Research On The Early Warning Of Corporate Financial Crisis Based On Transformer

Under the influence of multiple factors such as unstable market economy,fierce market competition,and poor enterprise management,the business risks faced by enterprises continue to increase,and the financial risks of some enterprises continue to accumulate,resulting in the continuous deterioration of their financial conditions and eventually financial crisis,even on the road to bankruptcy and liquidation.Once a financial crisis occurs,it will cause huge economic losses to enterprises,shareholders,creditors,etc.,and will also have a great negative impact on the stable operation of the market economy.A scientific and effective financial crisis early warning model can help enterprises manage and prevent the occurrence of financial risks,and provide judgment basis for investors to adjust investment strategies and creditors to enter into creditor-debt relationships.Therefore,constructing a scientific and effective financial crisis early warning model is of great significance and value to protect the interests of enterprises,investors and creditors.Deep learning is widely used in complex nonlinear tasks because of its strong learning ability,fast operation speed,no restrictions on the form of input data,and the ability to effectively fit complex nonlinear relationships between input data.The early warning of financial crisis of an enterprise is a complex nonlinear time series task,which can be realized by building a model.Therefore,this paper attempts to use the cutting-edge algorithm Transformer of deep learning to predict whether the company will have a financial crisis in the future,and then play an early warning role.This paper takes the listed A-share manufacturing enterprises as the research object,and proposes a model based on Transformer to warn the financial crisis of enterprises.First,by analyzing the reasons for the formation of corporate financial crisis,the financial crisis early warning indicators are selected from the perspective of available financial information,and two groups of different corporate financial crisis early warning indicator systems are constructed using correlation analysis.Secondly,considering the procedural nature of the occurrence of corporate financial crisis,a total of 5 data sets in the 4,6,8,10,and 12 quarters before the predicted time point of the sample corporate financial crisis were selected for demonstration.By comparing the accuracy rates of the models under different early warning index systems,it is concluded that when using Transformer to build a model to process classification tasks,it is not necessary to eliminate the high correlation indicators contained in the indicator system.The prediction accuracy of the model shows that the data set with a sequence length of 10 quarters has the best early warning effect,and the data set with a sequence length of 4 quarters has the worst early warning effect.Third,in order to reduce the impact of the insufficient sample size of ST enterprises on the accuracy of the model,this paper expands the data set by three times through data enhancement,and the accuracy of the model under the new data set can reach 98.25%,which proves that the model can be used for it provides an accurate basis for judging whether an enterprise has financial crisis in the future,and is an ideal model for enterprise financial crisis early warning.Finally,by comparing the Transformer-based enterprise financial crisis early warning model constructed in this paper with the typical deep learning models CNN and LSTM models,it is concluded that in the task of realizing corporate financial crisis early warning,the Transformer model has higher accuracy which means the model is superior.