Long-range correlation analysis of economic news flow intensity
Introduction
The output signals generated by a complex system often exhibit long-range correlations. To have a better understanding for the dynamics of the underlying complex systems it is important to estimate such long-range correlations. Different methods were developed to detect long-range correlations in time series, among them the rescaled range analysis [1], the wavelet transform module maxima approach [2], the fluctuation analysis (FA) [3], the detrended fluctuation analysis (DFA) [4], the detrending moving average analysis (DMA) [5], and many others.
DFA is one of the well-known and used method for indirect scaling the long-range auto-correlation in non-stationary time series. The method was proposed in the paper [4] extending the ideas of the work [3]. Since then DFA proved its effectiveness for many scientific and engineering problems such as DNA analysis [6], [7], [8], human gait [9], [10], analysis of dynamics for daily internet traffic [11], biomedical signal processing [12], [13], [14], [15], [16], finance and economical time series [17], [18], [19], [20], [21], and many others.
It should be noted that the DFA has some drawbacks. The paper [22] states that DFA
- 1.
can lead to uncontrolled bias;
- 2.
is more expensive than unbiased estimator;
- 3.
cannot provide protection against nonlinear nonstationaries.
In order to avoid one-sided results, we will compare the performances of four estimators (Fluctuation Analysis (FA), Detrended Fluctuation Analysis of orders 1, 2 and 3 (DFA1, DFA2 and DFA3), and Detrending Moving Average (DMA)) in their capability to study long-range auto-correlation in time series of news flow intensity. The work [23] points out that DMA and DFA remain “The Methods of Choice” in determining the Hurst index of time series.
Our empirical analytics used the time series of news analytics data from Raven Pack. News analytics is a relatively new tool designed to improve the trading strategies of investors. Knowing the characteristics of news in numerical indices one can use them in mathematical and statistical models [24] and automated trading systems [25]. Currently, the tools of the news analytics have been increasingly used by traders in the US and Europe.
There are more than 50 providers of economic news in the world. Bloomberg, Dow Jones and Thomson Reuters are the three largest of them. About 200 agencies are involved in providing financial analytics. In our research we use the Raven Pack data, one of the most well-known providers of news analytics data.
Two-month period was divided into non-overlapping consecutive segments of equal (small) length. Using news analytics data from Raven Pack we calculated the amount of economical and finance news reported in the world during each interval of time. Obtained in such a way time series of news flow intensity were used for the long-range auto-correlation analysis.
Section snippets
News analytics data
News analytics is a new approach to the analysis of news flow based on methods of artificial intelligence. The process of news analysis in information systems is automated and usually includes the following steps:
- 1.
collecting news from different sources;
- 2.
preliminary analysis of news;
- 3.
analysis of news-related expectations (sentiments) taking into account the current market situation;
- 4.
design and use of quantitative models.
Types of auto-correlation
Let be a time series with large and let , . The (auto) correlation between and is defined by the correlation function where denotes the expected value of . The following types of correlation can be distinguished:
- 1.
are uncorrelated; it is clear that if and are uncorrelated then must be equal to zero, ;
- 2.
the case of short-range correlations of the leads to exponentially declining of
Empirical results
We constructed the time series of the amount of news per 1, 2 and 5 min as it was described in Section 2 from the news analytics data set containing the list of all news (Raven Pack data set). We analyzed three data sets shown in Table 1. We use FA, DFA (of order 1, 2 and 3) and DMA methods to quantify the correlation and scaling properties of the time series.
Then, using FA, DFA and DMA methods, we obtained values of for different segment lengths .
Papers [30], [31], [32]
Conclusion
In attempt to detect the possible presence of long-range correlation of economic news flow intensity, we applied a different techniques (FA, DFA, DMA). The results clearly show that long-range power-law correlation exists in news flow. The examination of power-law slap by means of DFA (of order 1, 2 and 3) methods suggests that news messages intensity behave close to noise for . FA and DMA are run both on the original and on the deseasonalized data, with indistinguishable results.
The
Acknowledgment
The first author was supported by the Ministry of Education and Science of the Russian Federation (project 1.1520.2014K), the third author was supported by RFBR (grant 13-01-00175).
References (32)
- et al.
Statistical properties of DNA sequences
Physica A
(1995) - et al.
Analysis of DNA sequences using methods of statistical physics
Physica A
(1998) - et al.
Multiple-time scales analysis of physiological time series under neural control
Physica A
(1998) - et al.
Heart rate variability analysis using correlation dimension and detrended fluctuation analysis
ITBM-RBM
(2002) - et al.
Detrended fluctuation analysis of short-term heart rate variability in late pregnant women
Auton. Neurosci.-Basic Clin.
(2009) - et al.
Correlations in economic time series
Physica A
(1997) - et al.
Volatility distribution in the S&P500 stock index
Physica A
(1997) - et al.
Applications of statistical physics to economic and financial topics
Physica A
(1999) - et al.
Introducing false EUR and false EUR exchange rates
Physica A
(2000) - et al.
Detecting long-range correlations with detrended fluctuation analysis
Physica A
(2001)
On the scaling ranges of detrended fluctuation analysis for long-term memory correlated short series of data
Physica A
Long-term storage capacity of reservoirs
Trans. Amer. Soc. Civ. Eng.
On the wavelet transformation of fractal objects
J. Stat. Phys.
Long-range correlations in nucleotide sequences
Nature
Mosaic organization of DNA nucleotides
Phys. Rev. E
Second-order moving average and scaling of stochastic time series
Eur. Phys. J. B
Cited by (11)
Parameter identification for mixed fractional Brownian motions with the drift parameter
2019, Physica A: Statistical Mechanics and its ApplicationsDynamic relationship between Japanese Yen exchange rates and market anxiety: A new perspective based on MF-DCCA
2017, Physica A: Statistical Mechanics and its ApplicationsCitation Excerpt :takes into consideration partial correlation analysis to quantify the hidden multifractal nature of two time series. Empirically, Sidorov et al. [24] adopted FA, DFA and DMA to analyze the long-range dependence of economic news flow intensity.
Arbitrage with fractional Gaussian processes
2017, Physica A: Statistical Mechanics and its ApplicationsCitation Excerpt :This fact motivates the emergence of alternative models to the pioneering Black–Scholes model. A case which has drawn much attention is the fractional Brownian motion (fBm), which displays the long-range dependence observed in empirical data (see, [2–24] and the references therein). Since fBm is not a semimartingale, the general results of mathematical finance in [25] already imply that it allows a certain kind of arbitrage.
Fractality and multifractality analysis of news sentiments time series
2018, IAENG International Journal of Applied MathematicsGraph-based clustering approach for economic and financial event detection using news analytics data
2018, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)