联系我们

论文查询结果	返回搜索

论文编号：	7534
作者编号：	2120132350
上传时间：	2015/6/11 20:58:10
中文题目：	基于集成神经网络的上市公司财务危机预警研究
英文题目：	Financial Crisis Prediction of Listed Companies Based on Integrated Neural Network
指导老师：	安利平
中文关键字：	财务预警；数据挖掘；集成神经网络；AdaBoost
英文关键字：	financial forecast, data mining, integrated neural network, AdaBoost
中文摘要：	随着经济全球化的发展，市场经济竞争愈发激烈，我国上市公司因日益恶化的财务状况而被特殊处理的情形愈发严重。企业财务危机的发生，不仅对自身生存和发展构成威胁，也给利益相关者带来负面影响。因此，本文主要利用上市公司公开的财务报表数据，构建完善的财务危机预警指标体系及稳定可靠的财务危机预警模型，帮助企业管理人员和相关利益者进行决策。 本文首先介绍了研究背景及研究意义，回顾了前人研究采用的指标体系和建模方法，明确了本文的研究框架和研究方法。然后，对财务危机的相关理论和集成神经网络构造理论进行详细阐述，为后面模型的构建奠定了基础。在构建财务预警指标体系时，本文从公司偿债能力、营运能力、盈利能力、发展能力和现金支付五个方面进行选取，并加入能反映公司治理变量和审计意见的非财务指标，利用统计分析软件SPSS对建模指标进行科学地筛选和提取。其次，在原始AdaBoost算法分析的基础之上，本文提出了优化训练集和优化弱分类权重的改进算法。最后，本文对构建的基于BP_AdaBoost的预警模型、基于组合决策的BP_AdaBoost的预警模型和基于非平衡数据分类BP_AdaBoost的预警模型进行实证分析，并对三个模型的预测性能进行比较。结果显示，不同的模型在时效性、准确率等方面各具优势，在实际问题中可以灵活选择建模方法。 本研究的创新之处在于：一是统筹考虑财务指标和非财务指标，利用显著性检验、主成分分析等方法对财务预警指标进行科学地初选、筛选和提取。二是改进原始AdaBoost算法选取弱学习样本的方法，使得选择下一轮的训练样本不再来自上一轮分错的样本中间，而是来自前面T轮组合判决后分错的样本中间。三是改变传统的基于最小分类错误率的思想，考虑基于风险最低的集成神经网络对财务状况进行预警，改进AdaBoost算法权值调整方法，更加关注小类样本。
英文摘要：	With the development of globalization, competition in market economy grows fiercely. Many listed companies have been titled ‘ST’ because of its financial distress in China. It is not only the threat to the existence and development of company itself, but also brings great loss to stakeholders. Therefore, it is important that we could use public data and choose suitable financial crisis early-warning features and establish reliable and stable financial distress early-warning model in order to help stakeholders adjust the management strategy in time. This paper introduces the background and potential research significance at first, and then systematically reviews the literature research in the area of company financial distress on the index systems and the method of model building. This paper sets up the research framework and methods. It discusses the related theory of financial crisis and integrated neural networks technology, all of that supply theory basic for the building of financial distress forecasting model. In terms of choosing suitable index system, the features that reflect the areas of the company-the ability of debt paying、the ability to operate、the profitability、development ability and cash flow. This paper adds some non-financial indicators into the model, which can reflect the corporate governance and audit opinions to enable the predictors can be more comprehensive and better to reflect the company's business situation. Then, this paper uses traditional statistic tools SPSS to select appropriate index. After that, based on the detailed analysis of the AdaBoost algorithm, this paper presents two new AdaBoost algorithms which are the method based on optimizing training dataset and the method on optimizing the weights of weak classifiers. At last, this paper builds up the BP_AdaBoost model、the BP_AdaBoost model for imbalanced of financial class sample set and the BP_AdaBoost model of combined decision, and then compares the performance of three models. The empirical results show that the different models have respective advantage. The innovations of this paper are: Firstly, the preliminary selected early-warning features are reasonable and scientific. Secondly, this paper improves the original AdaBoost algorithm on the method of selection of weak learning samples. The new algorithm for the next iteration of training samples doesn’t come from the error class samples of previous iteration. It is from the previous T iterations of error class samples. Thirdly, this paper changes from the traditional concept based on the minimal classification error rate to the integrated neural network based on the minimal risk to forecast the financial situation.
查看全文：	预览  下载(下载需要进行登录)