Autocorrelation type, timescale and statistical property in financial time series

Earlier studies have documented that three types of autocorrelations exist in financial time series: sign, volatility, and return autocorrelation. In this paper, we examine how each type of the above autocorrelations affects the statistical properties of financial time series and its role in maintaining such statistical properties. Using three different shuffling series that correspondingly destroy each type of autocorrelation upon different timescales, we find that: (1) the statistical properties of the shuffling series significantly vary from the original ones; (2) volatility and return autocorrelations show greater impacts than sign autocorrelation; (3) the effects on the statistical properties are intensified as time scale expands; (4) the nonlinear component of autocorrelation is the major drive of the effect.