Uncertainty of financial time series based on discrete fractional cumulative residual entropy

Cumulative residual entropy (CRE) is a measure of uncertainty and departs from other entropy in that it is established on cumulative residual distribution function instead of density function. In this paper, we prove some important properties of discrete CRE and propose fractional multiscale cumulative residual entropy (FMCRE) as a function of fractional order α, which combines CRE with fractional calculus, probability of permutation ordinal patterns, and multiscale to overcome the limitation of CRE. After adding amplitude information through weighted permutation ordinal patterns, we get fractional weighted multiscale cumulative residual entropy (FWMCRE). FMCRE and FWMCRE extend CRE into a continuous family and can be used in more situations with a suitable parameter. Moreover, they can capture long-range phenomena more clearly and have higher sensitivity to the signal evolution. Results from simulated data verify that FMCRE and FWMCRE can identify time series accurately and have immunity to noise. We confirm that the length of time series has little effect on the accuracy of distinguishing data, and even short series can get results exactly. Finally, we apply FMCRE and FWMCRE on stock data and confirm that they can be used as metrics to measure uncertainty of the system as well as distinguishing signals. FWMCRE can also track changes in stock markets and whether adding amplitude information must be decided by the characteristics of data.