Scale Invariance of News Flow Intensity Time Series

Abstract

In this paper we examine the presence of self-similarity in flow intensity of economic and financial news taken from a nine-month period of 2015. Since there is a close relationship between long range dependent and self-similar processes, we use two methods – the detrended fluctuation analysis (DFA) and the averaged wavelet coefficient (AWC) method – to estimate both the long range correlation and the self-similarity exponent (the Hurst exponent), respectively. Empirical results obtained by this methods show that time series of news intensity exhibit self-similarity (as well as a long memory property). The Hurst exponent (as well as the long-range correlation exponent) is greater than 0.5 over three orders of magnitude in time ranging from several minutes to dozen of days. Estimates of the Hurst exponent obtained by AWC are very close to the estimates of the long-range correlation exponent obtained by DFA for almost all analyzed time series. By using decouple scales and multi-scale approaches, the DFA and the AWC methods allowed us to reveal a strong scaling behavior as well as to detect a distinct crossover effect.