Identifying Cross-Market Risk Contagion Amplifiers via Graph Attention Networks: Empirical Evidence from U.S. Financial Stress Periods

Abstract

The propagation of financial distress across heterogeneous asset markets has become one of the most pressing challenges for systemic risk monitoring in the post-crisis regulatory environment. This paper investigates cross-market risk contagion among U.S. equity, bond, and credit derivative markets through a graph attention network (GAT) framework embedded in a multilayer network structure. Constructing a directed, attention-weighted graph with 30 nodes spanning three market layers, and drawing on daily data from January 2018 to December 2023, the analysis identifies a persistent set of risk amplifier nodes whose elevated outgoing attention scores and centrality measures designate them as the primary drivers of cross-market stress transmission. The high-yield CDS index node emerges as the most consistently identified amplifier across three distinct stress episodes—the March 2020 pandemic shock, the 2022 interest rate stress cycle, and the March 2023 regional banking turmoil—while the financial sector equity index and long-duration bond node occupy episode-specific secondary amplifier roles. The multilayer network representation captures contagion pathways that single-layer analyses systematically underestimate, particularly the derivative-to-equity transmission channel active under tail-risk conditions. These findings carry direct implications for counterparty risk monitoring and macro-prudential early-warning design in the context of central clearing and financial stability regulation.