You are on page 1of 11

Physica A 429 (2015) 1727

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

Multifractal cross-correlation analysis in electricity spot


market
Qingju Fan , Dan Li
Department of Statistics, School of Science, Wuhan University of Technology, Luoshi Road 122, Wuhan, PR China

highlights
Electricity price and trading volume series are analyzed by MF-DFA.
The newly developed MFCCA methodology is applied.
The cross-correlations disappear on the level of relatively small fluctuations.
The MFCCA methodology provides more convincing results than MF-DXA.

article info abstract


Article history: In this paper, we investigate the multiscale cross-correlations between electricity price
Received 24 November 2014 and trading volume in Czech market based on a newly developed algorithm, called Mul-
Received in revised form 20 January 2015 tifractal Cross-Correlation Analysis (MFCCA). The new algorithm is a natural multifrac-
Available online 20 February 2015
tal generalization of the Detrended Cross-Correlation Analysis (DCCA), and is sensitive to
cross-correlation structure and free from limitations of other algorithms. By considering
Keywords:
the original sign of the cross-covariance, it allows us to properly quantify and detect the
Multifractal cross-correlation analysis
(MFCCA)
subtle characteristics of two simultaneous recorded time series. First, the multifractality
Generalized Hurst exponent and the long range anti-persistent auto-correlations of price return and trading volume
Scaling exponent variation are confirmed using Multifractal Detrended Fluctuation Analysis (MF-DFA). Fur-
thermore, we show that there exist long-range anti-persistent cross-correlations between
price return and trading volume variation by MFCCA. And we also identify that the cross-
correlations disappear on the level of relative small fluctuations. In order to obtain deeper
insight into the dynamics of the electricity market, we analyze the relation between gener-
alized Hurst exponent and the multifractal cross-correlation scaling exponent q . We find
that the difference between the generalized Hurst exponent and the multifractal cross-
correlation scaling exponent is significantly different for smaller fluctuation, which indi-
cates that the multifractal character of cross-correlations resembles more each other for
electricity price and trading volume on the level of large fluctuations and weakens for the
smaller ones.
2015 Elsevier B.V. All rights reserved.

1. Introduction

The power sector has gradually been transformed from an exclusive monopolistic state to a competitive market place
after deregulation, which has exposed market players to more uncertainty. Due to the unique nonstorability (outside
of hydro) of electricity, electricity price is much more likely to be driven by spot demand and supply considerations

Corresponding author. Tel.: +86 13995685880.


E-mail address: fanqj@bu.edu (Q. Fan).

http://dx.doi.org/10.1016/j.physa.2015.02.065
0378-4371/ 2015 Elsevier B.V. All rights reserved.
18 Q. Fan, D. Li / Physica A 429 (2015) 1727

