Академический Документы
Профессиональный Документы
Культура Документы
Silva
WINTS09
1 / 17
Contents
Introduction
High-order statistics
WINTS09
2 / 17
Introduction
WINTS09
3 / 17
Introduction
WINTS09
3 / 17
X = E[Xt ],
WINTS09
3 / 17
CX = E[Xt ] = X
CX (k) = X (k) X 2 = R(k), k Z
CX (k, m) = X (k, m) X X (k) + X (m) + X (k m) +2X 3 , k, m Z
Introduction
WINTS09
4 / 17
CX = E[Xt ] = X
CX (k) = X (k) X 2 = R(k), k Z
CX (k, m) = X (k, m) X X (k) + X (m) + X (k m) +2X 3 , k, m Z
CX (m)
= CX (m),
m>0
m, n > 0
Introduction
WINTS09
4 / 17
Introduction
WINTS09
5 / 17
Introduction
WINTS09
5 / 17
WINTS09
5 / 17
k>0
k>0
m>k>0
WINTS09
6 / 17
k>0
k>0
m>k>0
st
nd
WINTS09
6 / 17
i=1
i=1
if a = 0,
otherwise,
WINTS09
7 / 17
i=1
i=1
C3,X =
if a = 0,
otherwise,
CX (0, 0)
CX (1, 1)
CX (0, 1)
.
..
CX (0, p 1)
CX (p 1, p 1)
CX (0, 0)
.
..
..
.
CX (p 2, p 2)
.
..
CX (0, p 2)
CX (0, 0)
.
.
.
p
CX (0, 1)
CX (0, 2)
=
.
..
CX (0, p)
= c3,X
WINTS09
7 / 17
WINTS09
8 / 17
2nd block
Bth block
(1)
(2)
(2)
(B)
(B)
{X1 , . . . , XM , X1 , . . . , XM , . . . , X1 , . . . , XM }
{z
}|
{z
}
{z
}
|
|
1st block
2nd block
Bth block
WINTS09
8 / 17
2nd block
Bth block
(1)
(2)
(2)
(B)
(B)
{X1 , . . . , XM , X1 , . . . , XM , . . . , X1 , . . . , XM }
{z
}|
{z
}
{z
}
|
|
1st block
2nd block
(i)
Bth block
1 M (i)
Xj
M j=1
1 Mk (i)
(i)
(i) 2
(i)
(i)
Xj X
C X (k, k) =
Xj+k X
,
M j=1
1 Mk (i)
(i) 2
(i)
(i)
(i)
Xj X
Xj+k X ,
C X (0, k) =
M j=1
Parameter estimation based on third-order cumulants
k = 0, . . . , p 1
k = 1, . . . , p
WINTS09
8 / 17
k = 0, . . . , p 1
1 B (i)
C X (0, k) = C X (0, k),
B i=1
k = 1, . . . , p
WINTS09
9 / 17
k = 0, . . . , p 1
1 B (i)
C X (0, k) = C X (0, k),
B i=1
k = 1, . . . , p
C X (0, 0)
C X (0, 1)
..
.
C X (1, 1)
C X (0, 0)
..
.
C X (0, p 1)
C X (0, p 2)
..
.
C X (p 1, p 1)
C X (p 2, p 2)
..
.
C X (0, 0)
1
2
..
.
p
C X (0, 1)
C X (0, 2)
..
.
C X (0, p)
WINTS09
9 / 17
k = 0, . . . , p 1
1 B (i)
C X (0, k) = C X (0, k),
B i=1
k = 1, . . . , p
C X (0, 0)
C X (0, 1)
..
.
C X (1, 1)
C X (0, 0)
..
.
C X (0, p 1)
C X (0, p 2)
e = X 1 i ,
i=1
..
.
C X (p 1, p 1)
C X (p 2, p 2)
..
.
C X (0, 0)
p
e2 = V p X i2 ,
i=1
1
2
..
.
p
C X (0, 1)
C X (0, 2)
..
.
