Research Article  Open Access
AiJing Lin, PengJian Shang, HuaChun Zhou, "Effects of Exponential Trends on Correlations of Stock Markets", Mathematical Problems in Engineering, vol. 2014, Article ID 340845, 11 pages, 2014. https://doi.org/10.1155/2014/340845
Effects of Exponential Trends on Correlations of Stock Markets
Abstract
Detrended fluctuation analysis (DFA) is a scaling analysis method used to estimate longrange powerlaw correlation exponents in time series. In this paper, DFA is employed to discuss the longrange correlations of stock market. The effects of exponential trends on correlations of Hang Seng Index (HSI) are investigated with emphasis. We find that the longrange correlations and the positions of the crossovers of lower order DFA appear to have no immunity to the additive exponential trends. Further, our analysis suggests that an increase in the DFA order increases the efficiency of eliminating on exponential trends. In addition, the empirical study shows that the correlations and crossovers are associated with DFA order and magnitude of exponential trends.
1. Introduction
Financial markets have been known as representative complex dynamic systems, which are organized by various unexpected phenomena and affected by external factors. The study of financial markets is not easy because we do not know the control parameters that govern the systems. However, the bulk of the literature focused on the longrange correlations of financial time series in the past few years. New ideas and techniques have been introduced to measure the longrange correlation behaviors. The understanding of such longrange correlations would be helpful to design good portfolios. Longrange correlations were first observed by Barkoulas et al. in the discussion of Greek stock market [1]. In recent years, many researchers have found evidence of longrange correlations for stock markets [2–5]. To analyze the longrange correlations, previous studies presented various methods, such as rescaled range analysis (R/S analysis), wavelet transform modulus maxima (WTMM), and detrended fluctuation analysis (DFA) [6, 7]. The DFA [8] method invented by Peng et al. has become a widely used method for the determination and detection of longrange correlations in time series. It also has successfully been applied to diverse fields [9–25] such as DNA sequence, heart rate dynamics, solid state physics, longtime weather records, earthquakes, economics, and culturomics. One reason to employ the DFA method is to avoid spurious detection of correlations that are artifacts of nonstationarities in the time series. Financial time series are influenced by external trends. Strong trends in series may lead to a false detection of longrange correlations. For the reliable detecting of longrange correlations, it is essential to distinguish trends from the intrinsic longrange fluctuations.
An important concept used in the investigation by DFA method is fluctuation scaling. In case of a monofractal time series, the fluctuation scaling may be described by a single exponent (the Hurst exponent). But there is a class of series that is much more complex and whose scaling exponents are different in different magnitudes [26–31]. The turning points of scaling changing are the socalled crossovers. In this paper, detrended fluctuation analysis (DFA) method is employed to analyze the longrang correlations of stock markets. The main purpose of this study is to reveal how sensitive indeed are exponential trends to longrange correlation behaviors of stock markets. The effects of exponential trends on correlations of Hang Seng Index (HSI) were investigated with emphasis.
The organization of this paper is as follows. In Section 2, the methodology is introduced. In Section 3, the empirical data under consideration in this paper is described. In Section 4, the correlation behaviors of the original time series are discussed. The results and analysis of the effects of exponential trends on correlations are presented in Section 5. Finally, it ends with a conclusion.
2. Methodology
2.1. Absolute Returns
In recent years, the studies about stock returns and volatility have attracted much attention. There are extensive literature testing for longrange correlations in stock returns and volatility. One of the important reasons is that a few papers have suggested that economic systems gave long memory properties. There are many implications for both portfolio and risk management that stem from longrange correlation characteristics in stock returns and volatility. The empirical evidence has suggested strong evidence of longrange correlations in volatility and however weaker evidence for stock returns. In our paper, we use absolute log returns as proxies for volatility. We will give the definition of absolute log returns in the following.
Let denote the price of a financial asset at time . We define returns over a time horizon as . The absolute log returns over a time horizon are then defined as .
2.2. DFA Method
The DFA procedure consists of five steps. Let us suppose that is a series of length ; the DFA procedure involves the following steps.
Step 1. Determine the profile
Step 2. Cut the profile into nonoverlapping segments of equal length . Since the record length need not be a multiple of the considered time scale , 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, segments are obtained altogether.
