The Research Of Dynamic Stock Network Based On The Overlapping Community Structure

Community Structure, which is one feature, exists in the social network, neutral network, www and communication network. The study of community structure detection in complex networks is meaningful in topology structure analyzing, function acknowledging and dynamical behavior forecasting etc. Therefore, community structure detection attracts a lot of scholars’interest.The community detection includes overlapping detection and non-overlapping detection. There are sorts of algorithms of overlapping community structure detection have the capability of dividing the medium and small networks efficiently. However, there are some special nodes in the network and links to each other. With the increasing number of the special structure, the visibility of nodes and edges will decrease. As a result, many algorithms are not suitable for this kind of network.The algorithm of overlapping community structure detection based on the complete sub-graph has been raised to solve the problem mentioned above. S&P500and CSI300, which have community structure, are selected as the object of this study. By using the theory of overlapping networks, the characteristic of the security market has been studied, for instance, topological structure of network, clustering features, overlapping nodes and so on.The main contributions of the paper are summarized as follows:1. During processing the community detection, the steps are listed as follows:(1) Find cliques as the core of communities.(2)Merge cliques.(3)Merge non-clique nodes. The algorithm during merging the cliques has been improved in this paper. Thus, we use the overlapping coefficient to merge cliques, and then use the connecting degree to continue the process. This paper presents four methods in dividing non-clique notes for comparison purpose.2. Through detecting the communities of CSI300and S&P500prior to and during the financial crisis, the overlapping nodes have been found. Comparing the emerging market (CSI300) with the mature market(S&P500) under the financial strike, the similarities and the differences of the two markets have been detected. The similarity of the two markets is that the numbers of communities under the financial crisis were declined markedly. It proves that as long as the globalization, the clustering structure of industries is not obvious any more. The shock of financial crisis destructs the community structure of American market. On the contrary, Chinese market still has some community structures. It reflects the different reaction of the two markets under the financial crisis. It is found that the ocean shipping and national coal consumption industry in CSI300have been affected seriously under the financial crisis. However, the performance of the medical industries in S&P500is stable under the financial crisis.3. The daily closing price of CSI300from2005.1to2010.12is investigated. Through analyzing the time evolution of community structure and the annual change of overlapping nodes, the tendency of community structure has been detected. The community structure became from clear to obscure and recover clear at last. There was no overlapping node in2005. However, the number of overlapping nodes was rising in the following three years, and then it reached a peak in2008. There is a reduction tendency afterwards.In this study by the National Natural Science Foundation support, the project name "opening space in a weighted network evolution and optimization of topology design", item number:11075057.