Você está na página 1de 5

Economics Letters 114 (2012) 353357

Contents lists available at SciVerse ScienceDirect


Economics Letters
journal homepage: www.elsevier.com/locate/ecolet
On general periodic time-varying bilinear processes
Abdelouahab Bibi

, Ines Lescheb
Dpartement de Mathmatiques, Universit Mentouri Constantine, Algeria
a r t i c l e i n f o
Article history:
Received 11 October 2011
Accepted 15 November 2011
Available online 25 November 2011
JEL classification:
C22
C19
C13
Keywords:
Periodic models
Bilinear processes
Periodic bilinear processes
a b s t r a c t
We study in this note the class of bilinear processes with periodic time-varying coefficients. We give
necessary and sufficient conditions ensuring the existence of strict and second order stationary solutions
(in periodic sense) and for the existence of higher order moments. The given conditions can be applied
to periodic ARMA or periodic GARCH models. The central limit theorem and the law of iterated logarithm
(LIL) for higher order sample moments are showed.
2011 Elsevier B.V. All rights reserved.
1. The model and its Markovian representation
Bilinear time series models, introduced by Granger and
Andersen (1978), are certainly among the most popular nonlinear
models that are capable to model with success the non-Gaussian
data. The stupefying development of these models is based
on time-invariant parameter assumption. In many applications,
this assumption maybe not appropriate when the series to
be modelled exhibits structural changes and non-stationarity
behaviour. However, one possible way is to consider a bilinear
model whose coefficients may arbitrarily vary with respect to
time according to some specific structures. Periodic or almost
periodic coefficients are perhaps the most popular structures used
in time series exhibiting structural changes in regime at known
dates.
Inthis paper, we consider a general bilinear model (X
t
)
tZ
, Z :=
{0, 1, 2, . . . , } defined on some probability space (, , P)
which exhibits periodically time-varying coefficients, i.e.,
X
t
=
p

i=1
a
i
(t)X
ti
+
q

j=0
b
j
(t)e
tj
+
P

i=1
Q

j=1
c
ij
(t)X
ti
e
tj
(1.1)
denotes by PBL (p, q, P, Q) where (e
t
)
tZ
is a sequence of indepen-
dent and identically distributed (i.i.d.) random variables defined

Corresponding author.
E-mail addresses: a.bibi@umc.edu.dz (A. Bibi), i.lescheb@gmail.com(I. Lescheb).
on the same probability space (, , P) with E
_
log
+
|e
t
|
_
<
+ where log
+
(x) := max {log(x), 0} , x > 0 and e
l
is inde-
pendent of X
k
, k < l. The parameters (a
i
(t))
1ip
, (b
i
(t))
0iq
and
_
c
ij
(t)
_
1iP,1jQ
are periodic in t with period s, i.e., a
i
(t) =
a
i
(t +ks), b
i
(t) = b
i
(t +ks) and c
ij
(t) = c
ij
(t +ks) for any (t, k)
Z Z. Of course, the model (1.1) includes some popular models,
namely, periodic ARMA (PARMA) models and a large class of peri-
odic GARCH models (see Kristensen, 2009andBibi andLessak, 2009
for further discussion). These models are globally non-stationary,
but are stationary within each season, are becoming an appeal-
ing tool for investigating both non-Gaussianity and distinct sea-
sonal patterns appearing for instance in hydrology, climatology,
econometrics, etc. . . , and continue to gain a growing interest of
researchers.
By introducing the state r = (p + q)-vector X
t
:= (X
t
, X
t1
,
. . . , X
tp+1
, e
t
, . . . , e
tq+1
)

, and assuming that, without loss of


generality, that P = p, Eq. (1.1) can be expressed in the following
Markovian representation X
t
= H

X
t
and
X
t
= A
t
X
t1
+e
t
b
t
(1.2)
where H = (1, 0, . . . , 0)

