Evolution of Major Shareholder Networks in China’s Shanghai A-Share Market: A Dynamic Network Analysis

This paper investigates the dynamic evolution of major shareholder networks in Chinese listed companies from a complex network perspective. The shareholder–stock system is analyzed at the network level because ownership links are interdependent and may reveal collective structures that are not visible from firm-level statistics alone. Based on quarterly data on the top ten shareholders of Shanghai A-share companies covered by the historical dataset from 2011 to 2024, dynamic “shareholder–stock” bipartite networks are first constructed. Statistically validated projected shareholder networks are then derived using the Statistically Validated Network (SVN) method. Basic topological metrics, degree-distribution heterogeneity, community structures, and node-importance measures are analyzed to characterize the temporal evolution of network structure. Over the study period, expansion of the shareholder set was accompanied by declining density and average degree in the non-isolated subgraph, while its clustering coefficient increased, indicating increasing overall sparsity alongside persistent local interconnectedness. Relative to the surrounding periods, the 2020–2022 period exhibited a temporary structural reconfiguration, although the overall network scale remained relatively stable. This reconfiguration was characterized by a rebound in average degree, the expansion of large communities, and increased community-size heterogeneity. After 2023, density, average degree, and community concentration partially moved back toward their longer-term patterns, suggesting a self-organizing adjustment in the network structure. A quarterly multivariate common-break analysis within the Bai–Perron framework identified statistically supported structural transitions at 2020Q1 and 2022Q4 under the baseline threshold, providing quantitative support for this temporal organization. BIC selected a two-break specification at all four BH screening cutoffs. Bootstrap support was consistent across the examined block lengths at thresholds of 0.025, 0.05, and 0.10, whereas inference at 0.01 remained sensitive to block length. At the late-2020 peak in the number of isolated shareholders, the associated stocks spanned multiple industries, with modest differences in sectoral composition from the full-sample benchmark. Across the annual fourth-quarter networks, PageRank analysis of the non-isolated subgraphs further showed that highly ranked shareholders were concentrated predominantly in larger communities. These findings suggest that dynamic network analysis can effectively capture time-varying shareholder relationships and reveal structural adjustments in investment networks under changing market conditions.