The subtraction of the mean is not compulsory, since it would be eliminated by the later detrending in the third step anyway.
Step 3. Calculate the local trend for each segment by a leastsquare fit of the data. Then we define the detrended time series for segment duration , denoted by , as the difference between the original time series and the fits:
where is the fitting polynomial in the segment. Linear, cubic, or higher order polynomials can be used in the fitting procedure (DFA1, DFA3, and higher order DFA). Since the detrending of the time series is done by the subtraction of the polynomial fits from the profile, these methods differ in their capability of eliminating trends in the series. In order DFA, trends of order in the profile and of order in the original series are eliminated.
We calculate—for each of the segments—the variance as follows:
of the detrended time series by averaging over all data points in the segment.
Step 4. Average over all segments to obtain the fluctuation function: It is apparent that will increase with increasing. Of course, depends on the DFA order . By construction, is only defined for .
Step 5. Determine the scaling behavior of the fluctuation functions by analyzing loglog plots versus . If the series are longrange powerlaw correlated, increases, for large values of , as a powerlaw The exponent is called the scaling exponents or correlation exponents, which represents the correlation properties of the signal.
2.3. Correlations of Time Series
For , we are interested in the correlation of the values and for different time lags, that is, correlations over different time scales . In order to get rid of a constant offset in the data, the mean is usually subtracted. Quantitatively, correlations between values separated by steps are defined by the (auto)correlation function as follows:
If are uncorrelated, is zero for . Shortrange correlations of are described by declining exponentially and with a decay time . For socalled longrange correlations, declines as a powerlaw with an exponent . A direct calculation of is usually not appropriate due to noise superimposed on the collected data and due to underlying trends of unknown origin. For example, the average might be different for the first and the second half of the record, if the data are strongly longrange correlated. This makes the definition of problematic. Thus, we have to determine the correlation exponent indirectly. A similar approximation for is We find The correlation exponent can be determined by measuring the fluctuation exponent . The correlations of stock market are calculated from realized returns on each market. Correlations tend to increase in times of large shocks to returns such as a stock market crash. Larger markets are more liquid and display more price moment than smaller counterparts. Smaller markets may not react as quickly to relevant information because some stocks may be traded infrequently. In the aftermath of October 1987 stock market crash, more attention was afforded to stock market correlation and the related concept of longrange correlation.
3. Data Description
The data used in this paper are daily closing prices of two stock indices. Form the period January 1988 to the period June 2007. The data sets consist of New York Stock Exchange Composite Index (NYSE) and Hong Kong Heng Seng Index (HSI) covering 4896 and 4810 days, respectively. The original data set used in our studies is provided by Yahoo Finance website. Two original time series are shown in Figure 1. For each stock closing price, the monthly absolute log return is calculated by . The graphs of monthly absolute log returns versus the time are shown in Figure 2.
4. The DFA of Absolute Returns
We calculate the absolute returns over several intervals: 1 day, 5 days, 1 month (22 days), 6 months (132 days), corresponding to , , , and , respectively. Then we detect the correlation behaviors of absolute returns data of NYSE and HSI by using DFA method. Figures 3 and 4 present the fluctuation function as a function of time scale . It is demonstrated that multifractal properties of NYSE and HSI are similar for different order DFA. However, the positions of crossovers depend on time intervals . The time scales of crossovers increase significantly with the increase of . The crossover is interpreted as a result of different correlation properties for small and large scales in the signal; however, it may also be caused by external trends.
In order to provide more details about the correlations properties of absolute log returns, we calculate the scaling exponents and correlation exponents for . The scaling exponents and correlation exponents are shown in Tables 1 and 2. We observe that both NYSE and HSI absolute log returns show longrange correlated behaviors. It is noted that HSI is more correlated than NYSE based on DFA method.


5. Effects of Exponential Trends on Correlations
5.1. Corrupted with Exponential Trends
Time series from real world are often affected by trends, which have to be well distinguished from the original series. We have studied the correlations properties of stock market by applying the DFA method in Section 4. In this section, we survey the effect of exponential trend on DFA. The effects of exponential trends on correlations of HSI are investigated with emphasis. We attempt to generate a new artificial data by adding trends on as follows: In this paper, is the daily absolute log returns data of HSI. is the number of the data.
