Correlation dynamics in the cryptocurrency market based on dimensionality reduction analysis

Abstract

In this study, dimensionality reduction algorithms are applied to examine the correlation dynamics in the cryptocurrency market in which the time-dependent correlation matrix sequence is transformed into a distance matrix. The results based on multidimensional scaling (MDS) analysis show that information related to dynamic structural changes in the correlation coefficient matrix can be reconstructed well in low-dimensional Euclidean space. In particular, we found that the t-distributed stochastic neighbor embedding (t−SNE) algorithm can effectively exhibit the correlation dynamics in a two-dimensional (2D) space, thus providing a visualization method for analyzing dynamics in the market. Finally, we extract the clusters generated by the t−SNE algorithm using the k nearest neighbor (kNN) network, and study the difference in the yield distribution of the periods corresponding to different clusters. Based on a comparison with the CCI 30 index, we determined that the components in the kNN network correspond well to different states in the market. The results show that dramatic changes in the correlation matrix suggest that significant changes may occur in the distribution of yields, such as a decrease in the average return.