Abstract
Abstract
The proliferation of artificial intelligence (AI) applications introduces emergent AI-associated risks as novel and potent vectors for systemic contagion in inter-connected capital markets. Traditional complex network models, however, are often static and fail to capture the non-linear, multi-path coupling effects and the dynamic evolutionary trajectory of risk propagation. To address these gaps, this study develops a hybrid static-dynamic analytical framework. First, we conduct a static analysis to identify systemically important nodes by integrating multi-dimensional centrality metrics and employ the Leiden algorithm to partition the network into distinct risk communities. The core innovation lies in our dynamic analysis. We introduce a non-linear, multi-path risk coupling model to quantify the compounded influence intensity when risk propagates simultaneously through multiple channels. Building on this, we design a multi-stage, multi-level dynamic risk transmission algorithm that simulates the entire cascade, from initial infection and three-degree diffusion to the eventual stabilization of network-wide risk. Empirical analysis on a network of 2,052 Chinese listed companies demonstrates the framework’s effectiveness. This research provides a novel, quantifiable methodology for tracing risk from local shocks to systemic cascades, offering valuable tools for financial regulators.