than any other commodities, with demand in the short-term market being fairly inelastic. As a result, sizable shocks in
production or consumption may give rise to the price jumps [1]. Because of the potential instability of the electricity system,
many studies concentrate on the correlations and memory characteristics of electricity price in recent years, and most
literatures show that the electricity price is strongly non-stationary and the return series are long-range anti-persistent
auto-correlations by using different methods, such as rescaled range analysis, detrended fluctuation analysis, periodogram
methods, multiscale wavelet approach, Markov switching multifractal model and fractionally integrated autoregressive
moving average model [27].
Nonlinear auto-correlations or cross-correlations of time series are often grounded on the study of their multifrac-
tal structure [8]. There exist many different methods which are applied to characterize the auto-correlations and cross-
correlations of data in various area, such as physics [9], geophysics [10], economics [1114], hydrology [15] and so on. The
most popular method is Detrended Fluctuation Analysis (DFA) developed by Peng et al. [16], which can detect the scaling
properties of long-range power-law auto-correlation of nonstationary time series. In addition, another robust and pow-
erful technique is Multifractal Detrended Fluctuation Analysis (MF-DFA), which is proposed by Kantelhardt et al. [17]. To
quantify the cross-correlations between two non-stationary time series, a new method based on detrended covariance,
called Detrended Cross-Correlation Analysis (DCCA) is proposed by Podobnik and Stanley [18], which is the generalization
of the fractal auto-correlation (DFA). Subsequently, the multifractal extension (MF-DXA) of the DCCA method was pro-
posed by Zhou [19]. Other closely related methods to deal with multifractal cross-correlations have also been introduced in
Refs. [2026]. Recently, Oswiecimka et al. proposed a novel algorithm named Multifractal Cross-Correlation Analysis
(MFCCA) [27], which is a consistent extension of detrended cross-correlation analysis (DCCA). Different from the meth-
ods proposed earlier, MFCCA allows to compute the arbitrary-order covariance function of two signals, and at the same time
it properly takes care of the relative signs in the signals, which can avoid the limitation of taking modulus [21,28,29] of the
cross-covariance function, and properly quantify and detect the subtle characteristics of two simultaneous recorded signals.
By means of this new method, we can calculate the spectrum of the cross-correlation scaling exponent q and estimate the
scaling properties of the qth order cross-covariance function with respect to the original sign of the cross-covariance.
Therefore, we apply the MFCCA to investigate the cross-correlation properties between the electricity price and trading
volume in the Czech electricity spot market. To the best of our knowledge, there are few literatures focusing on the study of
the cross-correlations between the electricity price and trading volume. In fact, as a most important index, the trading vol-
ume contains much of useful information about the dynamics of price formation, which can help us to understand the behav-
ior of electricity markets. In this paper, first the multifractality of electricity price return series and trading volume variation
series are confirmed using MF-DFA. Furthermore, a multifractal cross-correlation analysis between electricity price return
and trading volume variation series in Czech electricity spot market is also conducted. We qualitatively analyze the cross-
correlations between the series of the electricity price return and trading volume variation employing the statistics proposed
by Podobnik et al. [30]. Then, we utilize the MFCCA method to investigate the cross-correlations between the two series.
The organization of this paper is as follows. Section 2 presents the methodology used in this paper and Section 3 is the
data description. Section 4 provides the empirical result. We conclude in Section 5.

2. Methodology

2.1. MF-DXA method

Multifractal Detrended Cross-Correlation Analysis (MF-DXA) method was used to quantify the cross-correlations of two
different non-stationary time series, which is introduced by Zhou [19] and has been applied in many empirical systems
[28,29,31]. The MF-DXA procedure consists of five steps, and the first step is the same as to the traditional DFA procedure.
Consider two time series xk and yk , where k = 1, 2, . . . , N and N is the length of the time series.
Step 1: Determine the profile
i
i

Y (i) = (x x) , Y (i) = (y y) (1)
k=1 k =1

where denotes averaging over entire time series.


Step 2: Cut the profile Y (i) and Y (i) into Ns = [N /s] nonoverlapping segments of equal length s. Since the record length N
need not be a multiple of the considered time scale s, a short part at the end of the profile will remain in most cases.
In order not to disregard this part of the record, the same procedure is repeated starting from the other end of the
record. Thus, 2Ns segments are obtained altogether. In practice, it is reasonable to take 10 < s < N /5.
Step 3: Calculate the local trend for each segment v by a least-square fit of the data. Then we calculate the difference
between the original time series and the fits.
Step 4: Calculate the covariance of the residuals in each segment v :
s
1
2
fxy (s, v) = (Yk Yk,v ),
Yk,v ) (Yk (2)
s k=1
Q. Fan, D. Li / Physica A 429 (2015) 1727 19

v = 1, 2, . . . , 2Ns . Here, Yk,v are the fitting polynomials in segment v , respectively. Then average over all
Yk,v and
segments to obtain the fluctuation function:
2Ns
1 1/q
q
Fxy ( s) = [fxy2 (s, v)]q/2 (3)
2Ns v=1

for any real value q = 0, and



2Ns
1
0
Fxy (s) = exp 2
ln[fxy (s, v)] (4)
4Ns v=1
q
Step 5: Determine the scaling behavior of the fluctuation functions by analyzing loglog plots Fxy (s) versus s:
q
Fxy (s) sq (5)

