A Note On The Stationarity and The Existence of Moments of The Garch Model

Statistica Sinica 8(1998), 505-510
A NOTE ON THE STATIONARITY AND THE EXISTENCE

OF MOMENTS OF THE GARCH MODEL
Min Chen and Hong Zhi An
Academia Sinica
Abstract: In the present paper we examine the strict stationarity and the existence
of higher-order moments for the GARCH(p,q) model under general and tractable
assumptions.
Key words and phrases: GARCH model, higher-order moments, nonlinear time
series, strict stationarity,
1. Introduction
Consider the following non-linear time series model
xt = t ht 2
ht = 0 + 1 x2t1 + + p x2tp + 1 ht1 + + q htq ,
(1.1)
where 0 > 0, i 0, i = 1, . . . , p, j 0, j = 1, . . . , q, {t } is a sequence of

independent identically distributed(i.i.d.) random variables with zero mean and
unit variance, and t is independent of xts , s > 0.
The model (1.1) is called the GARCH(p,q) model, which is proposed by
Bollerslev (1986) and is one of many generalizations of the so-called ARCH (autoregressive conditional heteroskedasticity) model proposed by Engle (1982) in
the literature. The GARCH models have been widely applied in modelling monetary and nancial data such as ination rate, interest rate and stock prices. The
recent review by Bollerslev et al. (1992) contains an extensive literature on this
subject.
The strict stationarity and the existence of moments for a time series model
are fundamental for statistical inference. Therefore, it is signicant to nd necessary and sucient conditions for the strict stationarity and the existence of
moments for a time series model. Bougeral and Picard (1992) gave a necessary
and sucient condition for the strict stationarity of the GARCH model. However, as they pointed out, the conditions proposed are dicult to verify and can
only be checked by Monte Carlo methods. Bollerslev (1986) discussed conditions
for the existence of higher-order moments for GARCH(1,1) model. So far it appears that there is not any paper about the existence of higher-order moments
506
MIN CHEN AND HONG ZHI AN
for the GARCH(p, q) model in the literature. The purpose of this paper is to give
some sucient conditions for the strict stationarity and the existence of moments
for the GARCH(p, q) model.
2. The Strict Stationarity of the GARCH Model
First, we introduce some notation. Let {xt } conform to model (1.1). Dene yt = xt 2 , t = t 2 , Y t = (yt , . . . , ytp+1 , ht , . . . , htq+1 ) , Bt = (0 t , 0, . . . ,
0, 0 , 0, . . . , 0) , where denotes the transposition of a matrix, and
1 t p1 t p t 1 t q1 t q t
1
0
0
0 0
0
. .
.
.
.
.
.
.
.
..
..
..
..
..
.. . .
.
0
At =

1
..
.
1
p1
0
..
..
.
.
0 0
0
q1
0
..
..
.
.
0 1
0
p
0
..
.
0
1
1
..
.
0
q
0
..
.
The expectation of the random matrix At is dened element-wise, hence it is

obvious that EAt is a constant matrix, and
1 p1 p 1 q1 q
1 0 0 0 0 0
. .
.. .. . .
.. ..
. . . ..
.
. . . . .
.
0 1 0 0 0 0
A.
EAt =

p1 p 1 q1 q
1
0 0 0 1 0 0
.
.. .. . .
. .
.. . .
. ..
.
. . . .. ..
0 0 1
Similarly, EBt = B (0 , 0, . . . , 0, 0 , 0, . . . , 0) . Then from (1.1) we have

2
xt = t
0 +
i xti +
i=1
which implies
yt =
p

i=1
j htj
j=1
i t 2 xti 2 +
i=1
j t 2 htj + 0 t 2 ,
(2.1)
j t htj + 0 t .
(2.2)
j=1
i t yti +
q

j=1
STATIONARITY AND MOMENTS OF GARCH MODELS
507
Thus, {yt } is a solution of (2.2) if and only if {Yt } is a solution of the following
stochastic dierence equation
Yt = At Yt1 + Bt .
Lemma 2.1. If
i=1 i
j=1 j
(2.3)
< 1, then the series of random vectors
k1

Atj Btk
j=0
k=1
converges almost surely. Furthermore if

Yt = Bt +
k1

k=1
Atj Btk ,
(2.4)
j=0
then {Yt } is a strictly stationary, vector-valued process satisfying (2.3).

