基于CatBoost模型的上市公司财务困境预测

Financial Distress Prediction of Listed Companies Based on CatBoost Model

摘要: 本文按照财务异常与财务正常的公司数量为1:2的比例选取了2019~2020年间63家被特殊处理的上市公司和126家财务正常的上市公司。从公司发展能力、经营能力、盈利能力、股本结构、人员变动等十个方面的51个变量中利用统计方法和随机森林方法筛选出17个重要变量，在此基础上首次运用CatBoost模型进行公司财务困境问题上进行建模和预测，并与之前学者经常使用的XGBoost、随机森林和逻辑回归这三种建模方法进行对比。实验结果显示CatBoost模型在特异性、准确率和AUC值方面都要优于其他三种模型，其中准确率达到了98.2%，在准确性方面要比近年广泛运用且预测效果较好的XGBoost模型高3.5%。

Abstract: According to the ratio of 1:2 between the number of abnormal and normal financial companies, this paper selects 63 specially treated listed companies and 126 normal financial listed companies from 2019 to 2020. From the 51 variables of the company’s development ability, operation ability, profitability, equity structure and personnel change, etc., 17 important variables are selected by using statistical method and random forest method. On this basis, CatBoost model is used to model and predict the company’s financial distress for the first time and it is compared with the three modeling methods of XGBoost, Random forest and Logistic regression that are often used by previous scholars. The experimental results show that CatBoost model is superior to the other three models in terms of specificity, accuracy and AUC value, of which the accuracy is 98.2%, which is 3.5% higher than XGBoost model, which is widely used in recent years and has good prediction effect.