基于组合预测技术的上市公司财务困境预测问题研究

发布时间：2018-06-06 21:37

本文选题：财务预测 + 逻辑回归　；参考：《山东财经大学》2013年硕士论文

【摘要】：我国的资本市场已走过了二十多年的历程，，企业所面临的竞争日趋激烈，企业经营稍有不慎就可能陷入财务困境，上市公司因财务状况异常而陷入困境的情况却屡见不鲜。因此，利用上市公司的财务数据，建立上市公司财务困境预测模型以揭示风险，己成为上市公司管理者、投资者和债权人等相关利益方共同关注的问题。本文共选取来自于多个行业的146家上市公司的财务数据，包括73家被特别处理（用ST表示）的上市公司的财务数据和与之配对的73家运行良好的公司的财务数据，把它们分为49对训练样本和24对测试样本。根据国内外学者对企业财务困境预测的研究成果，选择了20个财务指标构建了财务预警指标体系。利用T检验和非参数检验对原始预测变量进行显著性检验，使得预测的指标体系更具有合理性，并运用因子分析法消除指标间的共线性，得到了7个因子。分别采用多元判别分析、Logit回归分析和最小二乘支持向量机对训练样本数据进行了分类判断，构建了三个单项财务困境预测模型；并在此基础上，构建了一个线性组合预测模型和一个基于神经网络的组合预测模型，最后利用测试样本对五个模型进行了有效的验证，并对五个检验结果进行了比较。研究结果表明：组合预测模型对上市公司财务困境的预测效果，跟单一方法比较，预测精度没有显著提高，但组合预测模型更具有稳定性；基于神经网络方法的组合效果好于线性组合模型。
[Abstract]:The capital market of our country has gone through the course of more than 20 years, the competition that the enterprise faces is becoming fiercer day by day, the enterprise manages a little carelessly may fall into the financial distress, the listed company because of the financial condition unusual but the predicament situation is common but common. Therefore, using the financial data of listed companies to establish a prediction model of financial distress of listed companies to reveal the risks, they have become the managers of listed companies. This paper selects the financial data of 146 listed companies from many industries. It includes financial data of 73 listed companies specially processed (expressed in St) and 73 well-run companies matched with them, which are divided into 49 pairs of training samples and 24 pairs of test samples. According to the research results of domestic and foreign scholars on financial distress prediction of enterprises, 20 financial indicators were selected to construct the financial warning index system. Using T-test and non-parametric test to test the significance of the original predictive variables, the prediction index system is more reasonable, and the co-linearity between the indicators is eliminated by factor analysis, and seven factors are obtained. The training sample data are classified and judged by multivariate discriminant analysis logit regression analysis and least square support vector machine, and three single financial distress prediction models are constructed. A linear combination prediction model and a combined prediction model based on neural network are constructed. Finally, five models are validated with test samples, and the five test results are compared. The results show that: compared with single method, the forecasting accuracy of the combined forecasting model is not significantly improved, but the combined forecasting model is more stable; The combination effect based on neural network method is better than that of linear combination model.
【学位授予单位】：山东财经大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：F275;F832.51;F224

【参考文献】