The scaling exponent q is known as the generalized cross-correlation exponent, describing the power-law relationship
between two time series. In particular, if the time series xk is identical to yk , MF-DXA is equivalent to MF-DFA [17]. Moreover,
if scaling exponent q is independent of q, the cross-correlations between two time series are monofractal. If scaling
exponent q is dependent on q, the cross-correlations between two time series are multifractal. Furthermore, for positive q,
q describes the scaling behavior of the segments with large fluctuations. On the contrary, for negative q, q describes the
scaling behavior of the segments with small fluctuations.
2
The dubious sign of fxy (s, v) in Eq. (2) plays a determinant role. More complex cases have remained unexplained, such
2
as fxy (s, v) is negative. Most literatures [21,28,29] set a tiny modification on the above procedure to the problem. The
modification is as follows,
s
1
2
fxy (s, v) = Yk,v | |Yk
|Yk Yk,v |. (6)
s k=1
In most situations, we take Eq. (6) rather than Eq. (2), if sign information is neglected and only fluctuation information is
concerned.
Furthermore, in order to measure the degree of multifractality of cross-correlation, we define q :
q = max(q ) min(q ). (7)
The greater is q , the stronger is the degree of multifractality of cross-correlation [32].

2.2. MFCCA method

Multifractal Cross-Correlation Analysis (MFCCA) is based on the DCCA, and it also consists of five steps. The difference
between MFCCA and MF-DXA lies in the fourth step. Oswiecimka et al. [27] propose to consider the sign of cross-covariation
2
fxy (s, v), the most natural form of the qth order covariance function is postulated by the following equation:
2Ns
1 1/q
q
Fxy (s) = sign(fxy
2
(s, v))|fxy2 (s, v)|q/2 (8)
2Ns v=1
for any real value q = 0, and

2Ns
1
0
Fxy (s) = exp ln[ 2
fxy (s, v)] . (9)
4Ns v=1
2
As we can see in Eq. (8), for negative values of q, small values of the covariance function fxy (s, v) are amplified; while for large
q
q > 0, its large values dominate. Moreover, the formula for calculating Fxy (s) respects the genuine signs of the amplified
(or suppressed) fluctuations of the detrended cross-covariance function (Eq. (2)), and at the same time, it allows to avoid
q
complex numbers associated with the arbitrary powers of negative fluctuations. If the so-obtained function Fxy (s) does not
develop scaling, by for instance fluctuating around zero, there is no fractal cross-correlation between the time series under
study for the considered value of q. Multifractal cross-correlation is expected to manifest itself in the power-law dependence
(if the qth order covariance function is negative for every s, we may take Fxy (s) Fxy (s)): Fxy (s) sq , which is the same
q q q

as Eq. (5).

3. Data description

In our study, we focus on the cross-correlations between electricity price and trading volume in the Czech Republic. The
Czech electricity market was fully deregulated in 2006 and the network of power plants consists of less expensive hydro and
20 Q. Fan, D. Li / Physica A 429 (2015) 1727

Fig. 1. Price return and trading volume variation.

nuclear power plants, and more expensive hard coal and gas power plants, with lignite plants of moderate cost [33]. OTE (the
Czech electricity and gas market operator, established in 2001) has been organizing the day-ahead spot electricity market
since 2002. The Czech market has been coupled through implicit auctions with the organized day-ahead electricity market
in Slovakia since 2009 and with the day-ahead electricity market in Hungary since 2012. It has the form of a daily auction,
with a traded period of 1 h and a minimum tradable volume of 1 MWh, and with Euro as its trading currency. The market al-
ways closes at 11 a.m. the day before. We analyze hourly day-ahead spot price of electricity and trading volume in the Czech
Republic between January 1, 2009 and November 30, 2012, with a total of 34,320 observations. The price is denominated in
EUR/MWh and negative price was not allowed before 2013. Let Pt and Vt denote the price and trading volume of electricity
at time t, respectively. The hourly price return, rt , is calculated as its logarithmic difference, rt = log(Pt ) log(Pt 1 ), and
also the logarithmic variation in trading volume Vt , r t = log(Vt ) log(Vt 1 ). The data were obtained from OTEs Yearly Re-
ports, available at http://www.ote-cr.cz/statistics. The graphical representation of series of price return and trading volume
variation is illustrated in Fig. 1.

