A Network Perspective Of Comovement And Structural Change:evidence From The Chinese Stock Market

In the process of economic globalization,it has reached a broad consensus that the modern financial markets present complex system characteristics along with the electronization and informatization of trading technologies.An appropriate characterization of the comovement can provide us some light for deeply and clearly understanding the risk management and investor trading patterns in the stock market.This paper applies the complex network approach to deal with this issue for the Chinese stock market.Weselect the daily closing prices of the CSI 300 from January 2006 to December 2019 as the research sample.Firstly,in view of the topological properties,we investigate the time-varying comovement between individual stocks by constructing 14 directed weighted stock networks.Furthermore,the weighted LeaderRank algorithm is employed to describe the comovement structure of the entire market.Most importantly,from the perspective of fundamental factors and industry factors,we reveal the driving factors of the comovement and structural change of the entire market.The full text is divided into five chapters which are arranged as follows:In chapter one,the background and significance of this study are firstly expounded.Then,the existing works of domestic and foreign scholars using complex network theory to analyze the relevant projects are summarized.Finally,the main work and specific framework of this paper are introduced.In chapter two,we introduce the theoretical knowledge of graph theory and the definition of related network topology indicators,including network density,average clustering coefficient,global efficiency,node degree and degree distribution.The definitions of network centrality indicators are further presented,including degree centrality,vertex strength,betweenness centrality,closeness centrality,eigenvector centrality,PageRank algorithm and weighted LeaderRank algorithm.Finally,the statistical test methods involved in this paper are explained,including Kolmogorov-Smirnov(K-S)goodness-of-fit test and log-likelihood ratio test.The third chapter investigates the comovement of the CSI 300 individual stocks.First,we select the optimal parameter by comparing four special significance levels,with the optimal significant level λ=0.05 settled down.Secondly,14 dynamic CSI 300 constituent networks are constructed based on the cointegration relationship by employing the method of complex network and Engel-Granger test.Thirdly,the time-varying changes of the topological structure of the CSI 300 stock networks are described by exploring the changes of topological properties,including network density,average clustering coefficient,global efficiency,and node degree and degree distribution.The empirical results show that:(1)Compared with the normal period,the number of edges in the crisis period increases significantly,which means that the connection between nodes is more closely,and the comovement of the network is greatly enhanced;(2)During the financial crisis,the high comovement between individual stocks can increase the transmission channel and efficiency of risk diffusion in accordance with the real change of the Chinese stock market;(3)The results of the degree distribution prove the heterogeneity of constructed E-G networks.The fourth chapter analyzes the comovement structure of CSI 300 stock markets.Considering that different centrality indicators have different scopes of application,the scenarios are not completely consistent,and the selection of specific indicators should also be judged according to the research data.This paper firstly compares the correlation between different centrality indicators and financial characteristics,and figures out the most suitable centrality indicator for the data in this paper,that is,the weighted LeaderRank(WLR)algorithm.Then,the effectiveness of the weighted Leader Rank algorithm for the comovement structure is verified by the Q-Q(quantile-quantile)graph.Finally,we analyze the characteristics of the top 15(10%of 154)stocks obtained by the weighted LeaderRank algorithm.The empirical results show that:(1)According to the results of the correlation analysis and Q-Q graph,the weighted LeaderRank algorithm has more balanced and superior performance,which is considered to be the most suitable centrality indicator for this paper;(2)According to the results of the WLR score ranking,stocks with higher weighted LeaderRank algorithm scores generally have more long-term investment value;(3)Among the top 15 stocks,the average market value of 5 stocks is within 20%of the 154 stocks,and companies with larger market capitalization are usually listed in the stock market,which corresponds to the concept of" too big to fail" in some financial institutions;(4)Among the top 15 stocks,the average capitalization of 3 stocks rank between 80 and 90,which corresponds to the concept of“too connected to fail”.The fifth chapter further explores the driving factors of the comovement of the CSI 300 individual stocks and the comovemnt structure of the whole market.In the research on the driving factors of individual stocks’ comovement,this chapter selects the dummy variables that describe the cointegration relationship of paired stocks as the explained variables,and selects the Roe,To,BMratio,Mvalue,Beta,Alpha,Growth,Em as explanatory variables.To ensure the matching between the explanatory variables and the explained variables,we define the fundamental factors of the stock pair as explanatory variables.Meanwhile with the help of the SWS first-level industry classification standard,INDUSTRYi→j,t describing whether the paired stocks belong to the same industry is introduced as an explanatory variable.In the analysis of the driving factors of the comovement structure of the whole market,we select the weighted LeaderRank algorithm score as the explained variable,and still select 8 financial indicators that reflect the characteristics of different aspects of the stock as the fundamental factors.In the selection of industry factors,we define 24 dummy variables as the reflection of 25 sub-industries,and constructed a regression equation to study the driving factors of market comovement structure.Our results are verified to be robustness.The empirical results show that:(1)Investor preference will change in different periods.In normal times,investors pay more attention to stock returns,and only in times of crisis,will the risk be efficiently concerned;(2)In normal periods,we only find that the agriculture,forestry,animal husbandry&fishery and composite have significant influence on the comovement structure of the entire market.Besides,public utilities and medias also have a significant impact during the crisis.