Reframing Diversification as a Network Problem: Graph Neural Networks for Portfolio Management
提出将多元化分析视为网络问题，利用图神经网络从市场连接中学习，使依赖关系、隐藏集中度和冲击传导更透明，帮助投资经理实现更可重复、可扩展的分散化分析。
Diversification and portfolio analysis are traditionally framed through covariance, correlation, and factor exposures. While foundational, these tools summarize co-movement without revealing the structure through which information and risk propagate, particularly during periods of market stress. This article reframes diversification as a network problem. By viewing the market as an interconnected system of assets, network-based representations make dependence, hidden concentration, and shock transmission more transparent. Building on this intuition, the authors focus on Graph Neural Networks (GNNs) as a practical framework for learning from market connectivity. Through intuitive examples and portfolio-relevant applications, the article shows how GNNs formalize reasoning that portfolio managers already apply informally, making diversification analysis more transparent, repeatable, and scalable in modern asset management.