Cross-market Stock Asset Classification Structure Based on Complex Network Clustering

This study aims to uncover the latent topological structure of stock markets through complex network clustering. This framework constructs a correlation network based on the daily returns of international equities, transforming asset co-movements into a force-directed network layout. A rigorous benchmark is conducted across five community detection algorithms. Louvain, Girvan-Newman, Walktrap, Label Propagation, and Spectral Clustering to identify persistent asset clusters that transcend conventional industry classifications. By enforcing a uniform cluster cardinality across all models, we uncover a relatively stable correlation in stock relationships that differs from traditional categorizations based on industry. Quantitative topological metrics reveal that the detected clusters exhibit strong intra-community connectivity while maintaining intercluster bridges, suggesting the presence of cross-community investment opportunities. The findings offer new insights for enhancing portfolio diversification and assessing systemic risk