4. Empirical analysis

4.1. Multifractal detrended fluctuation analysis (MF-DFA)

In order to detect the scaling properties of electricity price and trading volume, we first conduct multifractal analysis on
the electricity price return and trading volume variation series, respectively.
Based on MF-DFA methodology, we calculated the fluctuations F q (s) for scales s ranging from 10 events to N/5, where
N is the total length of the series. Figs. 2 and 3 show the fluctuation functions F q (s) versus s of electricity price return and
trading volume variation series from q = 3 to q = 3 (polynomial order k = 2), where the upper and the lower curves are
the curves of q = 3 and q = 3, respectively.
Fig. 2 shows the loglog plots of F q (s) versus s for the electricity price return series. When s > 15, the scaling starts visible.
As can be seen from Fig. 2, there is a crossover time scales s (about s = 42, approximately two days). Table 1 shows the
scaling exponents of electricity price return series in different qth. For 15 < s < 42, the scaling exponents decrease from
0.9990 to 0.3641 when q increases, which implies multifractality in the short-term. The scaling exponents are all smaller
than 0.5 for s > 42, which shows that the price return series exist anti-persistent auto-correlations in the long-term, and
the scaling exponents q weakly depend on the values of q indicating that the long-term auto-correlated behavior is weakly
multifractal.
In Fig. 3, it seems that there are two crossovers (about s1 = 24, and s2 = 260) in the loglog plots of F q (s) versus s
for the trading volume variation series. To describe the scaling behavior under different time scales, we compute the scaling
exponents in different scale intervals based on MF-DFA, less than 1 day (s < 24), more than 1 day but less than 11 days
(24 < s < 260), and larger than 11 days (s > 260). In Table 2, we can find that, for s < 24, the scaling exponents
decrease from larger than 0.7 to smaller than 0.4 indicating that the auto-correlated behavior is multifractal in the short-
term. Moreover the scaling exponents for q > 1 are smaller than 0.5 indicating that the auto-correlated behavior of large
fluctuations is anti-persistent in the short-term. While for 24 < s < 260 and s > 260, the scaling exponents are smaller
than 0.5, indicating that the trading volume variation is anti-persistent in the long-term.
Q. Fan, D. Li / Physica A 429 (2015) 1727 21

Fig. 2. Loglog plots of F q (s) versus s for price return.

Fig. 3. Loglog plots of F q (s) versus s for trading volume variation.

Table 1
Scaling exponents of electricity price return
series obtained from MF-DFA.
15 < s < 42 s > 42

q = 3 0.9990 0.1798
q = 2 0.9202 0.1777
q = 1 0.8539 0.1704
q =0 0.7830 0.1485
q =1 0.6401 0.1019
q =2 0.4593 0.0408
q =3 0.3641 0.0115

Fig. 4 also shows the q-dependence of the generalized Hurst exponent for the electricity price return and trading volume
variation series in different time scale intervals, respectively. As shown in Fig. 4, when q varies from 3 to 3, both electricity
price return and trading volume variation series show stronger multifractality in the short-term than in the long-term.
22 Q. Fan, D. Li / Physica A 429 (2015) 1727

Fig. 4. The generalized Hurst exponents q of price return and trading volume variation in different scale intervals.

4.2. Multifractal cross-correlation analysis

4.2.1. Statistics test


Podobnik et al. [30] introduced a cross-correlation statistic in analogy to the LjungBox (LJB) test [34]. For two series {xk }
and {yk }, where k = 1, 2, . . . , N, the test statistic

m
Xi
Qcc (m) = N 2 . (10)
i=1
N i

Here, the cross-correlation function

N

xk yki
k=i+1
Xi = . (11)
N N
x2k y2k

k=1 k=1
Q. Fan, D. Li / Physica A 429 (2015) 1727 23

Fig. 5. The cross-correlation statistics.

