The Research On Financial Crisis Early Warning Of GEM Listed Companies Based On The Stacking Fusion Model

In the context of economic globalization,the social competition is becoming increasingly fierce,and the enterprises are facing more and more problems and crises,among which the financial crisis is the most prominent form of crisis.Once the listed companies have financial problems,it will not only lead to the loss of investors’ income,and the company will face a bankruptcy crisis,but also may cause a huge impact on the whole economy and society,especially for the listed companies on GEM in China.China’s GEM has problems such as low listing threshold,short establishment time and imperfect relevant laws and regulations.Compared with other markets,companies listed on GEM will face greater financial risks and be more prone to financial crisis.Based on this background,discussing and establishing a scientific and reasonable financial crisis early warning model to prevent and resolve the financial crisis in advance has great significance for the long-term and stable development of GEM listed enterprises.Based on this situation,this paper defines the financial crisis signs of GEM listed enterprises in China according to the GEM listing rules issued by Shenzhen Stock Exchange,select 680 GEM listed companies from 2012 to 2019 as research samples,and select 31 early warning indicators from the two dimensions of financial indicators and non-financial indicators,six aspects of profitability,cash flow ability,growth ability and so on according to the index selection principle.Subsequently,this paper focuses on the construction of financial crisis early warning index system and early warning model.In terms of financial crisis prediction index system: in order to ensure the effectiveness of the model prediction,before the model construction,all the index data are first cleaned and normalized,and then the final financial crisis warning indicators entering the model are screened out through normality test,significance test and correlation analysis.The empirical results show that 29 of the original 31 indicators have passed the significance test,which can significantly distinguish between normal financial companies and financial crisis companies.After correlation deletion,15 indicators constitute the final financial crisis early warning index system.In terms of financial crisis prediction model: after dividing samples and comprehensively sampling unbalanced samples,four algorithm models,logical regression,support vector machine,random forest and Gradient Boosting DecisionTree(GBDT),were constructed based on the established financial crisis warning index system.After the parameter adjustment and optimization of each model,the model prediction results are compare and analyzed by using five model performance evaluation indicators,among which the best prediction effect is the GBDT model,with an accuracy of 93.56%.In order to further enhance the prediction performance of the warning model,using four single models as primary learners,GBDT model as the secondary learners,using the output of the primary learners as the input variable of the secondary learners,A fusion model with overall better performance over all single models was obtained by model fusion,whose prediction accuracy was improved to 94.06% and with AUC values up to 0.9658.Finally,the research content is summarized and the corresponding policy suggestions and the shortcomings of this paper are proposed.