险境淘金:基于并购重组事件挖掘的财务脱困预测

Gold panning:Financial relief forecasting of listed enterprises based on the mining of mergers and acquisitions

针对我国上市企业财务困境预测问题,构造了一个多分类器集成学习模型,挖掘企业并购重组事件(mergers and acquisitions,M&A)以及年报中管理层讨论与分析(management discussion and analysis,MD&A),应用文本分析技术研究其能否提供增量信息,以及新特征的信息价值.研究结果表明,新模型在预测准确度(area under curve,AUC)与识别能力(true positive rate,TPR)上均显著优于基准模型;企业财务数据、M&A,MD&A等的多源异构特征,都帮助该模型获得更佳的预测效果;基于MD&A的文本情感挖掘发现,管理层语调越消极悲观,其企业越易于陷入财务困境;频繁发生M&A事件更易使企业趋于陷入财务困境;MD&A中语调夸大将不利于模型预测的准确性,但大规模M&A会削弱这种消极作用.

This paper constructs a multi-classifier ensemble learning model for listed enterprises in China to forecast their potential financial distress.The novel model incorporates the incremental information from their mergers and acquisitions(M&A)and annual report management discussion and analysis(MD&A)to mine them via text analysis techniques,and quantifies the values of various new features.The experimental results show that the proposed model significantly outperforms the benchmarks in the area under curve(AUC)and the true positive rate(TPR).From financial data,MD&A and M&A,the integration of multi-source new features helps the model obtain better results.The sentiment mining of the texts finds that the more negative the management tone,the more likely the enterprises falling into financial distress;in most cases,frequent large-scale M&A events are likely to incur financial difficulties;the model deteriorates if the management exaggerates in their texts,but large-scale M&A events help to weaken its effects.