Proof. By the denition of At and Bt , it is easy to see that both {At } and
{Bt } are sequence of independent, non-negative random vectors and Atj is
independent of Btk for k = j. Therefore, we have
E
k1

Atj Btk =
k1

j=0
EAtj EBtk = Ak B.
j=0
It is easy to veried that the characteristic polynomial of A is given by

i
det(A I) = 1 m
i=1 (i + i ) , where m = max{p, q}, and i = 0, for i > p,
i = 0, for i > q. Let (A) be the spectral radius of the matrix A, hence (A) < 1

if and only if pi=1 i + qj=1 j < 1. Thus, if pi=1 i + qj=1 j < 1, then (see
Horn and Johnson (1985))
Ak < ,
k=1
which implies that
k1

j=0
k=1
and hence
k1

k=1
Atj Btk < ,
Atj Btk < ,
a.s.
j=0
It is obvious that the vector-valued stochastic process {Yt } dened by (2.4)

is strictly stationary. Furthermore, we have

Yt = Bt + At Bt1 +
k1

k=2 j=0
Atj )Btk
508
= Bt + At Bt1 +
l1

At1l )Bt1l
l=1 j=0
= Bt + At Yt1 .
Lemma 2.2. If (2.3) admits a strictly stationary solution with finite first mo

ment, then pi=1 i + qj=1 j < 1. Moreover, the strictly stationary solution of
(2.3) is unique.
Proof. By (2.3), we have
Y0 = A0 Y1 + B0
= B0 + A0 B1 + A0 A1 Y2
..
.
= B0 +
n1
k1

k=1 j=0
(2.5)
n1

Aj )Bk + (
Aj )Yn .
j=0
Noting that all An , Bn and Yn are non-negative, {At } is a sequence of

independent random matrices, Anj and Bnk are independent for k = j, and
EY0 < . By taking expectation of each side of (2.5), it follows that
EY0
n1

k=1
This shows that
k1

j=0
Aj Bk =
n1

Ak B.
k=1
Ak B < .
k=1
Therefore,
lim An B = 0.
(2.6)
Let {i , i = 1, . . . , p + q} be the canonical basis of Rp+q , i.e. i = (i,1 , . . . ,

i,p+q ) , where ij = 0, for i = j, ii = 1. If we can prove that for 1 i p + q,
lim An i = 0,
(2.7)
then (2.7) implies that limn An = 0, which again implies that (A) < 1. As

we showed before the later is equivalent to pi=1 i + qj=1 j < 1, which leads
to the rst part of this lemma.
In fact, since B = 0 (1 + p+1 ) and 0 < 0 < , by (2.6) and the denition
of matrix A, (2.7) holds for i = 1 and i = p + 1. Again by the denition of A
and B, Ap+q = q (1 + p+1 ). If q = 0, then Ap+q = 0, hence (2.7) holds. If
q > 0, by the above equalities,
lim An p+q = lim An1 q (1 + p+1 ) = 0.
STATIONARITY AND MOMENTS OF GARCH MODELS
509
It is easy to see that for 2 i < p,

Ai = i 1 + i+1 + i p+1 .
Since (2.7) holds for i = 1 and i = p + 1, by a backward recursion (2.7) holds for
i = p, p 1, . . . , 1, respectively. Similarly, for p + 1 < i < p + q,
Ai = i (1 + p+1 ) + i+1 .
Noting that (2.7) holds for i = p + q, by a backward recursion, (2.7) holds for
i = p + q 1, p + q 2, . . . , p + 2. Finally, (2.7) holds for any i = 1, 2, . . . , p + q.
For the proof of the uniqueness, let {Ut } be another strictly stationary solution satisfying (2.3). Then {Ut } also satises an equation similar to (2.5). Note
that the third term on the right hand side of the last line in (2.5) goes to zero in
probability. Then the uniqueness follows immediately.
Theorem 2.1. The GARCH(p, q) model (1.1) admits a strictly stationary solu

tion with finite variance if and only if pi=1 i + qj=1 j < 1. Moreover, this
strictly stationary solution is also unique.
Proof. The desired resultis obtained by combining Lemma 2.1 and Lemma
2.2.
3. The Existence of Higher-Order Moments
Let
Ls = {x : xs = E s |x|s < , s > 0},
where x is a random variable.We need the Kronecker product (), the direct
operations (vec operations ), the notation Am = A A A, and the
basic identity vec(ABC) = (C A)vec(B). We denote the (i, j)th element of a
matrix D by (D)ij and dene m = E(At m ).
Theorem 3.1. Let {xt } be specified to be the strictly stationary solution of model

