Credit Network Modeling and Analysis via Large Language Models

TL;DR

The paper addresses the challenge of translating unstructured financial statements into a structured credit-network representation and using large language models to reason about network interventions. It develops a framework that encodes each firm as a local credit network with a liability matrix $\\mathbf{L}$ and external assets $\\mathbf{e}$, aggregates these into system-wide networks, and computes maximal clearing payments $\\mathbf{P}$ to obtain per-firm assets $a_i(\\mathbf{P})$ while aiming to maximize $\\sum_i a_i$. The key contributions are a scalable translation pipeline with automatic inconsistency alerts and a framework for LLM-guided execution of portfolio compression and debt removal, validated across diverse topologies and real-world data such as German banks. Findings show that LLMs generate coherent reasoning and effective executions, outperforming baselines and scaling to over $2{,}000$ firms, suggesting practical potential for AI-assisted financial-system analysis and stabilization. The work opens avenues for multi-operation optimization and development of specialized LLMs for financial network analysis.

Abstract

We investigate the application of large language models (LLMs) to construct credit networks from firms' textual financial statements and to analyze the resulting network structures. We start with using LLMs to translate each firm's financial statement into a credit network that pertains solely to that firm. These networks are then aggregated to form a comprehensive credit network representing the whole financial system. During this process, the inconsistencies in financial statements are automatically detected and human intervention is involved. We demonstrate that this translation process is effective across financial statements corresponding to credit networks with diverse topological structures. We further investigate the reasoning capabilities of LLMs in analyzing credit networks and determining optimal strategies for executing financial operations to maximize network performance measured by the total assets of firms, which is an inherently combinatorial optimization challenge. To demonstrate this capability, we focus on two financial operations: portfolio compression and debt removal, applying them to both synthetic and real-world datasets. Our findings show that LLMs can generate coherent reasoning and recommend effective executions of these operations to enhance overall network performance.

Figures (5)

• Figure 1: An example of a credit network. The credit network consists of three firms $v_i$, $v_j$, and $v_k$ with external assets of 6, 2, and 3, respectively. Firm $v_i$ owes 5 to $v_j$, and firm $v_j$ owes 5 to $v_k$. Liability matrix $\mathbf{L}$ and external assets $\mathbf{e}$ are shown on the right-hand side.
• Figure 2: An example for portfolio compression. Left: An initial credit network. Right: An updated credit network after portfolio compression with $\mu^c=2$.
• Figure 3: An example for debt removal. Left: An initial credit network. Right: An updated credit network after removing debt $l_{ij}$.
• Figure 4: Workflow for translating financial statements to credit networks.
• Figure 5: Inputs and outputs of LLMs for identifying effective execution strategies.