006
-0.018
-0.001
-0.011
0.007
-0.019
0.003
-0.001
-0.003
0.005
0.007
0.016
-0.009
-0.016
0.014
0.018
-0.016
0.001
-0.025
0.005
0.001
-0.002
0.002
-0.003
-0.004
0.000
0.003
0.013
0.000
0.007
0.014
Q-Stat
Prob
2798.4
5587.9
8369.0
11142.
13906.
16663.
19411.
22151.
24882.
27605.
30319.
33025.
35722.
38410.
41090.
43762.
46426.
49081.
51728.
54367.
56998.
59621.
62236.
64842.
67440.
70030.
72611.
75185.
77750.
80306.
82855.
85395.
87927.
90451.
92967.
95476.
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
Autocorrelation
Partial Correlation
AC
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
0.106
0.001
0.008
-0.009
-0.009
-0.012
-0.024
-0.001
-0.006
0.017
-0.004
0.049
0.058
0.009
0.017
-0.028
0.008
0.036
0.040
-0.007
-0.019
-0.046
-0.005
0.027
-0.002
0.013
0.032
0.002
0.020
0.014
0.005
0.004
0.037
-0.032
-0.037
-0.006
PAC
0.106
-0.010
0.009
-0.011
-0.007
-0.010
-0.021
0.004
-0.007
0.019
-0.009
0.051
0.048
-0.002
0.016
-0.032
0.017
0.034
0.038
-0.013
-0.017
-0.044
0.003
0.029
-0.012
0.014
0.025
-0.005
0.021
0.006
-0.002
0.000
0.041
-0.034
-0.022
-0.005
Q-Stat
Prob
31.459
31.465
31.659
31.904
32.123
32.514
34.069
34.071
34.171
35.025
35.079
41.844
51.463
51.686
52.507
54.694
54.875
58.504
63.052
63.174
64.219
70.298
70.379
72.511
72.527
73.018
75.957
75.968
77.127
77.648
77.714
77.767
81.689
84.586
88.467
88.571
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
Figure. Nasdaq Composite index autocorrelations for log levels and log differences (returns)
if
E[yt ]
, for all t
cov[yt , yt+k ]
k , for all t
0.011221
0.010868
1.091357
3338.542
-4224.341
1.997947
.11
Prob.
0.0002
0.0000
0.086119
1.097336
3.013434
3.017668
31.80961
0.000000
0.011323
0.010970
1.091301
3338.198
-4224.196
2.000051
Prob.
0.0002
0.0000
-.11
3
0.086119
1.097336
3.013331
3.017565
32.10153
0.000000
Correlogram of Residuals
Autocorrelation
Partial Correlation
AC
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
0.000
0.001
0.009
-0.010
-0.007
-0.009
-0.023
0.003
-0.008
0.019
-0.011
0.044
0.054
0.001
0.020
-0.031
0.008
0.031
0.038
-0.009
-0.014
-0.045
-0.004
0.029
-0.007
0.011
0.031
-0.003
0.019
0.011
0.004
0.000
0.041
-0.033
-0.034
0.003
PAC
0.000
0.001
0.009
-0.010
-0.007
-0.009
-0.023
0.003
-0.008
0.020
-0.012
0.044
0.053
0.001
0.020
-0.031
0.010
0.032
0.042
-0.008
-0.013
-0.046
-0.005
0.030
-0.010
0.010
0.027
-0.004
0.019
0.008
0.000
-0.004
0.043
-0.027
-0.025
-0.001
Q-Stat
Prob
2.E-05
0.0009
0.2428
0.4990
0.6319
0.8445
2.3176
2.3375
2.5307
3.5945
3.9433
9.5255
17.620
17.624
18.779
21.455
21.634
24.341
28.389
28.622
29.145
34.754
34.794
37.100
37.220
37.533
40.340
40.373
41.442
41.793
41.831
41.831
46.528
49.557
52.794
52.814
0.976
0.886
0.919
0.959
0.974
0.888
0.939
0.960
0.936
0.950
0.574
0.128
0.172
0.174
0.123
0.155
0.110
0.056
0.072
0.085
0.030
0.041
0.032
0.042
0.051
0.036
0.047
0.049
0.059
0.074
0.093
0.047
0.032
0.021
0.027
Partial Correlation
AC
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
0.278
0.272
0.192
0.193
0.217
0.154
0.141
0.145
0.074
0.106
0.107
0.127
0.115
0.124
0.120
0.137
0.133
0.091
0.148
0.076
0.126
0.144
0.105
0.188
0.088
0.120
0.142
0.142
0.120
0.117
0.119
0.106
0.081
0.076
0.087
0.073
PAC
0.278
0.211
0.084
0.089
0.118
0.026
0.022
0.047
-0.039
0.023
0.040
0.051
0.028
0.047
0.032
0.045
0.037
-0.022
0.063
-0.031
0.040
0.065
0.001
0.100
-0.029
0.013
0.046
0.042
-0.014
0.018
0.023
-0.007
-0.013
-0.015
0.007
-0.013
Q-Stat
Prob
216.92
425.28
529.00
633.28
765.20
832.06
888.03
947.14
962.58
994.10
1026.5
1072.0
1109.1
1152.2
1192.7
1245.3
1295.1
1318.5
1380.4
1396.7
1441.8
1500.7
1531.8
1632.3
1654.0
1694.7
1751.8
1808.9
1849.5
1888.4
1928.4
1960.4
1978.9
1995.3
2016.9
2031.9
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
models.
4
1.2 ARCH-models
The general setup for ARCH models is
yt = xt + ut
with xt = (x1t, x2t, . . . , xpt) , = (1, 2, . . . , p) ,
t = 1, . . . , T , and
ut|Ft1 N (0, ht),
where Ft is the information available at time
t (usually the past values of ut; u1, . . . , ut1),
and
ht = var(ut |Ft1) = + 1u2t1 + 2u2t2 + + q u2tq .
Properties of ARCH-processes
Consider (for the sake of simplicity) ARCH(1)
process
ht = + u2
t1
with > 0 and 0 < 1 and ut|ut1
N (0, ht).
(a) ut is white noise
(i) Constant mean (zero):
E[ut] = E[Et1[ut]] = E[0] = 0.
,
=0
..
= (1 + + 2 + + n)
]
+ n+1E[u2
, tn11
n
p 0, as
Q
n
i
= limn i=0
.
= 1
(iii) Autocovariances: cov(ut, ut+k ) = 0 for
all k = 0, because using again the law of
iterated expectations we get
E[u4
.
t] = 3
(1 )2 1 32
2 = 3 4,
=
3[var(u
)]
E[u4
t
t] > 3
u
(1 )2
ut = zt ht,
where the zt are independent standard normal
random variables (strict white noise). This
gives us a useful device to check after tting
an ARCH model the adequacy of the specication: Check the autocorrelations of the
squared standardized series.
Returning to the analogy with AR-processes, we note
that using standardized variables we may write
t = u2t ht = zt2ht ht = ht (zt2 1).
(yt xt )2
2h
t
e
T
11
+ k1ht+1
=
1
1 w
W
+ k1 ht+1
=
1
1
k
We note that for long forecast horizons k, the forecasts of the conditional variance ht+k approach the
unconditional variance u2 = /(1 ).
12