Professional Documents
Culture Documents
AN ADAPTATION OF
BOX-JENKINS METHODOLOGY
Joseph N Ladalla
University of Illinois at Springfield, Springfield, IL
KEYWORDS:
Multivariate time series, Box-Jenkins ARIMA
models, Principal components, Forecasting,
Eigenvalues, Eigenvectors.
SUMMARY:
A time domain analysis of Multivariate Time
series is suggested which exploits Box-Jenkins
methodology after the given time series data is
translated into the principal components based on
the dispersion matrix.
After fitting suitable
models for each of the principal components,
these models may be converted into
multivariate ARIMA models for the original
data. Similarly the forecasts for the individual
principal components may be put together into a
matrix of forecasts to obtain forecasts for the
original multivariate time series. We apply the
method on two examples - one simulated and the
other real data from Gregory C. Reinsel.
As a byproduct of this approach an
elegant method of determining the order of the
ARIMA model is provided for multivariate time
series.
(1.3) Xt = 1Xt-1 + at
where,
T
1 = P P , and = diag {1,1 , 2,1 ,.. , k,1 }
and at has MVN (0, ) with = PTDP;
2
2
2
D = diag( 1 , 2 ., k )
2.
MODEL ESTIMATION.
XP = W
.
(2.2)
Xt P = Wt
j = 1,2,, k
are
(2.3)
Wm
where P
matrix P.
= W m PT
0.234943 0.863473
P = 0.904492 -0.362364
0.355943 0.350865
-0.446336
-0.224914
0.866140
(4.1)
ut = 1 + 1 ut-1 + at - 1 at-1
vt = 2 + 2 vt-1 + bt - 2 bt-1
wt = 3 + 3 wt-1 + ct - 3 ct-1
(3.1)
Equivalently,
wt = + wt-1 + bt + bt-1
where,
T
= ( 1 2 3 )
T
bt = (at bt ct )
(3.2)
1
=
0
0
0 0
2 0
0 3
1
=
0
0
wt = + 1 wt-1 + 12 wt-12 + bt
(4.3)
0 0
12 = diag( 0
2 0
0 3
T
bt = (at bt ct )
0.23957)
and bt is N(0, D)
(3.4)
(4.4)
where,
T
T
= P ; = P P ; = P P ; at = Pbt;
xt = + 1 xt-1 +12xt-12 + at
where,
at = Pbt ; at is N(0, );
T
From (3.3) , (3.4) it follows that at is N( 0, PDP )
1 = P 1P
4. EXAMPLES:
Example 1.
T
12 = P 12P =
64.87
254.64
72.84
64.87
72.84
59.74
The matrices of 5-step ahead forecasts for U, V
and W, together with lower and upper 95%
confidence limits are given below:
a) Forecasts:
t
101
102
103
104
105
173.8353
179.0296
183.1772
186.4890
189.1335
64.9940
59.9175
57.9993
57.2746
57.0007
-30.8700
-31.9553
-32.4292
-33.2898
-33.4694
101
102
103
104
105
140.5930
136.4901
135.6501
136.0387
136.9046
41.4616
34.7612
32.6197
31.8633
31.5849
c)
W
-36.2033
-37.9833
-38.6362
-39.5456
-39.7386
101
102
103
104
105
207.0776
221.5692
230.7044
236.9394
241.3623
88.5265
85.0737
83.3789
82.6859
82.4166
W
-25.5367
-25.9273
-26.2221
-27.0340
-27.2001
101
102
103
104
105
forecast
110.74
108.06
107.59
108.13
108.59
forecast
101
102
103
104
105
140.62
147.41
151.96
155.41
157.94
forecast
101
102
103
104
105
57.94
57.07
57.46
57.64
58.33
136.49
137.09
137.90
139.13
140.00
113.14
99.11
110.46
97.46
110.92
Y
conf. limits
lower
upper
120.28
119.40
119.56
120.39
121.32
160.96
175.40
184.35
190.43
194.52
33.23 82.65
27.88 86.26
26.26 88.66
25.35 89.93
25.39 91.25
63.15
49.63
53.58
44.87
52.02
Example 2.
Here we consider a real life example:
Monthly flour price indices for three U.S. cities .
(c.f Gregory C. Reinsel (l997), Table 9) We use
the first 95 of the 100 observations for analysis
holding the last 5 observations to compare with
the forecasts.
