Dynamic Prediction Of Financial Distress Based On Imbalanced Data Stream Of An Industry

The research on financial distress prediction is always regarded as a hot topic in the area of the enterprise risk management,getting a lot of attention from the management department and researchers.Using the theory framework of financial distress and the practical model to improve the prediction accuracy is also an important task.Beginning with the concept of financial distress,theoretical and empirical study on financial distress prediction has gone through from traditional statistic analysis to artificial intelligent learning,from statical prediction to dynamic prediction,from binary classification to multi-class classification,from balanced data stream to imbalanced data stream,and the prediction accuracy gets increased gradually.In the 21th century,market competition is more diversified.As the competition model and business prospect of the industries develop to different orientations,the criterion of evaluating financial indexes of different industries should be different.For example,it is very normal that the enterprises in burgeoning industries have a high debt-to-assets ratio,while it is a sign of financial distress when it comes to traditional industries.But the existed financial distress prediction theories do not consider whether an enterprise will fall into financial distress or not in the context of a specific industry.Meanwhile different industries have different industry culture,operating procedure and government orientation.Thus the enterprise that have the same values of financial indexes may be categorized as different financial classes because of different industries.Then the concept of financial distress is no longer uniform and fixed.In order to determine the financial distress correctly,it should consider the environment of certain industry,but not simply according to the same criterion.What'more,competition in the market is constantly changing,but the statical prediction model ignores the mechanism that new sample data stream enters with time going by,and it cann't adapt to the trends of the enterprise's operating.Combing with theoretical knowledge of financial distress early warning system,this paper puts forward the concept of relative financial distress based on the same industry.Then proceeding dynamic financial distress forecasting with one industry's lateral financial data stream of the all listed companies,hoping to provide new directions and ideas in dynamic prediction of financial distress.Firstly,according to the meaning of financial distress and the concept of financial distress drift,this paper proposes the definition of financial distress based on the same industry:the category determination of different financial condition should be the result of judging under the same industry background.Secondly,in the module of financial index selection,this paper selects six types of indicators including the solvency,operational capacity,profitability,development capacity,structural ratio and per share index.In detail,totally 33 financial indicators are selected as alternative characteristic indexes,and plus L minus R is uesd method for feature selection.In financial distress determination model,this paper uses principal component analysis to determine the enterprises' financial class with the ranking order in the same year.In financial distress prediction model,six dynamic prediction models including SmoteBoost-SVM,SmoteBoost-DT,SmoteBoost-KNN,SmoteBoost-Logistic are trained based on the results of financial distress determination model.Finally,this paper chooses the steel industry as the object of empirical research,and collect this industry's financial data in 2000-2015 consolidated annual reports of all listed steel companies,establishing six dynamic training sets with every 10 years of financial data,eg.six training sets consisted by the years 2000-2009,2001-2010,2002-2011,2003-2012,2004-2013 and 2005-2014 financial data.By dynamic determinating and training the financial situation,this paper finds that the four prediction models all have good financial distress prediction accuracy.