Research On Early Warning System Of Financial Risk Based On Quantile Regression Neural Networks

In these years,financial frequent disarray made social attention focus on financial risks even more than ever.Therefore,constructing financial early warning model has great applied value for preventing financial risks.This paper conducts an in-depth study on China's financial risk early warning.Based on relevant research literature at home and abroad,firstly,in theory,it summarizes the theoretical methods and research results of financial risk early warning in domestic and foreign literatures,and compares the existing domestic and international mainstream financial risk early warning models,then summarizes the advantages and disadvantages of various models.Based on these theories,this thesis uses innovative methods to study China's financial risks.The neural network quantile regression algorithm and other two machine learning algorithms are used respectively for predicting the financial stability status of each quarter of 2018 and the first three quarters of 2019.Through the concrete realization of the model,it combines of theory and practice to study the financial risk warning method in China.This paper selects quarterly data from 2010 to 2017 of 24 indicators to establish an initial indicator system.Based on this,it applies the cluster analysis and nonparametric statistical methods such as ranksum test and Kruskal-Wallis one-way rank variance analysis are used to filtrate indicators,and retains 14 financial warning indicators finally.The k-means clustering and principal component analysis method divide financial risks into four risk states: safe,basic safe,vigilant and dangerous.Then,based on neural network quantile regression model,random forest algorithm and logistic regression algorithm,this paper establishs financial early warning models of China respectively to predict the operation state of the financial system of China.The prediction result shows that the neural network quantile model has higher accuracy than other models and can make a more reasonable explanation for the stable state of China's financial environment.The specific result shows that the financial risks in the third quarter of 2018 and the first quarter of 2019 are in highly vigilant status,which in the second quarter and the third quarter of 2019 is in a dangerous state and deserves attentions,on the other point,Based on the logistic regression algorithm in machine learning,also got the answer that about China's financial stability is insecure in 2018 and 2019.Preventing financial risks should still be an important task for China.This paper innovatively adopts a more time-sensitive indicator system,introducing statistical indicators in the fields of shadow banking and internet finance into the index system,The combination of neural network deep learning algorithm and quantile regression is used to predict the risk,and the scientificity and accuracy of the neural network quantile regression algorithm are verified by two other popular machine learning algorithms.