代理与信号传递视角下股权质押违约多模态风险识别体系构建

Establishing a Multi-Modal Risk Identification System for Equity Pledge Defaults from the Perspectives of Agency Theory and Signal Transmission

摘要: 股权质押风险的跨市场传染已成为威胁资本市场稳定的重大隐患，精准识别其关键驱动因子是防范系统性风险的重要前提。本文提出了一套融合结构化数据与非结构化文本信息的多维风险因子识别分析框架，为后续引入机器学习模型与Shapley值归因分解提供理论支撑与变量体系基础。研究从质押行为、财务经营、市场治理三大维度构建结构化指标，并引入年报MD&A文本的情感语调与语义风险因子两类非结构化指标，由此形成涵盖五大维度的理论指标体系，旨在系统揭示股权质押违约风险的潜在驱动因素。本文的边际贡献在于，将研究视角从单一财务维度拓展至结构化与非结构化的信息融合，丰富了股权质押风险识别的信息边界；同时，提出将Shapley值引入可解释的归因推断，为金融机构贷后管理和监管部门差异化监管提供量化依据。

Abstract: Cross-market contagion of equity pledge risks has become a major hidden threat to the stability of capital markets; accurately identifying its key drivers is a crucial prerequisite for preventing systemic risks. This paper proposes a multi-dimensional risk factor identification and analysis framework that integrates structured data with unstructured textual information, providing theoretical support and a foundation for the variable system required for the subsequent introduction of machine learning models and Shapley value attribution decomposition. The study constructs a set of structured indicators across three dimensions—pledging behaviour, financial performance, and market governance—and further introduces sentiment and semantic risk factors derived from the Management Discussion and Analysis (MD&A) sections of annual reports. Together, these components form a comprehensive theoretical indicator system spanning five dimensions in total, aimed at systematically revealing the potential drivers of equity pledge default risk. The marginal contribution of this paper lies in expanding the research perspective from a single financial dimension to the integration of structured and unstructured information, thereby broadening the informational scope for identifying equity pledge risks. Simultaneously, we propose incorporating Shapley values into interpretable attribution inference, which provides a quantitative basis for post-lending management by financial institutions and differentiated supervision by regulatory authorities.