R
r
, A
t
:= A
0
(t) +

Q
j=1
A
j
(t)e
tj
,
_
A
j
(t), b
t
_
0jQ
is a sequence of appropriate pairs of periodic
squared r r-matrices A
j
(t) and r-vector b
t
. Eq. (1.2) is the
same as the defining relation for random coefficient periodic
autoregressive models (see Franses and Paap, 2011 and Aknouche
and Guerbyenne, 2009).
0165-1765/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.econlet.2011.11.013
354 A. Bibi, I. Lescheb / Economics Letters 114 (2012) 353357
In this paper, we are interested first in the existence of causal
solutions i.e., solutions such that X
t
is measurable with respect to

(e)
t
:=
_
e
tj
, j 0
_
and its probabilistic properties. Second, we
focus on some interesting statistical properties such as the central
limit theorem and the law of iterated logarithm of higher order
empirical moments.
Some notations are used throughout the paper: I
(n)
is the n n
identity matrix, for any n m matrix M =
_
m
ij
_
, and ]0, 1],
define a matrix operator |M|

as |M|

:=
_

m
ij

_
. It is obvious
that

MX

|M|

for all X R
m
and

i
M
i


i
|M
i
|

where M N means m
ij
n
ij
for all i and j and N =
_
n
ij
_
. is the
usual Kronecker product of matrices and M
r
= M M
M r-times. If M is squared matrix (M) represents the maximum
modulus of the eigenvalues of the matrix M.
2. Strict and second order periodic stationarity
Iterating (1.2) s-times to get the following equation
X
t
= A(t)X
ts
+(t) (2.1)
in which A(t) =

s1
i=0
A
ti
and (t) =

s1
k=0
_

k1
i=0
A
ti
_
b
tk
e
tk
where, as usual, empty products are set equal to I
(r)
. It is worth
noting that since (e
t
)
tZ
is an i.i.d. process,
_
A(t), (t)
_
tZ
is a
strictly stationary and ergodic process. Processes similar to
_
X
t
_
tZ
in (2.1) have been examined by Brandt (1986), Bougerol and
Picard (1992) and Basrak et al. (2002) and thus we may apply
their results to derive conditions for the existence of the so-
called periodically stationary and ergodic solutions to (1.2) and
hence to (1.1). For this purpose, let . denote any operator
norm on the sets of r r and r 1 matrices. Define the top-
Lyapunov exponent associated with the strictly stationary and
ergodic sequence of random matrices A = (A(t))
tZ
by
(s)
(A) =
inf
t>0
1
t
E
_
log
_
_
_

t1
j=0
A(s(t j))
_
_
_
_
. Since, both E
_
log
+
A(t)
_
and E
_
log
+
_
_
_(t)
_
_
_
_
are finite then, from Brandt (1986), (see
Bougerol and Picard, 1992), the unique, causal, strictly periodically
stationary (s.p.s) and periodically ergodic (p.e) solution of (2.1)
is given by X
(1)
t
=

k0
_

k1
j=0
A(t js)
_
(t ks) whenever

(s)
(A) < 0. It is worth noting that
(s)
(A) < 0 (A) < 0
where (A) = inf
n>0
1
n
E
_
log
_
_
_

n1
j=0
A
tj
_
_
_
_
and thus (1.2) admits
a solution given by X
(2)
t
=

k0
_

k1
j=0
A
tj
_
b
tk
e
tk
which can
be expressed in Volterra expansion as follows
X
(2)
t
= b
t
e
t
+

r=1

l
1
,...,l
r
1

0k
i
=k
i+1
q
l
1
1

i
1
=0
A
k
1
(t i
1
)
l
r
1

i
r
=0
A
k
r
(t i
r
)b
t|l|
e
(l,k)
(2.2)
where e
(l,k)
=

{i:k
i
=0}

l
i
1
s=0
e
tk
i
|l
(i1)
|s
e
t|l
(r)
|
, l
(r)
= (l
1
, . . . ,
l
r
) and

l
(r)

r
i=1
l
i
.
For
(s)
(A) < 0 to be also necessary, the model (1.2) fur-
thermore, needs to be controllable and observable. Recalling here
that Eq. (1.2) is said to be controllable (respectively observable)
if rank {C
r
(t)} := r (resp. rank {O
r
(t)} := r) where C
r
(t) :=
_

