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
Day
c Professor
Seppo Pynn
onen, Department of Mathematics and Statistics, University of Vaasa, Box
700, 65100 Vaasa, phone: (06)324 8259 email:
sjp@uwasa., URL: http://www.uwasa./sjp/
1
-15
21.12.1999
Return
Log Index
Seppo Pynn
onen,
2008
nonstationary.
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
(2)
We denote ut W N (, u2 ).
Remark 1.1: Usually it is assumed in (2) that = 0.
Figure. Nasdaq Composite index autocorrelations for log levels and log differences (returns)
(1 L)yt = + (L)ut,
(5)
(L) = 1 + 1 L + 2 L + 3L +
such that
j=1 |j | < , (1) = 0, roots of (z) = 0
(4)
xt = + (L)ut ,
yt = + t + (L)ut.
(1 L)yt = + (L)u
t,
where
(8)
(L)
= (1 L)(L).
(1)
= (1 1)(1) = 0,
which does not satisfy the rule in Denition 1.3, and
hence a trend stationary series is not I(1).
0.011221
0.010868
1.091357
3338.542
-4224.341
1.997947
.11
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
Prob.
0.0002
0.0000
0.086119
1.097336
3.013434
3.017668
31.80961
0.000000
Inverted MA Roots
0.011323
0.010970
1.091301
3338.198
-4224.196
2.000051
Prob.
0.0002
0.0000
-.11
0.086119
1.097336
3.013331
3.017565
32.10153
0.000000
Because squared observations are the building blocks of the variance of the series, the
results suggest that the variation (volatility)
of the series is time dependent.
This leads to the so called ARCH-family of
models.
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
series.
The dependency, however, is nonlinear by nature.
yt = xt + ut
10
11
Properties of ARCH-processes
ht = + u2
t1
=0
=
...
(16)
= (1 + + 2 + + n)
]
+ n+1E[u2
tn1
n
0, as
i
= limn n
i=0
(14)
Var[ut] =
2
= E Et1[ut ]
E[ht] = E[ + u2
t1]
+ E[u2
t1 ]
E[u2
t]
.
= 1
(iii) Autocovariances: Exercise, show that
autocovariances are zero, i.e., E[utut+k ] = 0
for all k = 0. (Hint: use the law of iterated expectations.)
(15)
Et1 [Et2 [ut ]] = E [E[ut |Ft2 ]|Ft1 ] = E[ut |Ft1 ] = Et1 [ut ].
12
13
E[u4
t] = 3
2
1 2
.
(1 )2 1 32
14
2
= 3[Var(ut)]2,
2
(1 )
15
Write
(19)
ut
zt =
ht
(21)
yt = xt + ut
(20)
u t = zt ht ,
(22)
16
() =
t()
t=1
with
1
1
1
17
() = max
().
(25)
ht = + u2
t1 + ht1 ,
18
19
ht = +
r
j=1
j htj +
u2
ti
i=1
+
j1u2
ht =
tj
1
j=1
an ARCH() process.
Thus the GARCH term captures all the history from t 2 backwards of the shocks ut.
20
21
rt = + ut + ut1
ht = + u2t1 + ht1 .
Partial Correlation
0.007104
0.005685
1.094213
3352.444
-3775.938
2.129878
-.17
0.086119
1.097336
2.695856
2.706443
5.008069
0.000507
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
0.004
0.035
-0.007
-0.007
-0.009
0.001
-0.021
-0.023
-0.019
-0.016
-0.024
-0.008
-0.007
-0.004
0.005
-0.004
-0.008
-0.025
0.002
-0.030
0.000
-0.016
-0.004
0.030
-0.012
-0.012
0.011
-0.015
0.021
-0.002
-0.001
-0.008
0.037
-0.016
0.009
-0.013
PAC
0.004
0.035
-0.007
-0.008
-0.008
0.001
-0.020
-0.023
-0.017
-0.014
-0.023
-0.008
-0.006
-0.005
0.004
-0.005
-0.010
-0.027
0.001
-0.030
-0.002
-0.016
-0.005
0.030
-0.014
-0.016
0.010
-0.017
0.018
-0.004
-0.004
-0.008
0.036
-0.017
0.007
-0.013
Q-Stat
Prob
0.0519
3.5358
3.6727
3.8082
4.0159
4.0179
5.2184
6.7031
7.7149
8.4002
9.9573
10.148
10.278
10.324
10.405
10.448
10.639
12.405
12.421
14.903
14.903
15.671
15.710
18.231
18.627
19.046
19.387
20.051
21.360
21.372
21.377
21.542
25.370
26.108
26.352
26.810
0.060
0.159
0.283
0.404
0.547
0.516
0.460
0.462
0.494
0.444
0.517
0.592
0.667
0.732
0.791
0.831
0.775
0.825
0.729
0.782
0.788
0.830
0.745
0.772
0.795
0.820
0.829
0.810
0.845
0.876
0.897
0.791
0.797
0.823
0.838
8
4
0
-4
8
-8
-12
0
-4
-8
-12
2500
3000
3500
Residual
4000
4500
Actual
5000
Fitted
350
Series: Standardized Residuals
Sample 2276 5080
Observations 2805
300
250
Mean
Median
Maximum
Minimum
Std. Dev.
