Analysis of rhythmic patterns of the time series based on the statistical model of NBD

Abstract

The paper presents two methods for parameterizing quasi-periodic cycles in price returns time series (we shall call such patters, rhythmic structures). Both methods involve categorizing time series by discrete scaling of returns. The first method is based on the property of the negative binomial distribution which describes occurrence of named entities in natural texts (used in corpus linguistics and NLP). The classification problem is solved according to the parameters of the most frequently used words. The second method is based on reducing the dimensionality of incidence matrices of text sequences to a dictionary using neural networks. In both cases, it is possible to show the difference in the rhythmic structure of text financial sequences related to different industries