我国上市公司资产重组绩效预测方法的实证研究

发布时间：2018-07-22 20:18

【摘要】：随着我国证券市场的发展,上市公司面临财务状况异常或其他异常情况,资产重组便成为许多上市公司进行改善绩效的重要方法之一,它将作为证券市场里最能体现市场效率、最具创新活力的一环在我国上市公司中不断上演。究其原因,主要是由于资产重组可以帮助ST公司调整产业结构、优化资源配置、改善企业经营结构,提高资产营运绩效,从而实现成功摘帽。而资产重组绩效预测则是作为对ST公司在重组前进行判断其能否成功摘帽的一种方式,在公司的经营管理中扮演着举足轻重的角色。因此,如何对ST公司进行资产重组绩效预测以及如何提高其预测准确率是刻不容缓的。本文以上市公司资产重组作为研究对象,由于收集的数据集显示重组成功的企业与重组失败的企业的数目存在较大的非平衡性,虽然这是符合客观实际,但是相对传统的机器学习算法来说,通常会产生偏向多数类的结果,因而对于具有重要作用的少数类而言,预测的效果会相对较差。为了提高准确率,本文对收集的数据进行了平衡性处理。之后,利用单一预测模型来判断企业资产重组是否取得好的效果,并且选用十种方法进行性能比较,其中包括logit、probit、SVM、 MDA、CBR、BaggingLOGIT、BaggingSVM、BaggingPROBIT、BaggingCBR和BaggingMDA。结果显示：SVM和CBR模型对上市公司资产重组绩效的判断准确率要优于其它八种预测模型,也就是说这些模型可以判断出80%以上的ST公司能否在资产重组后1年恢复上级状态。通过此次研究,我们可以为ST公司能否通过资产重组来改善企业绩效提供一定的理论依据。其次,为了提高我国上市公司资产重组绩效预测的准确性,本文在单一预测模型的基础上对预测模型进行了改进,采用不同与前人对资产重组绩效研究的方法,应用聚类混合分类器与单一预测模型以及聚类融合分类器与单一预测模型相结合,建立了十种新的预测模型,即CLOGIT、CPROBIT、CMDA、CSVM、 CCBR、BaggingCLOGIT、BaggingCPROBIT、BaggingCMDA、BaggingCSVM和BaggingCCBR对资产重组绩效进行预测并且对这十种模型进行性能的比较。结果显示：聚类融合算法与聚类算法建立的模型无论是在准确率还是在真正率和真负率方面都有积极的表现。其中,在准确率和真负率方面,通过聚类混合建立的模型的预测均值要高于聚类融合分类方法建立的模型；而对于真正率,聚类融合分类得到的均值要优于聚类混合方法建立的模型。再次,为了对以上建立的二十种模型进行验证,本文另外收集了2011-2012年的ST公司作为新样本集进行分析,检验对上市公司资产重组绩效预测研究的重要性以及本文所用各个资产重组预测模型的实用价值。实验结果表明传统统计模型的预测结果次于人工智能方法建立的模型。其中,支持向量机及其集成、聚类形成的模型的预测结果在非平衡数据还是平衡数据中均相对稳定且效果较好,而案例推理及其集成、聚类形成的模型的预测结果在非平衡数据中相对较好。最后,本文对其研究进行总结,并指出了相关的管理启示,在实际应用中具有重要的作用,并为企业管理者进行决策提供了一定的理论依据。
[Abstract]:With the development of China's securities market, listed companies are faced with abnormal financial conditions or other abnormal conditions. Asset reorganization has become one of the most important ways for many listed companies to improve their performance. It will be performed in the listed companies in our country as the most efficient and innovative part of the stock market. The main reason is that the asset reorganization can help ST company to adjust the industrial structure, optimize the allocation of resources, improve the enterprise management structure, improve the performance of the operation of the assets, and thus achieve the success of the cap. It plays an important role. Therefore, how to predict the performance of ST assets restructuring and how to improve its accuracy is urgent.
This paper takes the assets reorganization of the listed companies as the research object. As the collection of data sets shows that the number of successful restructuring enterprises and the number of enterprises that have failed to reorganize has a large non balance. Although this is in accordance with the objective reality, the result is usually biased toward the majority of the traditional machine learning algorithms. In order to improve the accuracy, this paper deals with the collected data in order to improve the accuracy. Then, the single prediction model is used to determine whether the enterprise asset reorganization has achieved good results, and the ten methods are selected for performance comparison, including logit, probit, SVM, MDA, CBR, Ba. The results of ggingLOGIT, BaggingSVM, BaggingPROBIT, BaggingCBR and BaggingMDA. show that the accuracy of SVM and CBR model is better than the other eight forecasting models for the performance of the listed companies' assets reorganization, that is to say, these models can judge whether more than 80% ST companies can recover their superior state in 1 years after the restructure. Through this study We can provide a theoretical basis for ST company to improve its performance through asset restructuring.
Secondly, in order to improve the accuracy of asset reorganization performance prediction in China's listed companies, this paper improves the prediction model on the basis of a single prediction model, adopts the methods of different and previous research on the performance of asset reorganization, and applies the mixed classifier and the single predictive model, the Clustering Fusion classifier and the single prediction model. Combined, ten new prediction models are established, that is, CLOGIT, CPROBIT, CMDA, CSVM, CCBR, BaggingCLOGIT, BaggingCPROBIT, BaggingCMDA, BaggingCSVM and BaggingCCBR are used to predict the asset reorganization performance and compare the performance of the ten models. The results show that the cluster fusion algorithm and the clustering algorithm are in the model The accuracy rate is also positive in both real and true negative rates. In terms of accuracy and true negative, the mean value of the model established by cluster mixing is higher than that of the clustering fusion classification method, while the mean value of the clustering fusion classification is better than the model established by the clustering method for the real rate.
Thirdly, in order to verify the twenty models established above, this paper also collected 2011-2012 years' ST company as a new sample set to analyze the importance of the performance prediction research on the assets reorganization of the listed companies as well as the practical value of the asset reorganization prediction model used in this paper. The experimental results show the traditional statistical model. The prediction results are second to the model established by artificial intelligence methods. Among them, the prediction results of the support vector machine and its integration, the prediction results of the clustering model are relatively stable in the unbalanced data or in the balanced data, and the case reasoning and its integration, the prediction results of the model formed by the cluster are relatively better in the non balanced data.
Finally, this paper summarizes the research, and points out the related management enlightenment, and plays an important role in the practical application, and provides a certain theoretical basis for the enterprise managers to make decisions.
【学位授予单位】：浙江师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51;F275

【参考文献】