Skewness
Kurtosis
200
150
2
100
50
Jarque-Bera
Probability
0
-5.0
-2.5
0.0
-0.019280
0.050222
3.265924
-6.466048
0.999654
-0.731417
5.456912
955.6050
0.000000
2.5
0
2500
3000
3500
4000
4500
5000
22
23
The autocorrelations of the squared standardized residuals pass the white noise test.
(30)
ht = + u2
t1 + ht1 + xt .
24
25
ARCH-M Model
0.011379
0.009613
1.092049
3338.007
-3773.330
2.127550
-.17
0.086119
1.097336
2.694709
2.707413
6.443432
0.000006
Econometrica,
55, 391407.
26
27
rt = ht + ut + ut1
ht = + u2t1 + ht1 .
Below are the estimation results for the above model
0.010861
0.009448
1.092141
3339.759
-3774.146
-.17
0.086119
1.097336
2.694578
2.705165
2.132844
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
-0.003
0.034
-0.008
-0.009
-0.009
0.001
-0.022
-0.023
-0.018
-0.015
-0.023
-0.007
-0.006
-0.001
0.005
-0.003
-0.006
-0.025
0.006
-0.030
0.003
-0.016
-0.002
0.031
-0.011
-0.011
0.013
-0.015
0.022
-0.003
0.000
-0.006
0.041
-0.016
0.009
-0.013
PAC
-0.003
0.034
-0.008
-0.010
-0.009
0.001
-0.021
-0.023
-0.017
-0.014
-0.023
-0.007
-0.005
-0.002
0.003
-0.005
-0.008
-0.027
0.004
-0.030
0.000
-0.015
-0.004
0.031
-0.013
-0.015
0.011
-0.016
0.019
-0.004
-0.002
-0.006
0.040
-0.016
0.006
-0.012
Q-Stat
Prob
0.0213
3.2371
3.4198
3.6300
3.8611
3.8636
5.1942
6.6228
7.5391
8.1998
9.7327
9.8585
9.9568
9.9594
10.023
10.055
10.163
11.907
11.994
14.490
14.511
15.193
15.207
17.937
18.265
18.602
19.073
19.737
21.147
21.173
21.173
21.293
26.022
26.794
27.013
27.472
0.072
0.181
0.304
0.425
0.569
0.519
0.469
0.480
0.514
0.464
0.543
0.620
0.697
0.761
0.816
0.858
0.806
0.848
0.754
0.804
0.813
0.853
0.761
0.790
0.816
0.833
0.842
0.819
0.853
0.882
0.904
0.763
0.769
0.797
0.814
8
4
0
-4
8
-8
-12
0
-4
-8
-12
2500
3000
3500
Residual
4000
Actual
4500
5000
Fitted
4
2
350
300
250
-2
Mean
Median
Maximum
Minimum
Std. Dev.
Skewness
Kurtosis
200
150
-4
100
50
-6
Jarque-Bera
Probability
0
-5.0
-2.5
0.0
-0.056142
0.018376
3.239140
-6.557423
0.998195
-0.726040
5.409519
924.9855
0.000000
2.5
-8
2500
3000
3500
4000
4500
5000
Standardized Residuals
28
29
============================================================
Z,Z2(-i)
Z,Z2(+i)
i
lag
lead
============================================================
****|
|
****|
| 0 -0.3916 -0.3916
|
|
*|
| 1
0.0342 -0.0782
|
|
*|
| 2 -0.0055 -0.0842
|
|
|
| 3
0.0093 -0.0373
|
|
|
| 4
0.0066 0.0315
|
|
|
| 5
0.0134 -0.0046
|
|
|
| 6 -0.0134 -0.0019
|
|
|
| 7
0.0113 -0.0004
|
|
|
| 8 -0.0019 0.0045
|
|
|
| 9 -0.0034 0.0272
|
|
|
| 10 -0.0205 0.0128
============================================================
metry is to investigate the cross autocorrelations between standardized and squared standardized GARCH
residuals.
