ARCADIA: Scalable Causal Discovery for Corporate Bankruptcy Analysis Using Agentic AI

TL;DR

ARCADIA tackles the need for causal understanding in corporate bankruptcy risk by marrying agentic LLM-based hypothesis generation with rigorous statistical diagnostics to build temporally coherent DAGs. The framework iteratively proposes and evaluates causal graphs, incorporating domain knowledge and identifiability constraints to avoid spurious causal inferences. Empirical results on Italian firms show ARCADIA achieving near-perfect valid-DAG rates and stable, interpretable graphs compared with traditional baselines, with an intervention-ready pipeline. The work demonstrates how agentic AI can augment causal inference in high-stakes finance and points to broader applications in economic modeling and policy analysis.

Abstract

This paper introduces ARCADIA, an agentic AI framework for causal discovery that integrates large-language-model reasoning with statistical diagnostics to construct valid, temporally coherent causal structures. Unlike traditional algorithms, ARCADIA iteratively refines candidate DAGs through constraint-guided prompting and causal-validity feedback, leading to stable and interpretable models for real-world high-stakes domains. Experiments on corporate bankruptcy data show that ARCADIA produces more reliable causal graphs than NOTEARS, GOLEM, and DirectLiNGAM while offering a fully explainable, intervention-ready pipeline. The framework advances AI by demonstrating how agentic LLMs can participate in autonomous scientific modeling and structured causal inference.

Figures (1)

• Figure 1: Runtime comparison across feature budgets $M$.