5.2. DFA on Exponential Trends
We consider the exponential trends in the form with . In the case of and , we demonstrate the effects of exponential trends on DFA in Figure 5 and Table 3. In the case of and , we show the effects of exponential trends on DFA in Figure 6 and Table 4. In this paper, presents the scaling exponents before the crossover is and after the crossover is . For example, in Table 3, means that the scaling exponents before crossover is and after crossover is .


In the first case, we find that the slope of the fluctuation function versus the scales obtained from the DFA method slightly depends on the value of parameter . When we increase the parameter , the positions of the crossovers move to small scales, that is to say, the bigger the values of , the smaller the scales of the crossovers. In the second case, we obtain that scaling characteristics of fluctuation function versus scales are immune to parameters . There is no visible difference among the results obtained for all values of .
5.3. Effects of DFA on Exponential Trends
5.3.1. Effects of DFA on Strong Exponential Trends
In this section, to reliably describe the effects of DFA method on strong exponential trends, we construct artificial time series by adjusting the values of and . For artificial data with trends and , the DFA fluctuation versus the time scales is plotted in Figure 7. The curve exhibits a crossover at in Figure 7(a), with a slope for and for . For , the curve presents a longrange correlated behavior, but, for , there is no evidence of longrange correlation.
(a)
(b)
(c)
The positions of move to small scales for the same order DFA when we increase . As we expected, the strong exponential trends are markedly filtered by higher order DFA, although apparent crossovers are still observed for large time scales . The positions of the crossovers and the changes of the scaling exponents before and after the crossovers are shown in Table 5. The positions of the crossovers have been determined from the intersection of linear fits done on both sides of the crossovers. We can get the same results from Table 5, compared to Figure 7. Furthermore, the results are more credible.

Secondly, we study how different trends and affect the scaling behavior of by applying DFA1 to DFA5. For artificial data corrupted with trends and , the fluctuation functions versus the time scales are shown in Figure 8. In Table 5, the detail values of the crossovers and the changes of scaling exponents are presented.
In Figure 8 (), the curve exhibits a crossover at , with a slope for and for . For , the curve presents a longrange correlated behavior, but, for , there is no evidence of longrang correlation. We can find from Figure 8 and Table 6 that in case , for , the exponential trends cannot be eliminated; for , the exponential trends are mostly eliminated; for the exponential trends are significantly eliminated. In case , for , the exponential trends cannot be eliminated; for , the exponential trends are mostly eliminated; for , the exponential trends are significantly eliminated.

5.3.2. Effects of DFA on Weak Exponential Trends
For weak trends, we perform the similar procedure as we do in studying strong trends. For , the results of DFA on are the same as DFA on ; hence, we only give the corresponding scaling exponents of DFA1. The corresponding scaling exponents of DFA1 by means of fixing and varying are revealed in Table 7. The corresponding scaling exponents by means of fixing and varying are shown in Table 8.


When and is small, there are no obvious crossovers and the values of scaling exponents are decreasing with the reducing of the values of . When , the value of fluctuation exponents is 0.8328, which is the same as the value of the original time series. When and is small, the similar consequences can be concluded. When the value of fluctuation exponents is 0.8328. The results indicate that if , there are no obvious crossovers; if , the exponential trends have no effects on scaling exponents of each order DFA.
6. Conclusion
In this paper, the longrange correlation behaviors in absolute log returns are investigated by applying DFA. We find that the absolute log returns of NYSE and HSI are longrange correlated; however, the longrange correlation of NYSE is stronger than HSI. The effect of different exponential trends on the DFA scaling behavior and the crossovers of longrange correlated signals are also studied. The results suggest that longrange correlation behaviors of higher order DFA are not affected by exponential trends, while the effects of exponential trends depend on parameters and . The real financial data are often influenced by different trends which are caused by external effects. So it is of great importance to investigate the properties of the trends and the effects of the trends on the dynamical changes of original signal. There are several issues for further research. The first issue concerns application of our method to other types of trends. The second issue considers predicting the stock markets based on our current results due to the crossover being the representation of economic cycle in some sense. Our studies are relevant for a better understanding of the stock market mechanism and may be useful for identifying economic turning points and forecasting stock volatility. Further studies are necessary from theoretical and empirical perspective.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
This work is supported by the National Natural Science Foundation of China (Grant nos. 61304145, 61071142, and 61371130), the China Postdoctoral Science Foundation (Grant no. 2012M520156), the Fundamental Research Funds for Central Universities of China (Grant no. 2013JBM089), the Research Fund for the Doctoral Program of Higher Education (Grant no. 20130009120016), and the National High Technology Research Development Program of China (Grant no. 2011AA110306).