Table 2
Scaling exponents of trading volume variation series obtained
from MF-DFA.
s < 24 24 < s < 260 s > 260

q = 3 0.7403 0.2409 0.0749


q = 2 0.6314 0.2230 0.0698
q = 1 0.5414 0.2048 0.0649
q =0 0.4683 0.1870 0.0600
q =1 0.4076 0.1700 0.0553
q =2 0.3559 0.1545 0.0508
q =3 0.3105 0.1407 0.0465

The cross-correlation statistic Qcc (m) is approximately 2 (m) distributed with m degrees of freedom. The statistic can be
used to test the null hypothesis that none of the first m cross-correlation coefficient is different from zero. If there are no
cross-correlations between two time series, the cross-correlation test agrees well with the 2 (m) distribution. If the cross-
correlations test exceeds the critical value of the 2 (m) distribution, then the cross-correlations are significant at a special
significance level.
Fig. 5 shows the cross-correlation statistics for two pairs of series in the Czech electricity market. The degrees of freedom
vary from 100 to 104 . As a comparison, we also show the critical values for the 2 (m) distribution at the 5% level of
significance in Fig. 5. The results show that for the series of price return and trading volume variation, the empirical results
of Qcc (m) are larger than the critical values of the 2 (m) distribution at a 5% level of significance. Therefore, the null
hypothesis of no cross-correlations can be rejected in its entirety. This finding indicates that long-range cross-correlations
exists between electricity prices and trading volume in the Czech electricity market.

4.2.2. Multifractal cross-correlation analysis (MFCCA)


As suggested in Podobnik et al. [30], the cross-correlation test based on the statistics in Eq. (10) can only be used to test
the presence of cross-correlations qualitatively. In order to test the presence of cross-correlations quantitatively, we apply
the newly developed MFCCA methodology to estimate the multifractal cross-correlation exponents.
q
In order to characterize the cross-correlations, we show the fluctuation function Fxy (s) versus time scale s between
electricity price return and trading volume variation in Fig. 6 (polynomial order k = 2). As far as multifractal scaling is
q
concerned, the situation is significantly subtle. First, we found that the fluctuation function Fxy (s) fluctuates around zero
when q varies from 2 to 0, and Eq. (5) is not satisfied, which is depicted in Fig. 7. Fig. 6 shows the fluctuation function
q
Fxy (s) versus time scale s for q (0.6, 4). There are two crossovers which are s1 = 30(about 1 day) and s2 = 270 (about
11 days) in the loglog plot, and we also show cross-correlation scaling exponents of different scale intervals in Table 3.
We can observe that, when q = 2, the cross-correlation scaling exponent is 0.7227 for s < 30 indicating that the elec-
tricity price return and trading volume variation series are persistent cross-correlated in the short-term. For 30 < s < 270
and s > 270, the scaling exponents are 0.1794 and 0.0504, respectively, smaller than 0.5, indicating that they are anti-
persistent cross-correlated in the long-term. And when q vary from 0.6 to 4, the scaling exponents decrease from larger
1.4737 to smaller 0.2635 for s < 30 indicating that the cross-correlated behavior is strong multifractal in the short term.
24 Q. Fan, D. Li / Physica A 429 (2015) 1727

q
Fig. 6. Fxy (s) versus time scale s between electricity price return and trading volume variation obtained from MFCCA.

q
Fig. 7. When q (2, 0), Fxy (s) versus s obtained from MFCCA.

Moreover, the scaling exponents for q (0.6, 2.8) are larger than 0.5 indicating that the cross-correlated behavior of small
fluctuations is persistent in the short-term. The scaling exponents for q (2.6, 4) are smaller than 0.5 indicating that the
cross-correlated behavior of large fluctuations is anti-persistent in the short-term. However, the scaling exponents also de-
crease from larger 0.4594 to smaller 0.0678 for 30 < s < 270, indicating that the cross-correlated behavior is moderately
multifractal in the medium-term. And for s > 270, the values of q weakly depend on the values of q indicating that the
long-term cross-correlated behavior is weakly multifractal. Moreover, for both 30 < s < 270 and s > 270, all of the values
of q are smaller than 0.5 indicating that the cross-correlated behavior is anti-persistent in the long-term.
Additionally, we also conduct MF-DXA procedure (polynomial order k = 2) on the electricity prices return and trading
volume variation series. The result is shown in Fig. 8, and it clearly indicates a convincing multifractal scaling when q varies
from 5 to 5. This, of course, is a spurious cross-correlations.
Many literatures pointed out that the estimated fractal cross-correlations are often related to the fractal properties of
the individual signals themselves [17,35,36]. It is worth noticing that the relation between q and hxy (q) (which is the
average of the generalized Hurst exponents of individual time series [17]) depends on temporal organization of the signals as
determined by their Hurst exponents. For two signals whose Hurst exponents H are alike, their multifractal cross-correlation
characteristics described by q and hxy (q) are almost identical, while the divergence between q and hxy (q) becomes more
sizable for time series with more significant differences in auto-correlation (different Hurst exponent H).
Q. Fan, D. Li / Physica A 429 (2015) 1727 25