1
(t)
.
.
.
2
(t)
.
.
.
.
.
.
r
(t)
_
(resp. O
r
(t) := [O
1
(t)
.
.
.O
2
(t)
.
.
.
.
.
.O
r
(t)])
with
1
(t) = b
t
and
j
(t), j 2 are defined recursively by

j
(t) =
_
A
i
(t)
j1
(t 1), i = 0, . . . , q
_
, 2 j r (resp. O
1
(t) =
H

, O
j
(t) = [O
j1
(t 1)A
i
(t), 0 = 1, . . . , q], 2 j r). This
discussion is summarized in the following theorem.
Theorem 2.1. Let
_
X
t
_
tZ
be the process defined by (2.1). Then
1.
(s)
(A) < 0 is a sufficient condition for that (2.1) (respectively
(1.2)) to have an unique, causal, s.p.s and p.e solution given
by the series X
(1)
t
(respectively by the series X
(2)
t
). Moreover the
series X
(1)
t
, X
(2)
t
converges absolutely a.s, X
(1)
t
= X
(2)
t
a.s, and the
process
_
_
X

st
, X

st+1
, . . . , X

s(t+1)1
_

_
tZ
is strictly stationary.
2. If furthermore, the model (1.2) is controllable and observable then

(s)
(A) < 0 is a necessary condition of the existence of s.p.s
solution for (1.2).
Proof.
1. The sufficient condition is a consequence of Theorem 1
of Brandt (1986) and Theorem 1.1 of Bougerol and Picard
(1992). On the other hand, for any n 1, we set X
(1)
t
(n)
=

n
k=0
_

k1
j=0
A(t js)
_
(t ks) and X
(2)
t
(n) =

n
k=0
_

k1
j=0
A
tj
_
b
tk
e
tk
, 1 v s. Then it can be shown that
for any v {0, . . . , s 1}
X
(2)
t
(sn +v) = X
(1)
t
(n) +
sn+v

k=sn+1
_
k1

j=0
A
tj
_
b
tk
e
tk
= X
(1)
t
(n) +
n

j=0
A(t js)

v1

k=0
_
k+1

j=1
A
tsnj
_
b
tk1
e
tksn1
.
Thus
_
_
_X
(1)
t
(n) X
(2)
t
(sn +v)
_
_
_
_
_

n
j=0
A(t js)
_
_
.K(t) 0
as n +for some sequence K(t) of positive and bounded
random variables. For the last statement, see Bibi and Lessak
(2009).
2. If there exists a causal s.p.s solution of (1.2) given by the
series X
(2)
t
, then lim
k

_
H


k1
i=0
A
ti
_
b
tk

= 0, a.s, for all


t Z. By controllability and observability condition we obtain
lim
t
_
_
_

t1
i=0
A(s(t i))
_
_
_ = 0, a.s, and the result follows
from Lemma 3.4 of Bougerol and Picard (1992).
Example 2.1. For PBL (1, 0, 1, 1) model, with b
0
(t) = 1, we obtain

s1
i=0
A
ti
=

s1
i=0
(a
1
(t i) +c
11
(t i)e
ti
). Hence PBL (1, 0, 1, 1)
admit a causal s.p.s and p.e solution if and only if
(s)
(A) :=

s1
i=0
E{log(|a
1
(i) + c
11
(i)e
0
|)} < 0. It is worth noting that the
existence of explosive regimes (i.e., E{log(|a
1
(i)+c
11
(i)e
0
|)} > 0)
does not preclude the existence of s.p.s solution.
Example 2.2 (PARMA Models). For a PARMA (p, q) model with
b
0
(t) = 1, the sufficient condition reduces to
_

