基于Stacking集成模型的制造业上市企业财务困境预测研究  

Research on Financial Distress Prediction of Listed Manufacturing Companies Based on Stacking Ensemble Models

作　　者：梅宇 MEI Yu(School of Management,Hefei University of Technology,Hefei 230000,China)

机构地区：[1]合肥工业大学管理学院,安徽合肥230000

出　　处：《商业观察》2025年第8期42-48,共7页BUSINESS OBSERVATION

摘　　要：近年来,随着中国制造业的快速发展,上市制造企业的财务困境风险日益加剧,尤其是在2018—2023年间,制造业在ST企业中的比例一直很高,凸显了财务危机的严峻性。为应对这一挑战,文章基于沪深A股1914家制造业上市公司的财务和非财务数据,应用Stacking集成学习模型进行财务困境预测。集成模型以决策树、随机森林、XGBoost以及CatBoost模型作为基学习器,逻辑回归模型作为元学习器,通过集成多种算法的优势,提升了预测的准确性和稳健性。实验结果显示,模型在测试集上的AUC值达到85.49%,显著优于各单一分类模型,验证了Stacking集成模型在财务困境预测中的有效性。研究将有助于企业利益相关者及时发现可能陷入财务困境的制造业企业,从而更有效应对潜在的财务风险。In recent years,the rapid development of China's manufacturing industry has intensified the financial distress risk for listed manufacturing companies.Notably,from 2018 to 2023,the proportion of ST companies within this sector has remained high,underscoring the severity of financial crises.To address this challenge,this study applies the Stacking ensemble learning model to predict financial distress using both financial and non-financial data from 1914 Shanghai and Shenzhen A-share listed manufacturing companies.The ensemble model utilizes Decision Trees,Random Forests,XGBoost,and CatBoost as base learners,with Logistic Regression serving as the meta-learner.By leveraging the strengths of multiple algorithms,the model enhances both prediction accuracy and robustness.Experimental results demonstrate that the model achieves an AUC of 85.49%on the test set,which significantly outperforms individual classification models,thus validating the effectiveness of the Stacking ensemble model in predicting financial distress.This research will assist stakeholders in manufacturing companies in promptly identifying those financial distress,so as to effectively respond to potential financial risks.

关 键 词：ST企业 财务困境 Stacking集成模型 

分 类 号：F275[经济管理—企业管理]