A1228
Title: Structural learning and causal interference in multiplex networks: A two-stage framework for financial distress
Authors: Tianhai Zu - University of Texas at San Antonio (United States) [presenting]
Abstract: Analyzing systemic risk in high-dimensional corporate networks presents a dual statistical challenge: first, estimating the unobserved dependency structure among thousands of firms; and second, identifying causal effects in the presence of network interference. A novel framework is proposed that bridges these two objectives to analyze corporate bankruptcy. The structural learning problem is first addressed by developing a multi-layered latent position model for multiplex networks, integrating supply chain and co-investment data into a unified low-rank embedding. This approach effectively reduces dimensionality while capturing complex transitivity and homophily, significantly enhancing predictive performance over baseline models. Building on this estimated topology, the inquiry is then extended to network causal inference. A pilot framework is explored that utilizes the learned latent structure to model spillover effects, aiming to disentangle the causal propagation of financial distress from mere structural correlation. By synergizing latent space modeling with causal identification strategies, a robust statistical pathway is offered for understanding how risk is structured and transmitted in complex economic systems.
