基于软投票集成模型的房地产上市企业财务预警研究
Research on Financial Distress Predicting of Listed Real Estate Companies using Soft Voting Ensemble Learning
姜凤珍 1李毛 1王骏1
作者信息
- 1. 青岛理工大学管理工程学院,山东青岛 266520
- 折叠
摘要
以中国房地产上市企业为研究对象,选取了反映企业偿债能力、盈利能力、营运能力和杠杆比率在内的 20 个指标,收集了 137 家企业在 1991～2021 年期间的财务数据,构建了基于软投票的集成模型来预测中国房地产上市企业一年期和两年期的财务困境.结果显示,集成模型提前一年期和两年期预测的AUC值分别为0.946 和 0.880,与性能最好的单一分类器相比预测性能更高.然后通过SHAP解释模型对集成模型中输入变量的解释能力进行分析.无论提前一年期还是两年期,预测准确率在很大程度上受销售净利率、净资产收益率、应收账款周转率、利息保障倍数和总资产收益率的影响.提出的模型通过对房地产企业财务困境高准确率的预测,识别与预测相关的主要变量,有望帮助房地产企业和其他相关利益者通过早期预警防止破产.
Abstract
In this study,taking Chinese listed real estate companies as research objects,20 financial indicators reflecting solvency,profitability,operating capacity and leverage ratios are selected,and the financial data of 137 companies are collected for the period of 1991-2021,and a ensemble models based on soft voting is constructed to predict the financial distress of Chinese list-ed real estate companies in the one-year and two-year periods.The results show that the AUC value of the ensemble models are 0.946 and 0.880 for one-year and two-year advance prediction,respectively,which are higher compared with the best-performing single classifier.The explanatory power of the input variables in the ensemble models are analyzed by the SHAP explanatory model.Prediction accuracy is heavily influenced by ROS、ROE、ART、TIE and ROA,irrespective of the one-year or two-year advance pe-riod.The models proposed in this study predict the financial distress of real estate companies with high accuracy and identify the main variables related to forecasting.The study is intended to assist real estate companies and other relevant stakeholders to prevent financial failures through early warning.
关键词
房地产企业/财务困境/集成模型/软投票/Shapley值Key words
Real Estate Companies/Financial Distress/Ensemble Learning/Soft Voting/Shapley Value引用本文复制引用
基金项目
国家自然科学基金青年项目(72001121)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据