(1.1) and pi=1 i + qj=1 j < 1.
(i) If E|t |4 < and (2 ) < 1, then |xt |2 L2 .
(ii) If E|t |4(s1) < , for some integer s > 2, and (s ) < 1, then |xt |2 Ls .
Proof. Let Yt be generated according to (2.3) with the starting value Y0 = 0.
Let Y be a random variable having the same distribution as that of (2.4) with
t = 0, which is the (marginal) distribution of the unique stationary solution of
(2.3). It is clear that Yt Y in distribution. Let (Y ) be a random variable.
From weak convergence theory (Billingsley (1968)), it is known that to show
E(Y ) < , it suces to show lim inf E(Yt ) < . Let V (t) = E(Yt ). Then,
taking expectation on both sides of (2.3), we get
V (t) = AV (t 1) + B.
510
It is well know that lim V (t) exists and is nite if the spectral radius of A is less
than 1(cf. Subba Rao (1981)). Hence if the spectral radius of A is less than 1,
{V (t)} is bounded. Next, let V1 (t) = E(vec(Yt Yt )). Then
V1 (t) = E(At At )V1 (t 1) + E(Bt At )V1 (t 1)
+E(At Bt )V1 (t 1) + vec(E(Bt Bt )).
Note that the matices E(At At ), E(Bt At ), E(At Bt ) and vec(E(Bt Bt ))
are constant and nite matrices. As {V (t)} is bounded, it is clear that lim V1 (t)
exists and is nite if the spectral radius of 2 = E(At At ) is less than 1. Note
x4t ). Because x
4t converges in distribution to
that the rst element of V1 (t) is E(
4
4
x0 and lim V1 (t) is nite, E(x0 ) < . This completes the proof of (i). The proof
of (ii) is similar and hence omitted.
Acknowledgement
We are very grateful to referees for their careful reading and valuable suggestions which have enabled us to improve the earlier version of this paper. The
work was supported by NNCF of China and Prob. Lab. Inst. Applied Math.
Academia Sinica.
References
Billingsley, P. (1968). Convergence of Probability Measure. Wiley, New York.
Bollerslev, T. (1986). Generalized autoregressive conditional heteroscedasticity. J. Econometrics 31, 307-327.
Bollerslev, T., Chou, R. Y. and Kroner, K. F. (1992). ARCH modeling in finance. J. Econometrics 52, 5-59.
Bougeral, P. and Picard, N. (1992). Stationarity of GARCH processes and of some nonnegative
time series. J. Econometrics 52, 115-127.
Engle, R. F. (1982). Autoregressive conditional heteroscedasticity with estimates of the variance
of U. K. inflation. Econometrica 50, 987-1007.
Guegan, D. and Diebolt, J. (1994). Probabilistic properties of the -ARCH model. Statist.
Sinica 4, 71-87.
Horn, R. A. and Johnson, C. R. (1985). Matrix Analysis. Cambridge Univ. Press.
Subba Rao, T. (1981). On the theory of bilinear time series models. J. Roy. Statist. Soc. Ser.
B 43, 244-255.
Institute of Applied Mathematics, Academia Sinica, Beijing, 100080.
E-mail: mchen@iss01.iss.ac.cn
E-mail: hzan@sun.ihep.ac.cn
(Received October 1995; accetped March 1997)

A Note On The Stationarity and The Existence of Moments of The Garch Model

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

A Note On The Stationarity and The Existence of Moments of The Garch Model

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Note On The Stationarity and The Existence of Moments of The Garch Model

Uploaded by

Copyright:

Available Formats

Statistica Sinica 8(1998), 505-510

A NOTE ON THE STATIONARITY AND THE EXISTENCE

where 0 > 0, i 0, i = 1, . . . , p, j 0, j = 1, . . . , q, {t } is a sequence of

MIN CHEN AND HONG ZHI AN

The expectation of the random matrix At is dened element-wise, hence it is

Similarly, EBt = B (0 , 0, . . . , 0, 0 , 0, . . . , 0) . Then from (1.1) we have

STATIONARITY AND MOMENTS OF GARCH MODELS

< 1, then the series of random vectors

converges almost surely. Furthermore if

then {Yt } is a strictly stationary, vector-valued process satisfying (2.3).

It is easy to veried that the characteristic polynomial of A is given by

which implies that

Atj Btk < ,

Atj Btk < ,

It is obvious that the vector-valued stochastic process {Yt } dened by (2.4)

MIN CHEN AND HONG ZHI AN

Noting that all An , Bn and Yn are non-negative, {At } is a sequence of

This shows that

Let {i , i = 1, . . . , p + q} be the canonical basis of Rp+q , i.e. i = (i,1 , . . . ,

STATIONARITY AND MOMENTS OF GARCH MODELS

It is easy to see that for 2 i < p,

Ls = {x : xs = E s |x|s < , s > 0},

MIN CHEN AND HONG ZHI AN

You might also like

Ls = {x : xs = E s |x|s < , s > 0},