Ardalankia Jamshid, Habibnia Ali, Ausloos Marcel, Jafari G Reza
Department of Economics, Virginia Tech, Blacksburg, Virginia, United States of America.
Dataism Laboratory for Quantitative Finance, Virginia Tech, Blacksburg, Virginia, United States of America.
PLoS One. 2025 Sep 3;20(9):e0330372. doi: 10.1371/journal.pone.0330372. eCollection 2025.
Interdependent networks structurally influence each other so that the source network imposes hidden community structures into the target network. We propose a mathematical model so that when introducing an interlayer similarity function we generalize hierarchical clustering approaches for multilayer networks. The proposed methodology shows how a "source" network influences the "target" network via structural spillovers that are hidden and are not detectable by conventional community detection methods. The methodology reveals evidence that hidden interlayer interactions consequently generate hidden links on the target network. These hidden links construct hidden community structures on the target network (imposed from the source network) that are distinct from the community structures of the solo target network (without the presence of the source network). This model applies to systems with hidden interlayer interactions, such as, e.g., covert criminal groups, inter-platform social network interactions, scientific research groups, and financial markets. Financial markets are well known for complicated endogenous and exogenous, but often hidden, not to say the least, asymmetric layer interactions. We implement our model on multilayer financial networks: in particular, we find that trading value logarithmic changes (source) impose hidden community structures on the price return network (target). The main finding is that adding another relevant layer, such as the trading value layer, adds more information to systemic behaviors throughout the price return network. Dismissing it may yield less systemic information and underestimation of systemic risk because the footprint of some structures on the target network originated from another layer and is not detectable by singling out the target layer. As an empirical application, we exploit the methodology to define another perspective on portfolio diversification.
相互依存的网络在结构上相互影响，使得源网络将隐藏的社区结构强加于目标网络。我们提出了一个数学模型，在引入层间相似性函数时，我们对多层网络的层次聚类方法进行了推广。所提出的方法展示了一个“源”网络如何通过传统社区检测方法无法检测到的隐藏结构溢出影响“目标”网络。该方法揭示了隐藏的层间相互作用会在目标网络上产生隐藏链接的证据。这些隐藏链接在目标网络上构建了（由源网络强加的）与单独的目标网络（不存在源网络时）的社区结构不同的隐藏社区结构。该模型适用于具有隐藏层间相互作用的系统，例如秘密犯罪集团、跨平台社交网络互动、科研团队和金融市场。金融市场以内生和外生的复杂但通常隐藏的层间相互作用而闻名，更不用说不对称层间相互作用了。我们在多层金融网络上实现了我们的模型：特别是，我们发现交易价值对数变化（源）在价格回报网络（目标）上强加了隐藏的社区结构。主要发现是，添加另一个相关层，如交易价值层，会为整个价格回报网络的系统行为添加更多信息。忽略它可能会产生较少的系统信息并低估系统风险，因为目标网络上某些结构的痕迹源自另一层，仅单独考虑目标层是无法检测到的。作为一个实证应用，我们利用该方法为投资组合多元化定义了另一种视角。