s1
i=0
A
0
(i)
_
<
1. The last condition becomes also necessary if the model is
controllable and observable.
Now, we focus our attention on necessary and sufficient con-
ditions ensuring the existence of the square-integrable solution
for (1.1). A sufficient condition was given recently by Bibi and
Lessak (2009). Necessary conditions for the square-integrability
solution of general PBL (p, q, p, Q) models have not been obtained
A. Bibi, I. Lescheb / Economics Letters 114 (2012) 353357 355
to our best knowledge. These solution processes, when exist, be-
long to the class of periodically correlated (p.c) processes charac-
terized by E {X
t+s
} = E {X
t
} and Cov (X
t+s
, X
k+s
) = Cov (X
t
, X
k
)
for all t, k (see Gladyshev, 1961). Periodically correlated processes
have been received a particular interest because of their connec-
tion with multivariate second order stationary processes. Indeed,
define the v-th component of the s-dimensional vector X(t) by
X
t
(v) := X
st+v
, v = 0, . . . , s 1 and noting (v) = E {X
t
(v)} and

v
(h) = Cov (X
t
(v), X
t
(v h)) for all nonnegative h. Then the pro-
cess (X
t
)
tZ
is a p.c of period s if, and only if,
_
X(t)
_
tZ
is a second
order stationary process. However, in the special case when Q = 1
and q = 0 with b
0
(t) = 0, Bibi and Moon-Ho (2006) showed that
if E {e
t
} = 0 and E
_
e
2
t
_
=
2
, then PBL(p, 0, p, 1) has a causal p.c
solution whenever

(1)
=
_
s1

v=0
_
A
2
0
(v) +
2
A
2
1
(v)
_
_
< 1. (2.3)
In what follows, we show how to derive a condition analogous to
(2.3) for the existence of a p.c solution of general model (1.1) or
equivalently for (1.2). The following theorem gives sufficient and
necessary conditions for the existence of a p.c process solution of
Eq. (1.2).
Theorem 2.2. Assuming that
E
_
e
2
t
_
=
2
, E
_
e
2(Q+1)
t
_
< + and
E
_
e
2j+1
t
_
= 0 for any 0 j Q.
(2.4)
If

(Q)
=
_
s1

v=0
_
A
2
0
(v) +
2
Q

j=1
A
2
j
(v)
__
<
1
Q
(2.5)
then the infinite series (2.2) converging in mean square and absolutely
a.s and constitute the unique, causal, s.p.s and p.e solution to (1.2).
Conversely, if (1.2) is controllable and observable and has a causal
p.c solution, then

(Q)
=
_
s1

v=0
_
A
2
0
(v) +
2
Q

j=1
A
2
j
(v)
__
< 1. (2.6)
Sketch of proof of Theorem 2.2. To prove the theorem, we use
the same approach as in Ispany (1997). We first define the
equivalence relationship on the pair of the process e
(l,k)
by
(l, k) (l

, k

) E
_
e
(l,k)
e
(l ,

)
_
= 0 and let Q
_
l, k
_
=

__
l

, k

_
:
_
l, k
_

_
l

, k

__
. It can be seen that (l, k) (l

, k

and hence Q
_
l, k
_
Q
|k|
. With this notation, we have
from (2.2) and after some tedious algebra
E
_
X
2
t
_
K

l,k
Q
|k|

l
1
1

i
1
=0
A
k
1
(t i
1
)
l
r
1

i
r
=0
A
k
r
(t i
r
)b
t|l|

2

k
i
=0
k
i
,
and the result follows from Lemma 4.3 from Ispany (1997).
Corollary 2.1. For the model PBL(1, 0, 1, 1). The sufficient Condi-
tion (2.5) reduces to