q
Fig. 8. When q (5, 5), Fxy (s) versus s obtained from MF-DXA. The lowest and the highest lines refer to q = 5 and q = 5, respectively.

Table 3
Cross-correlation scaling exponents obtained from MFCCA.
s < 30 30 < s < 270 s > 270

q = 0.6 1.4737 0.4594 0.1220


q = 0.8 1.2363 0.3654 0.0988
q = 1.0 1.0889 0.3078 0.0841
q = 1.2 0.9852 0.2682 0.0738
q = 1.4 0.9049 0.2387 0.0660
q = 1.6 0.8379 0.2155 0.0598
q = 1.8 0.7783 0.1961 0.0547
q = 2.0 0.7227 0.1794 0.0504
q = 2.2 0.6690 0.1645 0.0468
q = 2.4 0.6166 0.1509 0.0436
q = 2.6 0.5651 0.1382 0.0409
q = 2.8 0.5149 0.1262 0.0385
q = 3.0 0.4666 0.1149 0.0364
q = 3.2 0.4207 0.1041 0.0346
q = 3.4 0.3774 0.0939 0.0331
q = 3.6 0.3370 0.0844 0.0319
q = 3.8 0.2992 0.0756 0.0309
q = 4.0 0.2635 0.0678 0.0302

Therefore, the calculated q and hxy (q) are presented in Fig. 9. It is clearly visible that q is larger than hxy (q) for
q (0.6, 4), and they converge to each other for large values of q. For smaller qs, the difference between q and hxy (q) is
more evident, which implies that the dynamics of these two series is significantly different, but still cross-correlative. While
for larger qs, the convergence between q and hxy (q) implies that the analyzed process is ruled by similar fractal dynamics.
We can conclude that large fluctuations are much more strongly cross-correlated than the smaller ones. Complexity of the
multifractal cross-correlation is expressed by the range of q that is approximately 0.3405.

4.2.3. Discussion
We have investigated the cross-correlations between electricity price return and trading volume variation in Czech
market. The results show that electricity price return and trading volume variation series are persistent cross-correlated in
the short-term. Moreover, short-term cross-correlations display high complexity (multifractality) indicating that the cross-
correlations are easily to be affected by small and large fluctuations. However, the long-term cross-correlations display anti-
persistent and low complexity. The short-term cross-correlated behavior is different from long-term situations implying
that modeling the cross-correlations between price return and trading volume variation should consider the different terms
and periodic trends. In spot market, electricity price and trading volume are based on supply and demand. Moreover, the
electricity supply is weather dependent and demand for electricity is highly inelastic. As the majority of the electricity
supply is based on medium- and long-term contracts, extreme price movements are caused mainly by unexpected external
events (extremes of temperature or humidity, macroeconomic news, etc.). The combined effect of non-storability and the
connection of less efficient power sources pushes electricity price toward persistent behavior.
26 Q. Fan, D. Li / Physica A 429 (2015) 1727

Fig. 9. Multifractal cross-correlation exponents q and the average generalized Hurst exponents hxy (q) (hxy (q) = (hxx (q) + hyy (q))/2, here hxx (q) and
hyy (q) refer to the scaling exponents of individual series, respectively.), when q (0.6, 4).

