Volume 2,Issue 7
基于代价敏感XGBoost 模型的企业ESG 评级预测
针对企业环境、社会与治理（ESG）评级分布不平衡导致的预测偏误，以及现有模型对不同行业ESG 行为解释力不足的问题，本文提出一种基于代价敏感极限梯度提升（XGBoost）与沙普利加法解释（SHAP）框架结合的改进预测模型。我们引入代价敏感学习策略，并结合合成少数类过采样技术（SMOTE），强化模型对少数类等级样本的捕获能力。同时量化财务指标的边际增益并利用SHAP 算法剖析行业异质性。研究结果表明代价敏感优化显著提升了极端评级的预测精度，“落后者”召回率91.72%，“领导者”召回率92.24%；财务基本面和行业信息的引入有效修正了历史评级的依赖性，宏平均受试者工作特征曲线下面积（Macro AUC）高达97.81%；驱动机制存在行业分化，重污染行业呈现出“资产密集型合规”逻辑，而非重污染行业则展现出“效率溢价与市场敏感”特征。本文为投资者精准锚定优质ESG 标的、排查尾部风险提供了高效的决策工具。
[1]JENSEN M C， MECKLING W H. Theory of the firm: managerial behavior， agency costs and ownership structure[J]. Journal of Financial Economics， 1976， 3(4): 305-360.
[2]D’AMATO V， D’ECCLESIA R， LEVANTESI S. ESG score prediction through random forest algorithm[J]. Computational Management Science， 2022， 19(2): 347-373.
[3]VELTE P. Does ESG performance have an impact on financial performance? Evidence from Germany[J]. Journal of Global Responsibility， 2017， 8(2): 169-178.
[4]SANDBERG H， ALNOOR A， TIBERIUS V. Environmental， social， and governance ratings and financial performance: evidence from the European food industry[J].
Business Strategy and the Environment， 2022， 32(4): 2471-2489.
[5] 史瑜港. 融入ESG 信息的制造业上市公司信用风险预警研究[D]. 兰州: 西北师范大学，2025.
[6] 袁蓉丽，江纳，刘梦瑶.ESG 研究综述与展望[J]. 财会月刊，2022(17):128-134.
[7] 梁四安， 李淑萍， 段军山， 等.ESG 表现如何影响企业绿色创新？—— 基于融资约束与分析师关注的中介效应[J/OL].(2025-12-26)[2026-01-21].https://link.cnki.net/urlid/61.1294.N.20251226.1133.008.
[8] 尚炀川，蒋建洪.ESG 评级对企业高质量发展的影响研究[J]. 商业观察，2025，11(34):106-111.
[9]DE LUCIA C， PAZIENZA P， BARTLETT M. Does good ESG lead to better financial performances by firms? Machine learning and logistic regression models of public enterprises in Europe [J]. Sustainability， 2020， 12(13): 5317.
[10]GARCÍA F， GONZÁLEZ-BUENO J， GUIJARRO F， et al. Forecasting the environmental， social， and governance rating of firms by using corporate financial performance variables: a rough set approach[J]. Sustainability， 2020， 12(8): 3324.
[11]D’AMATO V， D’ECCLESIA R， LEVANTESI S. Fundamental ratios as predictors of ESG scores: a machine learning approach[J]. Decisions in Economics and Finance， 2021， 44: 1087-1110.
[12]CINI F， FERRARI A. Towards the estimation of ESG ratings: a machine learning approach using balance sheet ratios [J]. Research in International Business and Finance， 2024， 73: 102653.
[13]Kartal M T， Taşkın D， Shahbaz M， Depren S K， Pata U K. Effects of environment， social， and governance (ESG) disclosures on ESG scores: investigating the role of corporate governance for publicly traded Turkish companies[J]. Journal of Environmental Management， 2024， 368: 122205.
[14]TASKIN D， SARIYER G， ACAR E， et al. Do past ESG scores efficiently predict future ESG performance?[J]. Research in International Business and Finance， 2025， 74: 102706.
[15]ZHANG J F， ZHAO Z X. Corporate ESG rating prediction based on XGBoost-SHAP interpretable machine learning model[J]. Expert Systems with Applications，2026， 295: 128809.
[16]CHEN T， GUESTRIN C. XGBoost: a scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco: ACM， 2016: 785-794.
[17] 张影. 基于代价敏感的XGBoost 算法在T2DM 患者并发糖尿病肾病预测中的应用[D]. 湘潭: 湘潭大学，2024.
[18] 李爱华，刘婉昕，陈思帆，等. 面向不平衡数据的SMOTE-BO-XGBoost 集成信用评分模型研究[J/OL].(2024-10-11)[2025-11-20].https://doi.org/10.16381/j.cnki.issn1003-207x.2023.0635.
[19]BEN YAN， ZIQI Y，YAN W. The measurement and early warning of daily financial stability index based on XGBoost and SHAP: Evidence from China[J]. Expert Systems with Applications， 2023，227: 120245.
[20]HAND D J， TILL R J. A simple generalisation of the area under the ROC curve for multiple class classification problems[J]. Machine Learning， 2001， 45(2): 171-186.
[21]MEHDARY A，CHEHRI A，JAKIMI A， et al. Hyperparameter optimization with genetic algorithms and XGBoost: A step forward in smart grid fraud detection[J]. Sensors，2024， 24(4): 1121.