Below are the cross autocorrelations between the standardized and squared standardized residuals of the tted MA(1)-GARCH(1,1) model.
30
31
2
(32)ht = + u2
t1 + ut1dt1 + ht1 ,
32
33
ones.
2/ 2805 0.038.
Cross autocorrelations of the standardized and squared
standardized MA(1)-TARCH(1,1)-M model.
============================================================
Z,Z2(-i)
Z,Z2(+i)
i
lag
lead
-----------------------------------------------------------****|
|
****|
| 0 -0.3543 -0.3543
|
|
*|
| 1
0.0304 -0.0488
|
|
*|
| 2 -0.0071 -0.0537
|
|
|
| 3
0.0112 -0.0156
|
|
|*
| 4
0.0069
0.0545
|
|
|
| 5
0.0113
0.0080
============================================================
The goodness of t improves and the statistically signicant positive asymmetry parameter estimate indicates presence of leverage.
34
35
(33)
37
18
16
14
ht
12
10
8
6
4
2
0
-10
-8
-6
-4
-2
38
39
The model is
=
(35)
1
is the unconditional variance of the series.
ht qt = (u2
t1 qt1 ) + (ht1 qt1 )
qt = + (qt1 ) + (u2
t1 ht1 ).
(36)
An asymmetric version for the model is
ht qt = (u2
t1 qt1 )
+(u2
t1 qt1)dt1 + (ht1 qt1 )
qt = + (qt1 ) + (u2
t1 ht1 ).
(37)
40
41
43
yt = xtSt + ut,
t = 1, . . . , T ,
where
(39)
2 ),
ut NID(0, S
t
(40)
(41)
2 = 2 (1 S ) + 2 S ,
S
t
0
1 t
t
Usually it is assumed that the possible dierence between the regimes is a mean and/or a
volatility shift, but no change in the autoregression parameter.
and
(42)
St = 0 or 1,
That is,
(Regime 0 or 1).
Thus, under regime 1 the coecient parameter vector is 1 and error variance 12.
(43)
with
(44)
2 ),
ut NID(0, S
t
45
Denote then
(45)
with
1
2S2t
exp
2S2t
(47)
47
.
Then the log-likelihood function to be maximized with respect to the unknown parameters is
(48)
() =
t(),
t=1
where
t () = log
1
1
St =0 St1 =0
(49)
(50)
with
(52)
(53)
we can write
(54)
and the problem reduces to calculating (estimating) the time dependent state probabilities, P [St1|Ft1], and weight them with
the transition probabilities to obtain the joint
probability.
(51)
In order to evaluate the log-likelihood function we need to dene the joint probabilities
P [St, St1|Ft1].
48
49
t().
The maximum likelihood estimates for is
then obtained iteratively maximizing the likelihood function by updating the likelihood
function at each iteration with the above algorithm.
(56)
with
P [St = st |Ft] =
st1 =0
(57)
50
51
Smoothed probabilities
P [St = j|FT ],
j = 0, 1,
52
53
Expected duration
Parameter
0
1
02
12
p (regime 1)
q (regime 0)
Estimate
2.605
-3.277
13.56
20.82
0.857
0.866
Std err
0.964
1.582
3.34
4.79
0.084
0.097
1
(61) E[D] =
kP [D = k] =
.
1
p
jj
k=1
7.5 quarters.
54
55
yt = St + St xt + t ,
where St = 0 (1 St ) + 1 St, St = 0 (1 St ) + 1 St
10.0
Parameter
Estimate
0 (low)
-0.0068
1 (high)
0.0802
0 (low)
0.9679
1 (high)
1.8040
2
0.5225
0 (low)
1.7050
12 (high)
State Prob
P (High|High)
0.96417
P (Low|High)
0.03583
P (High|Low)
0.01728
P (Low|Low)
0.98272
P (High)
0.67471
P (Low)
0.32529
Log-likelihood -3208.438
7.5
Finnish Returns
5.0
2.5
0.0
-2.5
-5.0
-7.5
-10.0
-6
-4
-2
World Returns
Std Err
0.0178
0.0508
0.0215
0.0690
0.0198
0.0711
t-value
-0.39
1.57
45.04
26.15
26.37
23.96
p-value
0.700
0.114
0.000
0.000
0.000
0.000
56
57
58