Research On Dynamic Financial Distress Prediction Based On Nonlinear Mixed Frequency Model

Mixed Data Sampling(MIDAS)is an important statistical model to solve the problem of mixed data analysis with inconsistent observation frequencies between response variables and explanatory variables.This model can overcome the information loss of the high-frequency data and the error of artificially filling data caused by the data preprocessing method of transforming the mixed frequency data to the same frequency,and keep the original data structure of the mixed frequency data.However,the existing mixed frequency models are still dominated by linear models.In this paper,two kinds of nonlinear mixed frequency models are proposed by combining the MIDAS technique with the Logistic model,and we apply them to the financial distress prediction of Chinese manufacturing listed companies.Firstly,we construct the MIDAS-Logistic model at different time windows,which considering the annually observable financial ratios and macroeconomic factors,the quarterly GDP growth ratio and the monthly Inflation rate as explanatory variables,and the annually observable financial status is used as the response variable.We explore the different performance in the short-term prediction for 2-3 years and the medium-long term prediction for 5 years before the manufacturing listed companies' financial distress.Secondly,the RRMIDAS-Logistic model is constructed by considering the quarterly financial status as response variable and the annually observable financial ratios,the quarterly and monthly observable macroeconomic factors as explanatory variables to capture the quarterly information of listed companies' ST status,which realize the relatively high frequency financial distress prediction.There are two innovations and characteristics in this paper: Firstly,we construct the nonlinear mixed frequency models,which can solve the problem of mixed frequency financial distress prediction based on the indicators and financial status at different frequency.The high frequency macroeconomic factors are introduced to explore the impact of macroeconomic fluctuations on the companies' financial distress.Secondly,different from the existing research of the annually financial distress prediction,we construct the relatively high frequency financial distress prediction model with the quarterly information of the listed companies' financial distress label,which can solve the problem of the financial distress lag disclosure.The empirical results show that:(1)In the short-term financial distress prediction of Chinese manufacturing listed companies for 2-3 years ahead,the MIDAS-Logistic model considering high frequency macroeconomic effect outperforms the traditional Logistic model constructed with the same frequency data,especially the AUC of the mixed frequency model is improved by 7.45% and the Type II error can be significantly reduced compared with the Logistic model.(2)The RRMIDAS-Logistic model constructed by considering the quarterly information of listed companies' ST has obvious time advantages compared with the traditional annually prediction model.It can predict the companies' financial distress in advance for one quarter,which are actually labeled ST at the second quarter,and the accuracy is more than 90%,which can provide timely decision-making basis for investors and regulators.