Research On China's Systemic Financial Risk Monitoring And Early Warning Based On Big Data Method

Historical experience shows that systemic financial risk not only threatens financial stability,but also brings huge losses to the whole macro economy.The outbreak of the international financial crisis in 2008 exposed that the traditional micro-prudential supervision mechanism could not identify and prevent systemic financial risk in the financial system,and it became the consensus of policy makers in most countries to strengthen macro-prudential supervision.The core of macro-prudential supervision is to accurately measure and give early warning of systemic financial risk.However,at present,the measurements based on traditional statistical and econometrical methods are the products of the era of "small data",with limited ability to collect and analyze data and insufficient foresight and accuracy in risk analysis and judgment.It is easy to lead to problems such as narrow vision of financial supervision,late timing of regulatory action and weak strength of regulatory measure.When the epoch of big data has arrived,the promotion and application of big data and big data analytics in various fields also provide important methods and experienced references for the research and practice of preventing systemic financial risk in the financial field.At present,financial regulatory authorities in major developed countries have been actively exploring the application of big data in macro-prudential supervision,especially in the monitoring and early warning of systemic financial risk.At present,China's research and practice in this field are still in the initial stage and need to be further deepened and developed.Based on the theoretical research on the generation,accumulation and diffusion of systemic financial risk,this paper explores the introduction of big data methods into the study of systemic financial risk in China,and innovatively uses big data indicators,machine learning,deep learning,text mining,network analysis and other big data method models.This paper probes into the accurate monitoring,early warning and contagion measurement of China's systemic financial risk in multi-dimension and deep depth,in order to understand China's systemic financial risk more scientifically and systematically,and puts forward the measures to prevent and resolve the systemic financial risk.The research contents and important conclusions include the following aspects:1.The theoretical study of systemic financial risk.This paper constructs the theoretical framework of systemic financial risk,including:(1)The systemic financial risk is defined as a probability that the continuous loss diffusion of the whole financial system leads to collapse,loss of function and even impact on the real economy,it has five characteristics such as universality,negative externality and contagion,and has two dimensions of time and space.(2)The theoretical basis of systemic financial risk is established by integrating Marxist money and credit theory,"debt-deflation" theory,financial instability hypothesis,asset price fluctuation theory,information asymmetry theory and financial cycle theory.(3)This paper expounds and analyzes the three-stage evolution mechanism of "accumulation-explosion-diffusion" of systemic financial risk.(4)Summarize the causes that lead to the occurrence of systemic financial risk include financial fragility,excessive financial liberalization,correlation of financial system,irrationality of market subjects and moral hazard in financial market internal influencing factors of financial system,macroeconomic cycle and policy regulation errors external influencing factors of the financial system.(5)From the theoretical level,this paper puts forward the transmission mechanism of the formation of systemic financial risk and its impact on the real economy.2.Research on real-time monitoring of China's financial stress.This paper holds that the deviation of economic subjects' emotional psychology and behavior is the essential root of the endogenous nature of systemic financial risk.Therefore,the financial stress index constructed from the perspective of economic subject emotion is used as an indicator to measure the level of systemic financial risk in the time dimension.By integrating big data and traditional financial statistical indicators,this paper constructs China's Financial Stress Index(FSI)with daily frequency based on the mixed frequency dynamic factor model,and investigates the time-varying characteristics and prediction effects on macroeconomic variables of financial stress index.The results show that :(1)The constructed FSI can accurately measure and monitor China's systemic financial risk.The phased change characteristics of the index and the identification results of regime state are highly consistent with the actual evolution of China's systemic financial risk.(2)From the parameter results of FSI estimation model,the uncertainty of expectation and risk perception of economic subjects measured by traditional financial indicators and big data indicators are positively correlated with financial stress.Financial stress has strong sustainability.Adding big data information can significantly improve the estimation of FSI model.(3)The out-of-sample prediction results of macroeconomic variables such as output and inflation show that FSI has good forecasting ability for macroeconomic variables,and the introduction of big data indicators can improve the forecasting effect of FSI.