s1
v=0
(a
2
1
(v) +
2
c
2
11
(v)) < 1 which is also nec-
essary if b
0
(t) = 0 for the existence of a causal, p.c and p.e., solution.
Proof. In this case, Q = 1, A
0
(t) = a
1
(t) and A
1
(t) = c
11
(t),
so

s1
v=0
_
A
2
0
(v) +
2
A
2
1
(v)
_
=

s1
v=0
_
a
2
1
(v) +
2
c
2
11
(v)
_
. If
b
0
(t) = 0 the model is controllable and hence the sufficient
Condition (2.5) is also necessary.
Corollary 2.2. For the PARMA (p, q) case, Condition (2.5) reduces to

s1
v=0
A

0
(v)
_
< 1.
Proof. Straightforward and hence omitted.
For the so-called subdiagonal PBL(p, q, p, Q) processes i.e.,
when c
ij
(t) = 0 if i < j in (1.1) an elegant Markovian represen-
tation can be given for which some interesting properties are es-
tablished. Indeed, Eq. (2.1) can be split up into
X
t
= A
0
(t)X
t1
+
Q

j=1
A
j
(t)X
tj
e
tj
+e
t
b
t
(2.7)
where in (2.7) the first row of matrices A
j
(t) becomes (c
jj
(t), . . . ,
c
pj
(t), 0, . . . , 0), j = 1, . . . , Q. By a simple extension of Pham
(1985) we obtain the following Markovian representation X
t
=
H

Z
t
and
Z
t
= M
t
(e
t
) Z
t1
+b
t
(e
t
) (2.8)
where Z
t
:= (X

t
, e
t
X

t
, e
t1
X

t1
, . . . , e
tQ+1
X

tQ+1
)

, H R
r
with r = (p +q) (Q +1) and M
t
(e
t
) and b
t
(e
t
) are finite order
polynomials in e
t
. More precisely M
t
(e
t
) = M
1
(t) + e
t
M
2
(t) and
b
t
(e
t
) = e
t
b
1
(t) + e
2
t
b
2
(t) with
_
b
1
(t), b
2
(t)
_
and {M
1
(t), M
2
(t)}
are periodic appropriate vectors of R
r
and r r matrices. For
the representation (2.8) we keep the same notation as in (2.1),
i.e., Z
t
= M(t)Z
ts
+ (t) where M(t) =

s1
i=0
M
ti
(e
ti
) and
(t) =

s1
k=0
_

k1
i=0
M
ti
(e
ti
)
_
b
tk
(e
tk
). Hence, we have the
following corollary.
Corollary 2.3. Let
(s)
(M) be the Lyapunov exponent associated with
respect to the sequence of random matrices M := (M(t))
tZ
and
assume that

:= E
_
|e
0
|

_
< + for some > 0. Then if

(s)
(M) < 0 there is

]0, 1[ such that E


_
|X
t
|

_
< +.
Proof. The proof is the same as in Basrak et al. (2002) by choosing
a norm. satisfying M

_
_
|M|

_
_
for some > 0.
Proposition 2.1. Suppose that
2(m+1)
< for some integer m
1. Then the following statements are equivalent: i. E
_
Z
2m
t
_
<
+, ii.
_

s1
v=0
E
_
(M
1
(v) +e
o
M
2
(v))
2m
_
_
< 1.
Proof. The proof follows closely the arguments of Theorem 2.2.

Proposition 2.2. Assume that E


_
e
2
0
_
=
2
< +, then (

s1
v=0
(M
2
1
(v) +
2
M
2
2
(v))) < 1 is a sufficient condition for both (2.8)
and (2.7) have an unique, causal, p.c solutions given by Z
t
=

k0
_

k1
j=0
M
tj
(e
tj
)
_
b
tk
(e
tk
) and X
t
= H

Z
t
where the
above series converges in mean square and absolutely a.s. The
variancecovariance matrix
v
(0) = E{(Z
t
(v) (v))(Z
t
(v)
(v))

