SJK wjk
และมีการปรับหรือเรียบและรายละเอียดหรือเวฟสัมประสิทธิ์ตามลำดับและจะได้รับโดย:
X (t) 0M (t) dt (3)
X (t) # เจ k (t) dt ญ = 1,2 ... J (4)
ทีวี. *
จับสัมประสิทธิ์เนื้อหาความถี่สูงของอนุกรมเวลาในขณะที่ค่าสัมประสิทธิ์SJK
เรียบเป็นตัวแทนของพฤติกรรมของซีรีส์ ขนาดของสัมประสิทธิ์เหล่านี้มีตัวชี้วัดของผลงานของที่สอดคล้องกันที่. ฟังก์ชั่นเวฟชุดรวมดังนั้นอนุกรมเวลาเดิมสามารถสร้างขึ้นใหม่เช่นXt = Sj (t) + Wj (t) + Wj_! (t) + ••• + W (t) (5) ที่ Sj = '£ KSU
1, ... , J เป็นเรียบ (หรือประมาณ) และรายละเอียด 1 ,...,J are the smooth (or approximation) and detail (or
(หรือเวฟ) ส่วนประกอบของสัญญาณตามลำดับ wavelet) components of the signal, respectively. This reconstruction
is known as multiresolution analysis.
In practice, the DWT is implemented via the pyramid
algorithm derived by Mallat (1989). This technique consists
of recursively applying a high-pass filter, which is based on
the mother wavelet, and its counterpart low-pass filter to
a given time series. The high-pass filter corresponds to a
differencing operation and extracts the detail (high frequency)
information of the signal, while the low-pass filter
is associated to an averaging operation and extracts the
coarse (low frequency) information. Specifically, the original
Xt series is decomposed into approximation Si(t) and detail
Wi (t) components by convolving the series with the lowpass
and high-pass filters, respectively. The approximation
component Sift) becomes the input for the next iteration
step, so that two new approximation S2(t) and detail W2(t)
components are obtained. This recursive procedure is continued
until the decomposition level J is reached. Despite its
great popularity, the standard DWT has several drawbacks.
First, it is subject to boundary effects, which may produce
unreliable results in the edge regions. Second, it is not shift
invariant. Third, it requires time series with a dyadic length.
The Haar a trous wavelet (HTW) transform
In this study, the HTW transform developed by Murtagh et al.
(2004) is applied as an alternative to traditional DWT. The
HTW combines the a trous wavelet, which is a type of redundant
wavelet transform, with the Haar wavelet, which is
one of the simplest wavelets. The HTW transform offers a
number of advantages over standard wavelet transforms.
First, the use of the Haar wavelet function respects the
asymmetric nature of the time-varying signal, so that scaling
and wavelet coefficients are calculated only from data
obtained previously in time. Therefore, the HTW does not
suffer from boundary problems caused by the application of
wavelet analysis on finite data sets. This absence of edge
effects in the HTW transform allows the conservation of the
whole information contained in the original series. Second,
the HTW transform is flexible enough to accurately reflect
the non-linear and chaotic dynamics of many financial time
series. Third, the redundancy inherent in the a trous wavelet
function implies that all wavelet components have the same
length as the original time series, so it is easy to relate information
at each resolution scale for the same time point. This
102 P. Moya-Marti'nez et al.
property of redundancy of also means that the HTW is shift
invariant.2
The HTW is a redundant version of the popular Haar
wavelet transform that uses a non-symmetric low-pass filter
h equal to (1, | ) and can be described as follows. The
scaling coefficients Sj+1/! of a time series Xt at any scale can
be obtained by convolving the smoothed version of the signal
at the previous scale with the low-pass filter h = ( | , j ) :
sj+1,k = ^(Slk-1 +Sj,k) (6)
The detail coefficients can be calculated as the difference
between the smoothed versions of the signal at two
consecutive scales:
Wj+1.k = sj.k ~ sj+1.k (7)
HTW-based variance, correlation and
cross-correlation
The wavelet coefficients can be manipulated to obtain
several statistical wavelet-based tools, such as the
wavelet variance, wavelet correlation, and wavelet crosscorrelation,
which provide an alternative representation of
the variability and association structure between time series
on a scale-by-scale basis. These three tools are defined
analogously to the usual variance, correlation and crosscorrelation
measures in time series analysis.
The wavelet variance decomposes the variance of a time
series on a scale-by-scale basis and helps to identify what
time scales are the dominant contributors to the overall
variability of the series. As noted by Gallegati (2008), the
wavelet variance at scale tj of a stationary stochastic process
Xt, o* (tj), is given by:
1
o-j(rj) = — var(Q)j,t ) (8)
where tj = 2-M and wjx denote the wavelet coefficients of
Xt at scale tj.
According to Jammazi (2012b), the HTW-based variance
estimator can be also expressed in terms of the normalized
sum of the squared wavelet coefficients:
1 Nr ' -1
ffx.HTw(rj) = 27Jf. = N E r f t
1 t=0 t=0
where ffxHTwIf/) 's the estimated wavelet variance at scale
tj, Nj = N/2j the number of wavelet coefficients at the resolution
level j and N the sample size. Since the HTW transform
is not affected by boundary conditions, the HTW variance
estimator allows to analyze a signal by using all the wavelet
coefficients.
The HTW framework also enables us to derive the
wavelet correlation and cross-correlation. These waveletbased
tools make it possible to quantify the degree of
association between two stochastic processes on a scale-byscale
basis. Analogously to the usual correlation coefficient
in time series analysis, the wavelet correlation coefficient
2 For a more complete and detailed discussion of the Haar a trous
wavelet transform, see Murtagh et al. (2004) and Jammazi (2012b).
at scale tj, px.Y(tj), can
การแปล กรุณารอสักครู่..