Community detection and clustering characteristics analysis of the stock market

In this paper, we study and analyze the community detection algorithms and the related evaluation metrics to explore the clustering characteristics of China's stock market. We build the stock market network by the constituent companies of the CSI 300 index and study the tightness between nodes. We first choose an appropriate adjacency matrix, which is obtained by the relation coefficient. Then, based on the singular value decomposition (SVD) of the adjacency matrix, we analyze the optimal network based on the directed spectral clustering on the ratio of eigenvectors and the spectral clustering followed by local refinement. The optimal network is chosen based on the best clustering effect according to the clustering effectiveness indexes Calinski–Harabasz and Davies–Bouldin. We explore the community structure in the stock market and study the influence of each community. Through the index out‐degree and in‐degree, we also obtain the most important companies in the market. Different from the traditional strategies that are always based on qualitative methods and undirected networks, this paper explores the characteristics of how to detect China's stock market. Specifically, we offer suitable strategies and measurements to evaluate the clustering results objectively and fairly. The qualities of community detection are measured too. Based on the above analysis, we further provide valuable insights by exploring the characteristics of the stock market.