从黑盒到白盒：应用SHAP值对ESG-股价波动关联进行统计稳健性检验与经济学解释

From Black Box to White Box: Statistical Robustness Checks and Economic Interpretation of the ESG-Stock Price Volatility Nexus Using SHAP Values

摘要: 在ESG投资理念深度融入中国资本市场的背景下，现有研究多聚焦ESG的收益效应，对其风险防控功能，尤其是极端尾部风险的影响研究存在不足，且普遍依赖线性模型，存在设定偏误。本文以2018~2024年沪深300指数成分股为研究样本，构建Kur全矩振幅值综合捕捉股价常规波动与极端尾部风险，采用双向固定效应模型、工具变量法系统检验ESG表现对股价全矩波动的因果性影响，并通过XGBoost非线性模型与TreeSHAP方法实现“从黑盒到白盒”的无偏拆解。研究发现：ESG表现对股价全矩波动具有显著的抑制效应，公司治理维度是核心驱动因子；二者呈U型非线性关系，存在明确的作用拐点，该效应在国有企业、制造业企业与市场下行区间更为突出；非线性模型的拟合精度与泛化能力显著优于线性基准模型，白盒拆解明确了信息披露质量、董事会独立性等核心细分指标，验证了治理与环境维度的正向协同效应。本文拓展了ESG与资本市场稳定的相关研究，为监管层完善ESG制度建设、上市公司优化ESG治理、投资者构建风险防控体系提供了经验证据。

Abstract: Against the backdrop of the deep integration of ESG investment principles into China’s capital market, existing literature has largely focused on the return implications of ESG, while paying insufficient attention to its risk-mitigation role, especially its impact on extreme tail risk. Furthermore, most prior studies rely on linear models, which are subject to inherent model misspecification. This paper uses a sample of constituents of the CSI 300 Index from 2018 to 2024, and constructs the Kur full-moment amplitude indicator to comprehensively capture both conventional stock price volatility and extreme tail risk. We employ a two-way fixed effects (TWFE) model and instrumental variable (IV) approach to systematically examine the causal effect of ESG performance on stock price full-moment volatility, and implement an unbiased “from black box to white box” decomposition via the XGBoost nonlinear model and TreeSHAP method. The main findings are as follows: First, ESG performance exerts a significant inhibitory effect on stock price full-moment volatility, with the corporate governance (G) dimension serving as the core driving factor. Second, the relationship between ESG performance and stock price volatility presents a significant U-shaped nonlinear pattern with a clear inflection point, and this inhibitory effect is more pronounced for state-owned enterprises (SOEs), manufacturing firms, and bear market periods. Third, the nonlinear model significantly outperforms the linear benchmark model in terms of fitting accuracy and out-of-sample generalization ability. The white-box decomposition identifies core sub-indicators including information disclosure quality and board independence, and verifies a significant positive synergy between the governance and environmental dimensions. This study extends the literature on ESG and capital market stability, and provides empirical evidence for regulators to improve ESG institutional frameworks, listed firms to optimize ESG governance, and investors to establish robust risk management systems.