C X (0, p)
V p = R(0)
i R(i)
i=1
WINTS09
9 / 17
WINTS09
10 / 17
WINTS09
10 / 17
The sample bias, variance and mean square error decrease as the sample size (of
the block) increases
WINTS09
10 / 17
The sample bias, variance and mean square error decrease as the sample size (of
the block) increases
For small sample size: evidence of departure from symmetry in the marginal
distributions, specially for values of the parameter near the non-stationary region
WINTS09
10 / 17
The sample bias, variance and mean square error decrease as the sample size (of
the block) increases
For small sample size: evidence of departure from symmetry in the marginal
distributions, specially for values of the parameter near the non-stationary region
Coefficients underestimated; innovation parameter overestimated
WINTS09
10 / 17
The sample bias, variance and mean square error decrease as the sample size (of
the block) increases
For small sample size: evidence of departure from symmetry in the marginal
distributions, specially for values of the parameter near the non-stationary region
Coefficients underestimated; innovation parameter overestimated
Fixed M statistics measures decrease as B increases and vice-versa
Monte Carlo results and application to real data
WINTS09
10 / 17
N=200
Bias( )
1
0
1
2
TOR_1B
TOR_2B
TOR_4B
TOR_1B
TOR_2B
TOR_4B
TOR_1B
TOR_2B
TOR_4B
TOR_1B
TOR_2B
TOR_4B
TOR_1B
TOR_2B
TOR_4B
TOR_1B
TOR_2B
TOR_4B
Bias( )
1
0
1
2
Bias()
Figure: Boxplots of the sample bias for the estimates obtained in 1000 realizations of 60 and 500 observations of the
INAR(2) model: Xt = 0.6 Xt1 + 0.1 Xt2 + et , where et Po(1).
Monte Carlo results and application to real data
WINTS09
11 / 17
(, )=(0.9, 3.0)
0.8
Bias()
0.4
0.4
0.8
WHT
CLS
LS_HOS
YW
TOR_1B
WHT
CLS
LS_HOS
YW
TOR_1B
WHT
CLS
LS_HOS
YW
TOR_1B
WHT
CLS
LS_HOS
YW
TOR_1B
Bias()
2.5
2.5
Figure:
Boxplots of the sample bias for the estimates obtained in 1000 realizations of 200 observations of the INAR(1)
models: Xt = 0.4 Xt1 + et and Xt = 0.9 Xt1 + et , where et Po(1).
Monte Carlo results and application to real data
WINTS09
12 / 17
Number of plants
40
35
30
25
20
15
10
5
1920
Figure:
1930
1940
1950
1960
1970
1981
The number of Swedish mechanical paper and pulp mills, from 1921 to 1981 [Brnns (1995) and Brnns and
Hellstrm (2001)]
Monte Carlo results and application to real data
WINTS09
13 / 17
WINTS09
14 / 17
e2
x2
MSE
CLS
0.9591
0.2017
15.2268
4.9279
192.2764
9.3254
LS_HOS
0.9269
1.3635
19.2253
18.6516
145.4513
9.2997
TOR_1B
0.9631
0.7518
14.7219
20.374
213.5073
8.9224
WINTS09
14 / 17
e2
x2
MSE
CLS
0.9591
0.2017
15.2268
4.9279
192.2764
9.3254
LS_HOS
0.9269
1.3635
19.2253
18.6516
145.4513
9.2997
TOR_1B
0.9631
0.7518
14.7219
20.374
213.5073
8.9224
e
(1 )( e + e2 )
and x2 =
1
(1 )2 (1 + )
WINTS09
14 / 17
e2
x2
MSE
CLS
0.9591
0.2017
15.2268
4.9279
192.2764
9.3254
LS_HOS
0.9269
1.3635
19.2253
18.6516
145.4513
9.2997
TOR_1B
0.9631
0.7518
14.7219
20.374
213.5073
8.9224
e
(1 )( e + e2 )
and x2 =
1
(1 )2 (1 + )
MSE between the observations and the fitted models based on TOR_1B, LS_HOS and
CLS estimates
Monte Carlo results and application to real data
WINTS09
14 / 17
45
40
Number of plants
35
30
25
20
15
10
5
1920
Figure:
1930
1940
1950
years
1960
1970
1981
The number of plants and the fitted values considering the TOR_1B and CLS estimates
Monte Carlo results and application to real data
WINTS09
15 / 17
Final remarks
Advantage of HOS: capability to detect and characterize the deviations from
Gaussianity and non-linearity of the processes
Final remarks
WINTS09
16 / 17
Final remarks
Advantage of HOS: capability to detect and characterize the deviations from
Gaussianity and non-linearity of the processes
INAR processes are non-Gaussian
Final remarks
WINTS09
16 / 17
Final remarks
Advantage of HOS: capability to detect and characterize the deviations from
Gaussianity and non-linearity of the processes
INAR processes are non-Gaussian
Final remarks
WINTS09
16 / 17
Final remarks
Advantage of HOS: capability to detect and characterize the deviations from
Gaussianity and non-linearity of the processes
INAR processes are non-Gaussian
Final remarks
WINTS09
16 / 17
Final remarks
Advantage of HOS: capability to detect and characterize the deviations from
Gaussianity and non-linearity of the processes
INAR processes are non-Gaussian
WINTS09
16 / 17
References
MENDEL, J. M. (1991).
BRNNS, K. (1995).
Explanatory Variables in the AR(1) Count Data Model.
Ume Economic Studies 381.
SILVA, I. (2005).
series.
processus GENAR(p).
49-71.
LATOUR, A. (1998).
References
WINTS09
17 / 17