} is given by

2
v
(0) =
_
M
2
1
(v) +
2
M
2
2
(v)
_

2
v1
(0) +
2
v
(0) (2.9)
where
v
(0) is the covariance matrix of e
t
(v)b
1
(v) + (e
2
t
(v)

2
)b
2
(v)+e
t
(v)M
2
(v)(v). Moreover the solution process is unique,
s.p.s and p.e.
356 A. Bibi, I. Lescheb / Economics Letters 114 (2012) 353357
Conversely, a necessary condition for the existence of a p.c solution
is that there exists a covariance matrix
v
(0) solution of Eq. (2.9).
Proof. Straightforward and hence omitted.
3. Limit theorems
The central limit theorem (CLT) and the law of the iterated
logarithm (LIL) concerning the sample moments of bilinear
processes have been established by Liu (1992) using some results
originated by Hall and Heyde (1980) in the context of stationary
and ergodic processes. This technique can be adapted to our model.
For this purpose and throughout this section, we shall assume
that the causality assumptions (2.4) and (2.5) are satisfied. The
following theorem (cf. Hall and Heyde, 1980, Theorem 5.4 and
Corollary 5.4), plays an important role in establishing both CLT
and LIL for sample moments of periodic bilinear processes. For
the model (1.1) defined by the first component of
_
X
t
_
tZ
which
satisfies (1.2), set Y
t
= X
t
E {X
t
}. The first partial sum of interest
is

S
n
(v) =

n1
t=0
(X
st+v
(v)) =

n1
t=0
X
st+v
n(v).
Theorem 3.1. Consider the model (1.1) and its vectorial represen-
tation (1.2). Under the causality assumptions (2.4) and (2.5), the
following limit laws hold (n)
1/2

S
n
(v) N
_
0,
2
(v)
_
where

2
(v) := lim
n
(n)
1/2
E
_

S
2
n
(v)
_
< +, and if
2
(v) > 0
then limsup
n
(2n log log n)
1/2

S
n
(v) = (v) and liminf
n
(2n log log n)
1/2

S
n
(v) = (v) almost surely.
Proof. To prove Theorem 3.1, we use the same approach as in Liu
(1992). We first define the following R
r
-valued processes S
n
(t) =
_
A
t
S
n1
(t 1) +e
t
b
t
_
I
{n0}
and for all n N:
n
(t) = S
n
(t)
S
n1
(t) =
_

n1
i=0
A
ti
_
b
tn
e
tn
. It is easily seen that for all n 0,
S
n
(t) and
n
(t) are respectively
(e)
t
and
(e)
t1
measurable, and for
any m > Q
X
t
= e
t
b
t
+
mQ1

k=1
_
k1

i=0
A
ti
_
b
tk
e
tk
+

k=mQ
_
k1

i=0
A
ti
_
b
tk
e
tk
=
0
(t) +
mQ1

k=1

k
(t) +

k=mQ

k
(t).
Hence, we have that
E
_
Y
t
|
(e)
tm
_
= E
_
H

(X
t
E
_
X
t
_
)|
(e)
tm
_
= E
_
H

k=mQ
_

k
(t) E
_

k
(t)
__
_
|
(e)
tm
_
.
By using the inequalities E
_
(E {Y|} E{Y})
2
_
Var(Y) and
E
_

n
(t)
n
(t)
_
C.
n/2
(Q)
(cf. Liu, 1992) we deduce that for m Q

m=1
_
E
_
(E {Y
t
|
tm
})
2
__
1/2

m=1
_
Var
_
H


kmQ

k
(t)
__
1/2

m=1
_

nmQ
Var
_
H

n
(t)
_
+
_

nmQ
_
Var
_
H

n
(t)
__
1/2
_
2
_
_
_
1/2
2

m=1
_

nmQ
E
_
_
H

n
(t)
_
2
_
+
_

nmQ
_
E
_
_
H

n
(t)
_
2
__
1/2
_
2
_
_
_
1/2
C.

