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

2025-11-30

AI Summary

In high-risk domains such as corporate bankruptcy prediction, causal modeling suffers from structural instability, temporal inconsistency, and insufficient interpretability. Method: This paper proposes an agent-based, AI-driven iterative causal discovery framework that synergistically integrates large language models’ symbolic reasoning with statistical causal tests—including conditional independence testing and intervention-effect evaluation. It generates an initial directed acyclic graph (DAG) via constraint-guided prompting and dynamically refines the graph structure using causal validity feedback, ensuring temporal consistency and intervention readiness. Results: Evaluated on real-world bankruptcy data, our method significantly outperforms state-of-the-art baselines (NOTEARS, GOLEM, DirectLiNGAM) in causal graph reliability, structural stability, and interpretability. It establishes a novel paradigm for autonomous, trustworthy causal modeling in high-stakes decision-making contexts.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search

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.

Problem

• Develops an agentic AI framework for causal discovery
• Constructs valid, temporally coherent causal structures
• Produces reliable, interpretable models for corporate bankruptcy analysis

Innovation

• Agentic AI integrates LLM reasoning with statistical diagnostics
• Iteratively refines DAGs via constraint-guided prompting and feedback
• Produces explainable, intervention-ready causal graphs for bankruptcy analysis