Research On Financial Crisis Early Warning Of Information Technology Industry In China

Information technology industry is one of the most important industries in China's national economy.Its main characteristics are capital concentration,strong innovation ability and high risk coefficient.At the same time,as the barometer and firewall of enterprise development,financial risk is a subject that all enterprises need to explore.If a reasonable and effective early warning model of financial crisis is established,enterprise managers can timely understand the financial and operational status of enterprises and make a prediction of financial crisis.Financial crisis prediction is an important field of financial analysis.The famous Z-socre model is the earliest financial crisis prediction model proposed.The Z-score model is established by five variables.However,the model is not sensitive to market changes and the weight needs to be adjusted frequently.After Z-score model,conditional probability model is applied to financial crisis prediction.With the wide application of machine learning methods,more and more scholars use machine learning methods to build financial crisis prediction models in recent years.However,few scholars consider class imbalance when using machine learning models,and some basic machine learning models have problems of over-fitting for data with more variables and fewer samples.The research object of this dissertation is listed companies in the information technology industry of the Shenzhen and Shanghai Stock Exchanges in China.Enterprises are classified into two categories according to the classification criteria of the financial crisis of listed companies of the Securities Regulatory Commission(the stock types are ST,* ST,SST,S*ST,S collectively referred to as ST stocks).Using the financial crisis situation of the company from 2015 to 2018 as the research object,selecting the corresponding T-2 years financial data and non-financial data as indicators,screening the indicators through differentiation analysis and correlation analysis,and finally using Twenty-two financial indicators and three non-financial indicators to establish a financial crisis early warning model based on XGBoost algorithm and LightGBM algorithm.Finally,the AUC value predicted by LightGBM algorithm model is 86%,and that predicted by XGBoost algorithm model is 83%.The results are similar,but the precision and recall of the financial crisis early warning model based on XGBoost algorithm are higher than that of LightGBM algorithm.At the same time,compared with the financial crisis early warning model constructed by decision tree,random forest and GBDT algorithm,it is found that XGBoost algorithm and LightGBM algorithm have better effect.In terms of running time,the training speed of LightGBM algorithm is nearly ten times faster than that of XGBoost algorithm.From the order of characteristic importance of the two models,interest guarantee multiple,liquidity ratio,business income growth rate,liquidity ratio,the proportion of R&D personnel,the top ten shareholders' shareholding ratio are all in a relatively high position of characteristic importance.