m1
_

m/2
(Q)
_
1/2
< +.
This completes the proof.
It is well known now that, even for stationary series, the first
and second empirical moments are not sufficient to distinguish
between linearity and nonlinearity. So, the resort to the higher
order moments is unavoidable. However, we need to consider the
limiting laws of the partial sums

S
(k(r))
n
(v) =

n1
t=0
X
t
(v)X
t
(v +
k
1
) X
t
(v + k
r
) n
k(r)
(v) where k(r) := (k
1
, . . . , k
r
) and

k(r)
(v) = E {X
t
(v)X
t
(v +k
1
) X
t
(v +k
r
)}. By using exactly the
same argument as in Theorem 3.1, we have the following general
result.
Theorem 3.2. Consider the model (1.1) and its vectorial represen-
tation (1.2). Assume that E
_

e
rQ
0

_
< + and E
_

e
j
0

_
= 0
for every odd positive integer j rQ. Under the causal-
ity assumptions (2.4) and (2.5), the following limit laws hold
(n)
1/2

S
(k(r))
n
(v) N
_
0,
2
k(r)
(v)
_
as n +where
2
k(r)
(v)
:= lim
n
(n)
1/2
_

S
(k(r))
n
(v)
_
2
< +, and if
2
k(r)
(v) > 0 then
limsup
n
(2n log log n)
1/2

S
(k(r))
n
(v) =
k(r)
(v), liminf
n
(2n log log n)
1/2

S
(k(r))
n
(v) =
k(r)
(v) almost surely.
Acknowledgements
This research was partially supported by Algerian Ministry of
Higher Education and Scientic Research contract PNR2011/8/u250/
738-CNEPRU.
References
Aknouche, A., Guerbyenne, H., 2009. Periodic stationarity of random coefficient
periodic autoregressions. Statistics & Probability Letters 79, 990996.
Basrak, B., Davis, R.A., Mikosch, T., 2002. Regular variation of GARCH processes.
Stochastic Processes and their Applications 99, 95115.
Bibi, A., Lessak, R., 2009. On stationarity and -mixing of periodic bilinear processes.
Statistics & Probability Letters 79, 7987.
Bibi, A., Moon-Ho, R., 2006. Estimation of periodic bilinear time series models.
Communications in Statistics-Theory and Methods 35, 17451756.
Bougerol, P., Picard, N., 1992. Strict stationarity of generalized autoregressive
processes. The Annals of Probability 20 (4), 17141730.
Brandt, A., 1986. The stochastic equation Y
n+1
= A
n
Y
n1
+ B
n
with stationary
coefficients. Advances in Applied Probability 18, 211220.
Franses, P.H., Paap, R., 2011. Random-coefficients periodic autoregressions.
Statistica Neerlandica 65 (1), 101115.
Gladyshev, E.G., 1961. Periodically correlated randomsequences. Soviet Mathemat-
ics 2, 385388.
A. Bibi, I. Lescheb / Economics Letters 114 (2012) 353357 357
Granger, C.W., Andersen, A.P., 1978. An Introduction to Bilinear Time Series Models.
Vandenhoek and Ruprecht, Gottingen.
Hall, P., Heyde, C.C., 1980. Martingale Limit Theory and its Applications. Academic
Press.
Ispany, M., 1997. On stationarity of additive bilinear state-space representation of
time series. In: Csiszar, I., et al. (Eds.), Stochastic Differential and Difference
Equations. Birkhauser, pp. 143155.
Kristensen, D., 2009. On stationarity and ergodicity of the bilinear model with
applications to the GARCH models. Journal of Time Series Analysis 30 (1),
125144.
Liu, J., 1992. On stationarity and asymptotic inference of bilinear time series models.
Statistica Sinica 2, 479494.
Pham, D.T., 1985. Bilinear Markovian representation of bilinear models. Stochastic
Processes and their Applications 20, 295306.

Você também pode gostar