Financial Distress Prediction Of Chinese Listed Companies Based On CNN Model And Diversity Indicators

In the past three decades,the financial distress prediction of listed companies has been a research hotspot in the field of risk management.Effective early warning of financial distress allows investors to grasp the profitability of listed companies,adjust investment strategies and reduce expected investment losses;it can also help corporate managers to take remedial measures to avoid financial crises.At present,the Chinese equity market has become the main investment market for international investors.Therefore,the early prediction of the financial distress of Chinese listed companies has attracted more and more attention.This paper uses the optimized convolutional neural network method to construct a financial distress prediction model for Chinese listed companies.Using 576 index data of3,424 Chinese listed companies in the entire industry from 2012 to 2016 as input variables of convolutional neural networks,and the probability of financial distress as output variables,a7-layer convolutional neural network financial distress prediction was established model.Among them,576 indicators include 333 financial indicators,108 non-financial indicators and 135 macro indicators,which fully reflects the diversity of indicators.By combining optimization models and assigning optimal weights to different time windows,the indicators for the five consecutive time windows(time t-5,t-4,t-3,t-2,t-1)from 2012 to 2016 The data will have a differential impact on the probability of financial distress of Chinese listed companies in 2017(time t)in order to improve the model's forecast performance.The empirical results show that,first,adding non-financial indicators and macro indicators related to the company on the basis of financial indicators can improve the prediction accuracy of the model;second,the CNN model after multiple time window optimization combinations and the current distinction between years Compared with the researched CNN model,the prediction accuracy and differentiation are improved;thirdly,the impact of index information of multiple different years on listed companies is more accurate than that of individual years.Compared with the existing CNN financial distress prediction methods,decision tree(DT),support vector machine(SVM),logistic regression(Logist),neural network(NN)and other single window financial distress prediction models,The prediction accuracy,AUC value and KS value of after multi-window index combinationoptimization CNN financial distress model are the best,and the prediction accuracy is as high as 94.5%.This work provides new ideas for the rational use of diversity and multi-window characteristic indicators in the study of financial distress early warning.At the same time,the optimization combination method proposed in this paper can be applied to the optimization of the results of various machine learning models except CNN prediction,especially for the study of the impact of feature heterogeneity in different years.