References
 J. T. Barkoulas, C. F. Baum, and N. Travlos, “Long memory in the Greek stock market,” Applied Financial Economics, vol. 10, no. 2, pp. 177–184, 2000. View at: Google Scholar
 T. Jagric, B. Podobnik, and M. Kolanovic, “Does the efficient market hypothesis hold? Evidence from six transition economies,” Eastern European Economics, vol. 43, no. 4, pp. 79–103, 2005. View at: Google Scholar
 Y.W. Cheung and K. S. Lai, “A search for long memory in international stock market returns,” Journal of International Money and Finance, vol. 14, no. 4, pp. 597–615, 1995. View at: Google Scholar
 J. Nikkinena, M. M. Omranb, P. Sahlström, and J. A. Äijö, “Stock returns and volatility following the September 11 attacks: evidence from 53 equity markets,” International Review of Financial Analysis, vol. 17, no. 1, pp. 27–46, 2008. View at: Google Scholar
 D. O. Cajueiro and B. M. Tabak, “Testing for longrange dependence in world stock markets,” Chaos, Solitons and Fractals, vol. 37, no. 3, pp. 918–927, 2008. View at: Publisher Site  Google Scholar
 A. Leonidov, “Long memory in stock trading,” International Journal of Theoretical and Applied Finance, vol. 7, no. 7, pp. 879–885, 2004. View at: Publisher Site  Google Scholar
 D.S. Ho, C.K. Lee, C.C. Wang, and M. Chuang, “Scaling characteristics in the Taiwan stock market,” Physica A, vol. 332, no. 1–4, pp. 448–460, 2004. View at: Publisher Site  Google Scholar
 C.K. Peng, S. V. Buldyrev, S. Havlin, M. Simons, H. E. Stanley, and A. L. Goldberger, “Mosaic organization of DNA nucleotides,” Physical Review E, vol. 49, no. 2, pp. 1685–1689, 1994. View at: Publisher Site  Google Scholar
 R. Nagarajan, “Effect of coarsegraining on detrended fluctuation analysis,” Physica A, vol. 363, no. 2, pp. 226–236, 2006. View at: Publisher Site  Google Scholar
 C.K. Peng, J. E. Mietus, Y. Liu et al., “Quantifying fractal dynamics of human respiration: age and gender effects,” Annals of Biomedical Engineering, vol. 30, no. 5, pp. 683–692, 2002. View at: Publisher Site  Google Scholar
 J. W. Kantelhardt, S. A. Zschiegner, E. KoscielnyBunde, S. Havlin, A. Bunde, and H. E. Stanley, “Multifractal detrended fluctuation analysis of nonstationary time series,” Physica A, vol. 316, no. 1–4, pp. 87–114, 2002. View at: Publisher Site  Google Scholar
 K. Hu, P. C. Ivanov, Z. Chen, P. Carpena, and H. E. Stanley, “Effect of trends on detrended fluctuation analysis,” Physical Review E, vol. 64, no. 1, Article ID 011114, 19 pages, 2001. View at: Google Scholar
 Z. Chen, P. C. Ivanov, K. Hu, and H. E. Stanley, “Effect of nonstationarities on detrended fluctuation analysis,” Physical Review E, vol. 65, no. 4, Article ID 041107, 15 pages, 2002. View at: Publisher Site  Google Scholar
 R. Nagarajan and R. G. Kavasseri, “Minimizing the effect of periodic and quasiperiodic trends in detrended fluctuation analysis,” Chaos, Solitons and Fractals, vol. 26, no. 3, pp. 777–784, 2005. View at: Publisher Site  Google Scholar
 P. Shang, Y. Lu, and S. Kamae, “Detecting longrange correlations of traffic time series with multifractal detrended fluctuation analysis,” Chaos, Solitons and Fractals, vol. 36, no. 1, pp. 82–90, 2008. View at: Publisher Site  Google Scholar
 P. Shang, A. Lin, and L. Liu, “Chaotic SVD method for minimizing the effect of exponential trends in detrended fluctuation analysis,” Physica A, vol. 388, no. 5, pp. 720–726, 2009. View at: Publisher Site  Google Scholar
