J Chongqing Univ.-Eng. Ed.
Mathematics & Application
Vol. 5 No. 3
September 2006
Article ID: 1671-8224(2006)03- 0165-05
Vector FIGARCH process, its persistence
and co-persistence in variance
∗
LI Song-chen
1,a
, ZHANG Shi-ying
2
1
School of Science, Tianjin University, Tianjin 300072, P.R. China
2
Institute of Systems Engineering, Tianjin University, Tianjin 300072, P.R. China
Received 22 February 2006; Revised 29 April 2006
Abstract: In this paper, the definition of the vector FIGARCH process is established, and the stationarity and some properties
of the process are discussed. According to the stationarity and the results of Du and Zhang [1], we verify the persistence in
variance of the vector FIGARCH process, and finally establish the sufficient and necessary condition for the co-persistence in
the variance of the process and also discuss the constant related vector FIGARCH(,,)pdq process as a special case.
Keywords: vector FIGARCH process; stationarity; persistence; co-persistence.
CLC number: O213 Document code: A
1 Introduction
∗
Since the ARCH model and the GARCH model
were introduced by Engle [2] and Bollerslev [3],
respectively, many different models in the GARCH
family have been extensively used in modeling
financial time series. For example, LGARCH,
MGARCH [4], EGARCH [5], GJR-GARCH [6],
NGARCH [7], VGARCH [7], TGARCH [8],
Augmented GARCH [9], FIGARCH [10], FI-
Augmented-GARCH-M [11] , etc.
However, financial market is a system involving
many stochastic variables. To describe the volatility
of these variables and the relation among them, the
vector ARCH [12] was proposed, then the concept of
persistence and co-persistence was introduced by
Bollerslev and Engel [13]. Furthermore, the co-
persistence of vector GARCH processes was
considered by Du and Zhang [1]. This paper will
mainly concern the persistence and the co-persistence
of the vector FIGARCH process. Hereinafter, we will
first give the definition of the vector FIGARCH
a
LI Song-chen (李松臣): Male; Born 1978; PhD candidate;
Lecture; Research interests: time series analysis, multivariate
statistical analysis; Email: lisongchen@tju.edu.cn
∗
Funded by the Natural Science Foundation of China (No.
70471050).
process, then discuss the stationarity and some useful
results, and verify the persistence and the co-
persistence of the vector FIGARCH process.
2 Definition of the vector FIGARCH process
Consider the real-valued stochastic ARCH process
{}
t
,
ttt
z
σ
= , where
1
() 0
tt
Ez
−
= ,
1
VAR ( ) 1
tt
z
−
,
and
t
is a positive time-dependent and measurable
function with respect to the information set available
at time ( 1)t
. Throughout,
1
()
t
E
−
g and
1
VAR ( )
t−
g
respectively denote the conditional expectation and
variance with respect to this same information set.
Thus, by definition, the { }
t
process is serially
uncorrelated with mean zero, but the conditional
variance of the process
2
t
is changing over time.
In the classic ARCH (
q
) model of Engle [2], the
conditional variance
2
t
is postulated to be a linear
function of the lagged squared innovations implying
Markovian dependence dating back only to
q
periods;
i.e.,
2
ti
for 1,2, ,iq
K . The GARCH ( , )pq
specification of Bollerslev [3] provides a more
flexible lag structure. Formally, this model is defined
by
222
() ()
ttt
LL
ωα ε β σ
=+ + , (1)
where L denotes the lag or backshift operator; ( )L