5. Conclusion

In this paper, we investigate the multiscale cross-correlations between electricity price and trading volume in Czech
market. First, we analyze the auto-correlations of electricity price return and trading volume variation. We calculate the
scaling exponents by MF-DFA methodology, and find that both price return and trading volume variation series exhibit
long-range anti-persistent auto-correlations and multifractality. Furthermore, we discuss the cross-correlations between
electricity price and trading volume. We find the cross-correlations to be generally significant based on the analysis of the
significance of the statistic Qcc (m). The empirical result obtained by MFCCA methodology implies that there exist long-term
anti-persistent cross-correlations between electricity price and trading volume for q (0.6, 4), and the fluctuation function
q
Fxy (s) fluctuates around zero when q varies from 2 to 0. The results obtained by MFCCA indicate that the multifractal
cross-correlation characterizes only relatively large fluctuations of the signals under investigation. From the perspective of
multifractal cross-correlation, smaller fluctuations that are filtered out by q < 0 may be considered mutually independent.
We also analyze the relation between generalized Hurst exponent and the multifractal scaling parameter q . We find
that the difference between the generalized Hurst exponent and the multifractal scaling parameter is significantly different
for smaller fluctuations, which indicates that the multifractal character of cross-correlations resembles more each other for
electricity price and trading volume on the level of large fluctuations and weakens for the smaller ones.
The MFCCA method considers the original sign of the cross-covariance function fxy2
(s, v), so this new method can eliminate
the risk of appearance of complex values that might lead to problems with their correct interpretation, and prohibit losing
information that is stored in the negative cross-covariance. It provides us more convincing and complete information about
the fractal cross-correlations of empirical data under investigation.

Acknowledgments

Qingju Fan and Dan Li sincerely thank the main Editor H.E. Stanley and the anonymous reviewers for their helpful
comments and suggestions. The authors are grateful for the financial support from the Fundamental Research Funds for
the Central Universities (Grant No. 2014-Ia-038), the National Natural Science Foundation of China (Grant No. 61304145),
and the Research Fund for the Doctoral Program of Higher Education (Grant No. 20130009120016).

References

[1] H. Geman, A. Roncoroni, Understanding the fine structure of electricity prices, J. Bus. 79 (3) (2006) 12251261.
[2] I. Simonsen, Measuring anti-correlations in the nordic electricity spot market by wavelets, Physica A 322 (2003) 597606.
[3] S. Koopman, M. Ooms, M. Carnero, Periodic seasonal Reg-ARFIMA-GARCH models for daily electricity spot prices, J. Amer. Statist. Assoc. 102 (477)
(2007) 1627.
[4] P. Malo, Modeling electricity spot and futures price dependence: a multifrequency approach, Physica A 388 (2009) 47634779.
[5] J. Alvarez-Ramirez, R. Escarela-Perez, Time-dependent correlations in electricity markets, Energy Econ. 32 (2010) 269277.
[6] L. Kristoufek, P. Lunackova, Long-term memory in electricity prices: Czech market evidence, Czech J. Econ. Finance 63 (5) (2013) 407424.
[7] F. Wang, G.P. Liao, J. Li, X. Li, T. Zhou, Multifractal detrended fluctuation analysis for clustering structures of electricity price periods, Physica A 392
(22) (2013) 57235734.
[8] J. Kwapien, S. Drozdz, Physical approach to complex systems, Phys. Rep. 515 (5) (2012) 115226.
Q. Fan, D. Li / Physica A 429 (2015) 1727 27

