Research On Financial Risk Early Warning Of GEM Based On ADASYN-logit

Since its establishment on October 30,2009,the Growth Enterprise Market(GEM)has been trading openly for nine years.During these nine years,the market scale has been growing and the trading mechanism has gradually improved.However,according to the Financial Safety Assessment Report of China's Non-financial Listed Companies in 2018,the overall financial safety situation of the GEM market in 2018 is the worst among the three sub-markets: the main board,the small and medium-sized board and the GEM.The decline rate is also the fastest among the three sub-markets.The financial risk is higher than that of the main board and the small and medium-sized board,and the overall investment risk is further increased.Therefore,with the rapid development of the GEM as a whole,how to establish a financial risk early warning model to help GEM companies predict the crisis ahead of time,so as to avoid the occurrence of financial crisis is an urgent problem to be solved in the GEM market.China's GEM market is an emerging market,which is different from foreign GEM market,so the research of domestic and foreign scholars is relatively small.On the issue of sample selection conditions,because GEM does not have ST warning system similar to the main board and SMEs,and the financial risk situation of GEM itself is worse than that of the main board and SMEs.In order to better and more timely warn the financial risk of GEM companies,this paper defines GEM financial crisis company as "the latest accounting" on the basis of analyzing the relevant system documents.A company whose annual net profit is negative or whose net assets are negative in the latest fiscal year is a company with normal financial position if one of the two conditions is not met.According to this condition,this paper selected 160 GEM companies,including 40 financial crisis companies and 120 financial normal companies.The ratio of the two samples is 1:3.The sample set is divided into two groups,one is training group and the other is testing group.This paper does not follow the 1:1 balanced proportion used in many previous studies,in order to be closer to the real proportion of financial risk companies and financial normal companies.Because there are many financial indicators in a company,it is impossible to take them all as characteristic indicators.After referring to a large number of previous studies,this paper selects 24 financial indicators from five aspects: profitability,operating capacity,solvency,development capacity and cash flow status.The financial crisis happened in the year of T,and according to a large number of studies,the model based on the financial data of T-1 year,that is,the year before the financial crisis,has the highest accuracy.Therefore,this paper also selected the financial data of these sample companies in T-1 year as the sample data.Because the random sampling method is used in the selection of samples,and the difference of financial data between different sectors and companies of different sizes in GEM market is not considered,so in order to reduce the adverse impact of the abnormal values on the modeling,the Winsorize extreme value of the collected sample data is processed by up to and down 1%.After data pretreatment,in order to eliminate the financial indicators which have no significant difference between the financial crisis and the normal financial company,this paper first carries out T-test and Wilcoxon sign rank test on the sample data,and the test results show that all the sample data are different.Secondly,in order to study which kind or types of financial indicators have a significant impact on the variable of financial risk,and to alleviate the interference of multi-collinearity between financial indicators on the follow-up empirical research,after data preprocessing,this paper carries out PCA dimensionality reduction on 24 financial indicators,and finally obtains eight public factors,each of which contains its own wealth.The business information is different.So far,the sample selection and the construction of the characteristic index system have been completed.In the empirical part,according to the principle of Logit model,this paper carries out two-class Logit regression on eight characteristic indicators,after eliminating the insignificant independent variables,establishes a financial risk early warning model of GEM based on Logit.The result of regression shows that the financial indicators of GEM company's profitability and operating ability will have a significant impact on the company's financial risk.At the same time,after testing the model with the test group samples,it is found that the classification results of the general Logit model established under the unbalanced samples will be significantly biased towards the normal financial companies,unable to accurately judge the existence of financial crisis companies,which has no practical significance.Therefore,this paper introduces the ADASYN method,which enlarges the number of financial crisis samples by three times,and then with the original wealth.The normal sample combination forms a new sample set.Based on the new sample set,this paper establishes a financial risk early warning model of GEM based on ADASYN-Logit.The result of regression shows that GEM's profitability,operating ability and solvency will have a significant impact on the company's financial risk.The test results show that the overall prediction accuracy of the improved GEM financial risk early warning model based on ADASYN is 75%.The ability to identify financial crisis samples is also significantly stronger than that based on the ordinary Logit financial risk early warning model.In order to analyze and judge the prediction performance of the two models more accurately,this paper uses two evaluation criteria of F value and G value based on previous research.At the same time,because ADASYN is an improvement of SMOTE oversampling method,in order to compare the effect of the improvement,the financial risk early warning model of GEM based on SMOTE-Logit is added to compare.From the F and G values of the three models,under unbalanced samples,ADASYN-Logit-based company financial risk early warning model has the best prediction performance,SMOTE-Logit-based company financial risk early warning model takes the second place,and ordinary company financial risk Logit early warning model is the worst.