Research On The Prediction Of Financial Distress Of A-share Manufacturing Listed Companies

Manufacturing has become an important pillar industry of the economic growth in China.Monitoring the financial situation of the manufacturing-listed company and giving timely warning of financial risks are of great significance to the investor,the company and the economic society.This thesis establishes financial distress forecast models to provide early warning of financial risks,where the data is from the information disclosures of manufacturinglisted company,including the financial indicators in the financial statements and Management Discussion and Analysis(MD&A)in the annual reports.The main contents are given as follows.Firstly,we choose the indicators that can reflect the solvency,profitability,operation ability and development potential of the company and test if these indicators have significant difference between the normal company and the ST(Special Treatment)company by using various statistic methods.The indicators are filtered by the significance test and correlation test and only 13 indicators are finally kept for modeling in this thesis.Secondly,we construct the superficial tone indicator and the implicit propensity for default indicator for the quantification of text from the MD&A in annual reports.The superficial tone indicator is obtained by counting the frequency of positive and negative words according to the existing emotion vocabulary.The implicit propensity for default indicator is extracted from the neutral characteristic words with distinction which exist both in the MD&A of normal and ST companies by extension of TF-IDF(Term Frequency-Inverse Document Frequency)method.Thirdly,we establish a financial distress prediction model for the A-share manufacturinglisted company based on the financial indicators and text characteristics by using Support Vector Machine,where the choice of kernel function is discussed in modeling.Empirical results indicate that the financial distress prediction model based on Support Vector Machine with linear kernel achieves 89.36% accuracy on the test set.In addition,we verify the effectiveness of two indicators,superficial tone and implicit propensity for default,and find that these two features extracted from MD&A report provide useful information for the prediction of company’s financial distress.Finally,we build a Logistic regression prediction model for the financial distress prediction of the manufacturing-listed company by using Bayesian method.In particular,we choose normal priors for the unknown coefficients in the Logistic model and derive the posterior distribution according to Bayesian rule.In numerical simulations,the No-U-Turn Sampler(NUTS)is used,the fitting effects and prediction accuracy are compared under different hyperparameters.The numerical results show that using informative priors can avoid overfitting to some extent and improve the model’s prediction ability.