| At present,in the context of the ongoing global epidemic and the war of Russia-Ukraine and other local conflicts,the global economy is suffering from an unprecedented impact.The business,finance and operation of enterprises around the world have been severely impacted,and they are facing the possibility of loss and bankruptcy at any time.Many enterprises are actively looking for a steady path of stable development,many scholars at home and abroad also related research on enterprise financial distress early warning,from univariable warning to multivariable early warning,to now machine learning algorithm warning,are seeking a more accurate and more stable early warning method,so that enterprise decision makers ready before the dilemma.Therefore,it is very necessary to study the early warning of enterprise financial difficulties.This paper selected the Shanghai A shares in 2020,2021 by the CSRC mark * ST manufacturing listed enterprises 134,financial normal 787 manufacturing enterprises as a research sample,according to the * ST rules,climb from the Tai’an database target enterprise T-3 years(2017,2018)operating capacity,profitability,solvency,growth,cash flow,indicators per share,relative value index,dividend distribution 8 dimensions of 56 financial indicators for financial distress warning research.First,Smote algorithm is used to overcome the influence of unbalanced data on training.Secondly,based on the dimension reduction of data in Lasso regression and Auto-encoder(AE),the performance of two dimension reduction methods in the early warning of financial difficulties is given.Then using the grid search to determine the optimal number of hidden layer nodes(ELM),build the limit learning machine financial dilemma warning model,the empirical results show that the Lasso regression and autocoder dimension reduction limit learning machine model average warning accuracy is about 85.49% and87.33% respectively,with significant classification effect.In view of the weights and deviation of the random training output instability,this paper uses the particle group algorithm(Particle Swarm Optimization,hereinafter referred to as PSO)to improve the improved limit learning machine in classification index and stability,and the numerical experiments show that Lasso regression and autoencoder and the average warning accuracy reached 90.52% and 92.97%,the variance is also significantly improved.Meanwhile,compared with the traditional financial dilemma warning models such as logistic regression,support vector machine and neural network,the improved limit learning machine of the particle swarm algorithm has the best performance in the financial dilemma warning.In addition,the autoencoder dimension reduction method is better than Lasso regression in terms of classification performance and data integrity.Overall,the improved limit learning machine model of the autoencoder has the best performance in the financial dilemma warning of Shanghai A-share manufacturing industry.Enterprise decision-makers can effectively conduct effective prevention and investigation through effective early warning models and important financial indicators,and actively safeguard the interests of enterprises in the overall environment of economic instability to achieve the purpose of sustainable and stable operation.In the context of global economic turbulence,this paper takes the Shanghai A-share manufacturing industry as the research object,explores the optimal financial dilemma warning model,uses Lasso regression and autoencoder to reduce data,combines the limit learning machine with the financial distress warning,and improves the improved classification performance,and compares the empirical results of different dimension reduction methods,obtaining the relatively optimal financial dilemma warning model.This paper explores the early warning of financial distress in Shanghai A-share manufacturing industry,which can provide another thinking for the research of financial distress early warning. |
PDF Full Text Request