[9] M.S. Movahed, G.R. Jafari, F. Ghasemi, S. Rahvar, M.R.R. Tabar, Multifractal detrended fluctuation analysis of sunspot time series, J. Stat. Mech. Theory
Exp. 2 (2006) 115.
[10] P. Varotsos, N. Sarlis, E. Skordas, Long-range correlations in the electric signals that precede rupture: further investigations, Phys. Rev. E 67 (13) (2003)
021109.
[11] B. Podobnik, D. Horvatic, A.M. Petersen, H. Stanley, Cross-correlations between volume change and price change, Proc. Natl. Acad. Sci. USA 106 (2009)
22079.
[12] A. Lin, P. Shang, X. Zhao, The cross-correlations of stock markets based on DCCA and time-delay DCCA, Nonlinear Dynam. 67 (2012) 425435.
[13] W. Shi, P. Shang, J. Wang, A. Lin, Multiscale multifractal detrended cross-correlation analysis of financial time series, Physica A 403 (2014) 3544.
[14] B. Podobnik, D. Wang, D. Horvatic, I. Grosse, H.E. Stanley, Time-lag cross-correlations in collective phenomena, Europhys. Lett. EPL 90 (2010) 68001.
[15] C. Matsoukas, S. Islam, Detrended fluctuation analysis of rainfall and streamflow time series, J. Geophys. Res.: Atmos. 105 (23) (2000) 2916529172.
[16] C. Peng, S. Buldyrev, A. Goldberger, S. Havlin, M. Simons, H. Stanley, Mosaic organization of DNA nucleotides, Phys. Rev. E 49 (2) (1994) 16851689.
[17] J. Kantelhardt, S. Zschiegner, E. Koscielny-Bunde, S. Havlin, A. Bunde, H. Stanley, Multifractal detrended fluctuation analysis of nonstationary time
series, Physica A 316 (14) (2002) 87114.
[18] B. Podobnik, H. Stanley, Detrended cross-correlation analysis: a new method for analyzing two nonstationary time series, Phys. Rev. Lett. 100 (2008)
084102.
[19] W. Zhou, Multifractal detrended cross-correlation analysis for two nonstationary signals, Phys. Rev. E 77 (2008) 066211.
[20] G. Gu, W. Zhou, Detrending moving average algorithm for multifractals, Phys. Rev. E 82 (2010) 011136.
[21] Z. Jiang, W. Zhou, Multifractal detrending moving average cross-correlation analysis, Phys. Rev. E 84 (2011) 016106.
[22] L. Kristoufek, Multifractal height cross-correlation analysis: a new method for analyzing long-range cross-correlations, Europhys. Lett. EPL 95 (6)
(2011) 68001.
[23] L. Hedayatifar, M. Vahabi, G. Jafari, Coupling detrended fluctuation analysis for analyzing coupled nonstationary signals, Phys. Rev. E 84 (2) (2011)
021138.
[24] G. Zebende, DCCA cross-correlation coefficient: quantifying level of cross-correlation, Physica A 390 (2011) 614618.
[25] B. Podobnik, Z. Jiang, W. Zhou, H. Stanley, Statistical tests for power-law cross-correlated processes, Phys. Rev. E 84 (2011) 066118.
[26] F. Wang, G.P. Liao, J.W. Li, R.B. Zou, W. Shi, Cross-correlation detection and analysis for Californias electricity market based on analogous multifractal
analysis, Chaos 23 (1) (2013) 013129.
[27] P. Oswiecimka, S. Drozdz, M. Forczek, S. Jadach, J. Kwapien, Detrended cross-correlation analysis consistently extended to multifractality, Phys. Rev.
E 89 (2) (2014) 066118.
[28] X. Zhao, P. Shang, A. Lin, G. Chen, Multifractal Fourier detrended cross-correlation analysis of traffic signals, Physica A 390 (2011) 36703678.
[29] G.X. Cao, L.B. Xu, J. Gao, Multifractal detrended cross-correlations between the Chinese exchange market and stock market, Physica A 391 (2012)
48554866.
[30] B. Podobnik, I. Grosse, D. Horvatic, S. Ilic, P. Ivanov, H. Stanley, Quantifying cross-correlations using local and global detrending approaches, Eur. Phys.
J. B 71 (2009) 243.
[31] F. Wang, G. Liao, X. Zhou, W. Shi, Multifractal detrended cross-correlation analysis for power markets, Nonlinear Dynam. 72 (2013) 353363.
[32] Y. Yuan, X. Zhuang, Z. Liu, Price-volume multifractal analysis and its application in Chinese stock markets, Physica A 391 (2012) 34843495.
[33] F. Sensfuss, M. Ragwitzm, M. Genoese, The merit-order effect: a detailed analysis of the price effect of renewable electricity generation on spot market
prices in Germany, Energy Policy 36 (8) (2008) 30863094.
[34] G.M. Ljung, G.E.P. Box, On a measure of a lack of fit in time series models, Biometrika 65 (1978) 297303.
[35] E. Siqueira, T. Stosic, L. Bejan, B. Stosic, Correlations and cross-correlations in the Brazilian agrarian commodities and stocks, Physica A 389 (14) (2010)
27392743.
[36] S. Shadkhoo, G. Jafari, Multifractal detrended cross-correlation analysis of temporal and spatial seismic data, Eur. Phys. J. B 72 (2009) 679683.