Since the procedure is a repetition of
the one adopted in the previous example, we
skip details to look only at the relevant facts:
Let
Observed
142.96
139.53
153.54
138.49
132.82
X
conf. limits observed
lower
upper
84.99
79.04
77.28
77.13
77.17
Z
conf. limits observed
lower upper
P2
P1
P=
P3
0.529453 0.513652
0.573376 0.369889
0.625236 -0.774173
eigenvalues
Proportion of
variation
2425.77
0.9799
0.675160
-0.731041
0.098677
43.72
0.0177
6.01
0.0024
W = XP
96
97
98
99
100
156.089
146.085
137.706
132.072
126.444
174.741
174.475
173.292
173.190
172.415
193.394
202.866
208.879
214.307
218.386
observed
187.5
190.7
190.4
192.4
192.9
where,
1 = diag (1.2363
2= diag ( 0.2363
3= diag (
0
7= diag (
0
0.7442 0.5750
0.2558
0
0.2064
0
0.2204
0
D= diag ( 184.42 8.56 1.54
)
)
)
)
)
Y
95%
95%
t lower limit forecast upper limit
96
97
98
99
100
0.1516
0.1886
0.9349
166.782
166.097
165.155
164.823
164.082
182.358
191.410
197.203
202.180
206.005
Z
95%
95%
lower limit forecast upper limit
179.8
179.0
179.2
181.4
181.8
observed
3 = P 3P ;
0.2329
0.8156
0.1886
151.205
140.784
133.107
127.466
122.159
observed
96
97
98
99
100
162.095
151.901
145.806
139.090
134.239
174.505
173.597
173.718
172.382
172.008
186.915
195.293
201.630
205.674
209.777
178.2
182.0
188.6
190.8
192.2
CONCLUSIONS:
APPENDIX:
Simulated trivariate time series data:
X
106.405
103.199
125.213
149.28
118.061
135.681
123.609
119.67
107.386
103.184
114.094
120.054
84.621
99.35
104.366
125.856
93.452
102.997
115.066
128.531
101.679
103.118
91.619
106.05
100.201
113.581
126.056
108.847
97.912
83.83
107.601
110.36
97.757
104.182
114.859
105.747
93.911
94.041
98.611
100.201
88.707
77.038
Y
190.627
186.567
216.202
192.67
182.183
178.08
183.006
169.606
179.878
155.704
144.206
166.466
155.456
131.834
121.94
109.621
117.757
124.693
156.691
132.815
129.522
138.485
145.72
142.91
147.96
152.793
154.536
165.975
136.182
143.16
145.963
153.876
163.029
148.128
157.864
152.697
149.361
145.982
138.33
107.725
135.912
147.826
Z
69.8236
64.0717
82.9996
86.1037
73.3294
79.8028
73.1775
71.5903
69.5109
59.1079
52.309
64.2461
46.125
45.9178
43.8563
49.5452
44.4521
48.3801
65.4364
66.0735
49.5604
49.2726
44.2546
51.4856
53.6439
58.5262
65.6212
58.81
47.0019
45.97
65.1901
70.4322
59.213
53.7676
58.5882
53.1915
45.1395
45.5159
48.2072
38.5165
47.7301
42.6818
88.54
101.416
115.772
101.246
114.453
113.326
125.198
126.252
120.437
126.726
110.813
112.05
108.752
106.33
118.326
124.521
124.204
108.119
97.963
93.941
89.856
85.456
99.458
101.673
117.352
117.677
114.288
121.491
113.204
100.379
115.707
111.279
105.084
107.303
115.289
125.418
104.866
99.596
116.503
106.761
100.369
115.777
107.297
110.316
96.653
84.422
114.216
109.858
Y
154.424
138.881
162.512
186.815
192.601
190.862
222.856
209.395
201.562
232.348
206.013
194.48
190.653
186.528
200.918
172.932
172.396
158.274
172.891
148.599
153.749
144.248
145.553
154.751
143.449
140.059
163.917
162.065
169.533
167.589
182.598
175.07
167.03
177.034
183.738
175.354
166.7
139.441
118.784
158.826
177.363
196.3
200.037
190.499
179.838
197.826
205.546
187.162
Z
50.2841
56.4294
71.8461
68.8396
79.7266
75.5322
86.5211
80.2245
70.4401
83.1272
71.6925
73.5274
65.5909
62.7813
71.8392
73.1649
74.4356
59.9918
59.4257
53.2992
50.943
46.2048
51.9204
55.9632
56.7054
52.1146
60.2398
61.0048
63.6008
54.1047
63.7947
58.8569
61.8931
66.3935
73.8751
72.9356
58.9892
44.6021
54.6768
53.6773
59.6366
71.5243
73.8049
65.213
61.2701
55.4614
70.3773
65.9143
X
124.751
122.008
125.239
104.81
119.74
126.335
140.156
116.937
120.667
119.445
113.135
99.113
110.462
97.458
110.919
Y
156.648
160.353
197.26
167.966
174.405
168.119
139.09
160.776
167.939
129.182
142.962
139.533
153.538
138.493
132.818
Z
66.1368
62.7405
75.3086
60.6377
71.4816
71.8745
76.9747
66.2836
69.1031
62.9973
63.149
49.6332
53.5748
44.8671
52.0201
REFERENCES
1.
2.
3.