Long-range correlation analysis of economic news flow intensity

doi:10.1016/j.physa.2015.10.025

Physica A: Statistical Mechanics and its Applications

Volume 444, 15 February 2016, Pages 205-212

https://doi.org/10.1016/j.physa.2015.10.025 Get rights and content

Highlights

•
Auto-correlation properties for time series of the news flow intensity are examined.
•
News analytics data shows the presence of long-range correlations for the time series of news intensity data.
•
The detrended fluctuation analysis and the detrending moving average analysis are run both on the original and on the deseasonalized data.

Abstract

The goal of the paper is to examine the auto-correlation properties for time series of the news flow intensity using different methods, such as the fluctuation analysis, the detrended fluctuation analysis and the detrending moving average analysis. Empirical findings for news analytics data show the presence of long-range correlations for the time series of news intensity data.

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.

The review of methods and tools of news analytics can be found in book [25].

Types of auto-correlation

Let $X = {(x_{t})}_{t = 1}^{n}$ be a time series with large $n$ and let $s \in N$ , $s ≪ n$ . The (auto) correlation between $X_{1} = {(x_{t})}_{t = 1}^{n - s}$ and $X_{2} = {(x_{t + s})}_{t = 1}^{n}$ is defined by the correlation function $C (s) = E (X_{1} X_{2}) = \frac{1}{n - s} \sum_{t = 1}^{n - s} x_{t} x_{t + s},$ where $E (\cdot)$ denotes the expected value of $(\cdot)$ . The following types of correlation can be distinguished:

1.
$x_{t}$ are uncorrelated; it is clear that if $X_{1}$ and $X_{s}$ are uncorrelated then $C (s)$ must be equal to zero, $C (s) = 0$ ;
2.
the case of short-range correlations of the ${(x_{t})}_{t = 1}^{n}$ leads to exponentially declining of $C (s$

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 $F (s_{i})$ for different segment lengths $s_{i} \in [10^{1}, 10^{3}]$ .

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 $1 / f$ noise for $s > 10^{2}$ . 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)

C. Peng et al.
Statistical properties of DNA sequences
Physica A
(1995)
S. Buldyrev et al.
Analysis of DNA sequences using methods of statistical physics
Physica A
(1998)
C.-K. Peng et al.
Multiple-time scales analysis of physiological time series under neural control
Physica A
(1998)
R. Acharya et al.
Heart rate variability analysis using correlation dimension and detrended fluctuation analysis
ITBM-RBM
(2002)
R.G. Yeh et al.
Detrended fluctuation analysis of short-term heart rate variability in late pregnant women
Auton. Neurosci.-Basic Clin.
(2009)
Y. Liu et al.
Correlations in economic time series
Physica A
(1997)
P. Cizeau et al.
Volatility distribution in the S&P500 stock index
Physica A
(1997)
M. Ausloos et al.
Applications of statistical physics to economic and financial topics
Physica A
(1999)
M. Ausloos et al.
Introducing false EUR and false EUR exchange rates
Physica A
(2000)
J.W. Kantelhardt et al.
Detecting long-range correlations with detrended fluctuation analysis
Physica A
(2001)

D. Grech et al.

On the scaling ranges of detrended fluctuation analysis for long-term memory correlated short series of data

Physica A

(2013)

H. Hurst

Long-term storage capacity of reservoirs

Trans. Amer. Soc. Civ. Eng.

(1951)

M. Holschneider

On the wavelet transformation of fractal objects

J. Stat. Phys.

(1988)

C. Peng et al.

Long-range correlations in nucleotide sequences

Nature

(1992)

C. Peng et al.

Mosaic organization of DNA nucleotides

Phys. Rev. E

(1994)

E. Alessio et al.

Second-order moving average and scaling of stochastic time series

Eur. Phys. J. B

(2002)

Cited by (11)

Parameter identification for mixed fractional Brownian motions with the drift parameter
2019, Physica A: Statistical Mechanics and its Applications
This paper deals with the problem of estimating the unknown parameters in the drift mixed fractional Brownian motion based on discrete-sampled observations. We construct two different estimators for the drift parameter based on the approximation by random walks of the driving noise and prove the consistency of these two estimators. Some numerical simulations are also presented to illustrate the performance of these maximum likelihood estimators. Empirical examples are given to illustrate the potential applications of these proposed estimators.
Dynamic relationship between Japanese Yen exchange rates and market anxiety: A new perspective based on MF-DCCA
2017, Physica A: Statistical Mechanics and its Applications
Citation 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.
This paper investigates the dynamic relationship between Japanese Yen exchange rates and market anxiety during the period from January 5, 1998 to April 18, 2016. A quantitative technique of multifractal detrended cross-correlation analysis (MF-DCCA) is used to explore the multifractal features of the cross-correlations between USD/JPY, AUD/JPY exchange rates and the market anxiety gauge VIX. The investigation shows that the causal relationship between Japanese Yen exchange rates and VIX are bidirectional in general, and the cross-correlations between the two sets of time series are multifractal. Strong evidence suggests that the cross-correlation exponents tend to exhibit different volatility patterns in response to diverse external shocks such as financial distress and widening in interest rate spread, suggesting that the cross-correlated behavior between Japanese Yen exchange rates and VIX are susceptible to economic uncertainties and risks. In addition, the performances of two market anxiety gauges, the VIX and the TED spread, are compared and the sources of multifractality are also traced. Thus, this paper contributes to the literature by shedding light on the unique driving forces of the Yen exchange rate fluctuations in the international foreign exchange market.
Arbitrage with fractional Gaussian processes
2017, Physica A: Statistical Mechanics and its Applications
Citation 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.
While the arbitrage opportunity in the Black–Scholes model driven by fractional Brownian motion has a long history, the arbitrage strategy in the Black–Scholes model driven by general fractional Gaussian processes is in its infancy. The development of stochastic calculus with respect to fractional Gaussian processes allowed us to study such models. In this paper, following the idea of Shiryaev (1998), an arbitrage strategy is constructed for the Black–Scholes model driven by fractional Gaussian processes, when the stochastic integral is interpreted in the Riemann–Stieltjes sense. Arbitrage opportunities in some fractional Gaussian processes, including fractional Brownian motion, sub-fractional Brownian motion, bi-fractional Brownian motion, weighted-fractional Brownian motion and tempered fractional Brownian motion, are also investigated.
Greek Government-Debt Crisis Events and European Financial Markets: Surprises of Greek Bond Yields and Inter-Relations of European Financial Markets
2019, SSRN
Fractality and multifractality analysis of news sentiments time series
2018, IAENG International Journal of Applied Mathematics
Graph-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)

View all citing articles on Scopus

View full text

Long-range correlation analysis of economic news flow intensity

Highlights

Abstract

Introduction

Section snippets

News analytics data

Types of auto-correlation

Empirical results

Conclusion

Acknowledgment

Physica A

Physica A

Physica A

ITBM-RBM

Auton. Neurosci.-Basic Clin.

Physica A

Physica A

Physica A

Physica A

Physica A

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