Research On Dynamic Financial Distress Prediction With Multi SVM Ensembles Based On Class Imbalance And Time Weighting

Since the 21 st century,China's economic and trade cooperation with the international community has become closer.While sharing the trade convenience of "global village",Chinese enterprises are also suffering from risks and challenges under complex international economic environment.Affected by the financial crisis,China's economy has been in a downward trend in recent years,and there are many companies running into financial distress due to poor management every year.At the same time,banks and other financial institutions have also been affected as many companies run into financial distress and are unable to repay the loan,which to some extent affected the normal operation of the economic system.Based on this economic situation,there is an urgent need to establish a scientific and efficient financial distress prediction system.However,most of the current research is based on class balanced and static data samples,which makes the models constructed cannot effectively adapt to the financial distress prediction under class imbalance and dynamic data stream environment.In this context,this paper performs research of financial distress prediction from the perspective of class imbalance and dynamic data flow.Firstly,the article indicates the background and significance of the research,and introduces the research content,research methods,technical routes and innovations of the thesis.Secondly,by reading a large amount of literature,summarizing the research results of predecessors,the concept,type and causes of financial distress are analyzed.Then,the construction methods of financial distress prediction index system are summarized,and the emerging financial distress prediction methods are sorted out.For the problem of financial distress concept drift caused by dynamic data flow,the meaning and causes are pointed out,and its solutions are summarized.On this basis,the listed companies in Shanghai and Shenzhen stock market are taken as the research object.The 438 listed companies that are specially treated for the first time from 2002 to 2016 are collected as the financial distress sample companies.Then,according to the matching principle,2190 financial normal sample companies are selected in proportion of 1:5,thus forming a sample of 2628 companies.Furthermore,the sample companies' financial data that are produced two years before financial distress year are taken as the initial data set.After data preprocessing,normality test,the mean comparison,stepwise discriminant analysis and multiple collinearity test,the financial distress prediction sample dataset containing 19 financial indicators was obtained.Through the descriptive statistics and financial indicator management analysis of the sample dataset,the results show that the financial indicators of the two types of sample companies generally change dynamically.Besides,compared with the financial normal sample companies,the financial distress companies' asset turnover is slow,their profitability,compensation debt capacity and shareholder return rate are low,their development capacity is limited and overall business performance is poor.In this paper,the Support Vector Machine(SVM)is taken as the core algorithm,which is used to construct financial distress prediction model.As the economic environment changes will lead to the phenomenon of financial distress concept drift,the model built with the old samples is not suitable for the dynamic financial distress prediction.Therefore,this paper introduces the time weighting method to give higher weight to the new samples,and by combining the multiple classifier integration algorithm,the benchmark model is constructed.At the same time,because the class imbalance problem will greatly weaken the financial distress prediction effect,this paper introduces oversampling technology to deal with class imbalance problem.Therefore,this paper uses the external integration oversampling method and the internal embedded oversampling method to construct two financial distress prediction models based on class imbalance and time weighting.On this basis,Matlab software is used to carry out the simulation experiment of the constructed financial distress prediction models,and the obtained empirical results show that:(1)the prediction accuracies of the model constructed with the external integration oversampling method and the model constructed with the internal embedded oversampling method are both higher than that of the benchmark model.It indicates that the class imbalance problem will reduce the discriminant accuracy of the financial distress prediction model for minority class samples.The introduction of oversampling method can increase the management information of minority class financial distress samples,and effectively achieve the information balance between two kinds of samples,thereby improving the model's prediction performance.(2)Comparing the two oversampling methods,the model constructed by the internal embedded oversampling method is of higher classification accuracy,which shows that combining the oversampling method with the time weighting method can more effectively utilize the new important financial distress samples.It not only solves the problem of class imbalance,but also effectively improves the adaptability of the model to dynamic data streams.On the basis of empirical research,combined with the case of listed companies dealing with financial distress,this paper proposes the following suggestions and management implications for listed companies to prevent and respond to financial distress:(1)the enterprises should reasonably select and use modeling samples,scientifically construct and select prediction models,thereby improving the prediction performance of the financial distress prediction model.(2)In order to promote financial pre-existing prediction and control,enterprises should pay attention to the selection and control of cost and expense indicators,capital liquidity indicators and enterprise risk indicators,so as to effectively prevent financial distress.(3)Once the enterprise is found to be in financial distress by the financial distress prediction mechanism,enterprises should adjust their business strategies in a timely manner according to changes in the economic environment,conduct in-depth analysis of financial indicators and operating conditions,and identify key factors that cause financial distress.Then,enterprise should take timely measures to prevent financial distress from getting worse.