 A. Lin and P. Shang, “Minimizing periodic trends by applying Laplace transform,” Fractals, vol. 19, no. 2, pp. 203–211, 2011. View at: Publisher Site  Google Scholar  Zentralblatt MATH  MathSciNet
 A. Lin, P. Shang, and H. Ma, “The Orthogonal Vsystem detrended fluctuation analysis,” Fluctuation and Noise Letters, vol. 10, no. 2, pp. 189–206, 2011. View at: Publisher Site  Google Scholar
 D. Horvatic, H. E. Stanley, and B. Podobnik, “Detrended crosscorrelation analysis for nonstationary time series with periodic trends,” Europhysics Letters, vol. 94, no. 1, Article ID 18007, 2011. View at: Publisher Site  Google Scholar
 L. Telesca and M. Macchiato, “Timescaling properties of the UmbriaMarche 19971998 seismic crisis, investigated by the detrended fluctuation analysis of interevent time series,” Chaos, Solitons and Fractals, vol. 19, no. 2, pp. 377–385, 2004. View at: Publisher Site  Google Scholar
 R. L. Costa and G. L. Vasconcelos, “Longrange correlations and nonstationarity in the Brazilian stock market,” Physica A, vol. 329, no. 12, pp. 231–248, 2003. View at: Publisher Site  Google Scholar
 M. C. Mariani, I. Florescu, M. P. Beccar Varela, and E. Ncheuguim, “Study of memory effects in international market indices,” Physica A, vol. 389, no. 8, pp. 1653–1664, 2010. View at: Publisher Site  Google Scholar
 J. B. Gao, J. Hu, X. Mao, and M. Perc, “Culturomics meets random fractal theory: insights into longrange correlations of social and natural phenomena over the past two centuries,” Journal of the Royal Society Interface, vol. 9, no. 73, pp. 1956–1964, 2012. View at: Google Scholar
 A. M. Petersen, J. N. Tenenbaum, S. Havlin, H. E. Stanley, and M. Perc, “Languages cool as they expand: allometric scaling and the decreasing need for new words,” Scientific Reports, vol. 2, article 943, 2012. View at: Google Scholar
 M. Perc, “Selforganization of progress across the century of physics,” Scientific Reports, vol. 3, article 1720, 2013. View at: Google Scholar
 R. B. Govindan, D. Vjushin, S. Brenner, A. Bunde, S. Havlin, and H.J. Schellnhuber, “Longrange correlations and trends in global climate models: comparison with real data,” Physica A, vol. 294, no. 12, pp. 239–248, 2001. View at: Publisher Site  Google Scholar
 J. A. Scheinkman and B. LeBaron, “Nonlinear dynamics and stock returns,” Journal of Business, vol. 62, no. 3, pp. 311–337, 1989. View at: Google Scholar
 M. H. Pesaran and A. Timmermann, “Predictability of stock returns: robustness and economic significance,” The Journal of Finance, vol. 50, no. 4, pp. 1201–1228, 1995. View at: Google Scholar
 K. J. M. Cremers, “Stock return predictability: a Bayesian model selection perspective,” Review of Financial Studies, vol. 15, no. 4, pp. 1223–1249, 2002. View at: Google Scholar
 W. Breen, L. R. Glosten, and R. Jagannathan, “Economic significance of predictable variations in stock index returns,” The Journal of Finance, vol. 44, no. 5, pp. 1177–1189, 1989. View at: Google Scholar
 C.K. Peng, S. Havlin, H. E. Stanley, and A. L. Goldberger, “Quantification of scaling exponents and crossover phenomena in nonstationary heartbeat time series,” Chaos, vol. 5, no. 1, pp. 82–87, 1995. View at: Google Scholar
Copyright
Copyright © 2014 AiJing Lin et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.