Firm-Level Supply Chain Risk Measurement and Its Determinants: A Generative Large Language Model-Based Analysis

Abstract

In the context of economic globalization and increasing uncertainty, supply chain risk has emerged as one of the major challenges for enterprises. Recent events such as geopolitical conflicts, the US-China trade war, and the COVID-19 pandemic have created significant disruptions that threaten both firm performance and broader economic stability. In response to these challenges, the Chinese government has introduced a series of policies to strengthen supply chain resilience and security. Therefore, accurately measuring and effectively managing supply chain risk is a major concern for both corporate managers and policymakers.

Despite its practical importance, measuring supply chain risk remains challenging. Existing methods either rely on theoretical models with restrictive assumptions, employ subjective survey data, or use indirect proxy variables that capture only specific risk dimensions, all of which have significant limitations. Additionally, while existing literature has explored supply chain risk management strategies, most studies are theoretical or case-based and lack empirical evidence that can provide practical guidance for firms.

To address these gaps, this study develops and validates a method for measuring firm-level supply chain risk using a generative large language model (LLM). Using Chinese A-share companies listed on the Shenzhen Stock Exchange from 2014 to 2023, I apply an LLM to identify and quantify supply chain risk discussions from corporate site visit transcripts, thereby measuring firm-level supply chain risk exposure. A series of validation tests confirm the reliability and validity of the LLM-based measure. The results show that the measure varies intuitively over time and across industries. The measure is closely correlated with proxies theoretically related to supply chain risk, including inventory levels, supply chain concentration, and purchasing managers’ indices. Based on a human-annotated benchmark dataset, I find that the LLM-based approach significantly outperforms dictionary-based methods in both risk identification and risk quantification.

Using this firm-level supply chain risk measure, I examine the impact of digital innovation, vertical alliances, and corporate culture on supply chain risk exposure and investigate the underlying mechanisms.

First, digital innovation represents the technological infrastructure for risk management. Using patent-based measures of digital innovation, I find that firms with more digital innovation exhibit significantly lower supply chain risk exposure. These results are robust to a series of robustness checks. Moreover, the mitigating effect of digital innovation is particularly pronounced in firms with greater geographical supply chain distances, extensive overseas operations, short-term relationships with partners, and those in the manufacturing sector. Further analysis demonstrates that digital innovation enhances information sharing and improves operational efficiency, serving as potential mechanisms for supply chain risk reduction.

Second, corporate culture serves as the internal “soft power” that shapes firms’ risk preferences and behaviors. Using textual analysis based on the competing values framework, I quantify corporate culture and find that it negatively affects supply chain risk exposure. This negative relationship remains robust across robustness checks. This effect varies by cultural type, as flexibility-oriented cultures, namely clan and adhocracy, are associated with lower supply chain risk, while control-oriented cultures, such as hierarchy and market, show no significant effect. Mechanism tests confirm that corporate culture operates through two channels. First, culture serves as an informal governance substitute, with stronger effects in firms with weaker formal governance. Second, culture improves external adaptation, providing greater benefits during periods of high economic policy uncertainty or industry volatility.

Third, vertical alliances create coordination networks between firms and their supply chain partners. Using a large language model to identify and characterize alliance agreements, I document that firms engaging in vertical alliances experience significantly lower supply chain risk exposure. The main findings are robust to a series of robustness checks. Further analysis shows that alliance design characteristics matter, as alliances featuring more partners, broader cooperation scope, formal equity structures, and explicit governance mechanisms are associated with greater risk mitigation effects. The mechanism analysis reveals that vertical alliances work through two channels: reducing transaction costs by limiting opportunistic behavior and promoting resource sharing between partners.

This study makes three main contributions. First, the study develops and validates an LLM-based approach to measure firm-level supply chain risk that overcomes limitations of existing methods. This methodological innovation facilitates empirical research on the determinants of supply chain risk and provides a framework for measuring other complex constructs in finance and accounting.

Second, the empirical results extend the literature on supply chain risk management. Although various management methods and conceptual frameworks have been proposed, they have not been empirically validated. By examining digital innovation, vertical alliances, and corporate culture, this research provides empirical evidence of their effects and underlying mechanisms, thereby advancing the theoretical understanding of supply chain risk management.

Third, the findings of this study offer important practical implications. For corporate managers, the results provide practical guidance for developing supply chain risk management strategies. Companies can strengthen their supply chains by investing in digital technologies, forming strategic partnerships with suppliers and customers, and developing flexibility-oriented corporate cultures. For policymakers, this study provides firm-level empirical evidence to guide policy decisions. The risk measurement method developed here also helps governments identify vulnerable areas and high-risk companies in supply chains, allowing for more targeted policies and support programs. Additionally, the findings can help regulatory agencies improve early warning systems and assist financial institutions in making better lending decisions, contributing to a more comprehensive approach to supply chain risk management.

Date of Award	4 Dec 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Honghai YU (External Supervisor) & Teng Stephen SUN (Supervisor)