Predicting Default Of Chinese A-stock Listed Firms Through The Logit Model With High Dimensional Mixed Frequency Data

In the era of big data,high-dimensional mixed frequency data is widespread,which brings opportunities and challenges for credit evaluation and risk management.It is particularly important for operators,investors,financial institutions and government regulatory departments to make full use of high-dimensional mixed frequency data information,extract important features from high-dimensional variables,establish a timely and accurate credit evaluation model,and predict the default risk of listed companies.Considering the high-dimensional mixed frequency data information,this dissertation modifies the standard Logit model,establishes a new high-dimensional mixed frequency data Logit model and applies it to the default prediction of Chinese A-stock listed firms.This research contains mainly two aspects:(1)In order to solve the problem of mixed frequency data in default prediction of listed companies,this dissertation introduces the Mixed Data Sampling(MIDAS)method into the Logit regression context to develop the Logit-MIDAS model and its maximum likelihood estimation method is proposed.The constructed Logit-MIDAS model can handle raw mixed frequency data directly,avoiding the information loss caused by frequency conversion.And it can make full use of high-frequency information and make timely prediction.In a real-world application to credit risk prediction of listed company in mainland,the Logit-MIDAS model is superior to the standard Logit model in terms of classification and forecasting ability.Considering the imbalance of sample data,three data sampling methods are used to enhance the accuracy of prediction.The empirical results show that the Logit-MIDAS model also outperforms the standard Logit model for both in-sample and out-of-sample tests.(2)In order to solve the problem of identifying important factors in default prediction of listed companies,this dissertation introduces Group Lasso method to select high-dimensional variables in the unrestricted mixed frequency data sampling Logit(Logit-U-MIDAS)model,and constructs Logit-U-MIDAS-Group Lasso model,which is adaptive to high dimensional mixed frequency data analysis by performing group selection and model estimation simultaneously.In the default prediction of Chinese listed companies,the Logit-U-MIDAS-Group Lasso model can effectively identify the important influencing factors from the high-frequency financial factors and low-frequency corporate governance variables,which not only effectively reduces the cost of financial supervision,but also accurately predicts the default of listed companies.With the help of sampling methods on imbalanced data,the Logit-U-MIDAS-Group Lasso model is further improved in prediction ability,and performance better than the standard Logit model and LASSO Logit(L1 Logit)model for both in-sample and out-of-sample tests.The research results show that the Logit model on high-dimensional mixed frequency data has good prediction,which can not only effectively identify key factors and reduce the cost of financial supervision,but also improve the accuracy of prediction and assist financial regulatory decision-making.Furthermore,the results can provide basic tools and decision-making reference for companies to improve their business ability,investors to reduce investment risks,and regulatory departments to improve regulatory efficiency.The research has a certain academic value and practical significance.