(4)Based on staged data and all data,the trends of FSI are basically consistent,and the robustness of the estimation model is relatively good.3.Research on early warning of China's systemic financial risk.Based on K-Nearest Neighbor,Random Forest and Support Vector Machine and other machine learning models,as well as deep learning models such as Long Short-Term Memory neural network and Gated Recurrent Unit neural network,this paper carries out early warning research on China's systemic financial risk.Using the model performance evaluation methods such as confusion matrix,ROC curve and cross validation,this paper finds the best early warning model for predicting China's systemic financial risk.Using the SHAP interpretation technology of machine learning model,this paper identifies the main causes of pushing up the level of China's systemic financial risk.The results show that :(1)The machine learning and deep learning algorithm models are obviously superior to the traditional econometrical model of Logistic Regression in terms of various performance metrics and prediction results.(2)When dynamic warning of systemic financial risk is carried out,the warning performance of Random Forest model is better than that of other machine learning models and deep learning models.(3)Using SHAP machine learning model interpretation technology,it is identified that the main causes of raising the level of China's systemic financial risk includes financial institutions,foreign exchange market,shadow banking,real estate market and policy intervention factors.(4)The prediction results of the Random Forest early warning model with the best performance show that the probability of systemic financial risk in China in the next year is small.4.Research on contagion of systemic financial risk in China's financial system.In this paper,the machine learning Gaussian Graph Model(GGM)based on Bayesian estimation is established.The partial correlation coefficient estimated by the model is used to construct a financial network reflecting the correlation between financial institutions,and the contagion of systemic financial risks in the whole financial network is analyzed.At the same time,the model estimation results are used to construct an index system to compare the degree of systemic importance of each financial institution,and to investigate the evolution trend of systemic financial risk in the financial system.The results show that :(1)Most financial institutions in the current financial risk contagion network in China are connected with other financial institutions to some extent,and systemic financial risk can be transmitted among different financial institutions.(2)From the perspective of contagion channels of systemic financial risk,the contagion channels based on investor sentiment and tail risk are relatively obvious,while the contagion channel based on financial market is relatively insignificant.(3)In different types of financial institutions,the risk contagions of securities companies,urban commercial banks and large state-owned commercial banks are relatively strong,and the risk contagions of large insurance companies can not be ignored.(4)Considering the asset scale,correlation and financial network effect,large state-owned commercial banks are the most systemically important,followed by joint-stock commercial banks.However,large insurance companies,local urban commercial banks and head securities companies whose associated financial institutions have large assets also have certain systemical importance.(5)The sum of partial correlation coefficients in different periods estimated based on the model reveals the evolution trend of systemic financial risk in China's financial system in recent years.The systemic financial risks in different intervals are more consistent with the operation of China's internal and external economy and the operation of the financial system.5.Research on the contagion effect of regional systemic financial risks in China.This paper also extends the research perspective of systemic financial risk to the regional level,uses the big data network analysis method to investigate the spatial correlation network characteristics and contagion of China's regional systemic financial risks.Combined with the characteristics of network structure,and uses the SIRS model on scale-free network to numerically simulate the epidemic evolution of risk in each province after the occurrence of extreme regional systemic financial risks.The results show that :(1)The regional systemic financial risks among various provinces in China show a typical network pattern,with obvious "scale-free characteristics" and "small-world phenomenon".In recent years,the correlation between systemic financial risks in various provinces shows a trend of gradually strengthening.(2)At present,Qinghai,Ningxia and other provinces with remote geographical location and slow economic development have become the concentrated areas of risk spillover due to large in-degree centralities of vertex.Jiangsu,Beijing and other provinces with superior geographical position and rapid economic development have become the main regions of risk spillover due to large out-degree centralities of vertex.Shanghai,Tianjin and other provinces have become important paths of risk transmission due to their betweenness centralities and Page Rank rankings.(3)Each province can cluster into different block in the regional systemic financial risk contagion network.Nine provinces such as Jiangsu and Beijing are clustered as "net spillover block" that is prone to outward spillover effect,four provinces such as Tianjin and Inner Mongolia are clustered as "broker block" that plays the role of "bridge" and "intermediary" in risk contagion networks,and seven provinces such as Shanghai and Heilongjiang are clustered as "two-way spillover block" that has two-way risk spillover effect on intra block and other blocks,nine provinces such as Qinghai and Ningxia are clustered as the "net overflow block" that is the center of risk agglomeration.(4)Numerical simulation shows that extreme risk will spread rapidly among provinces within a short time after occurrence.Compared with the occurrence of extreme risk in a single province,the time of transmission of extreme risk in multiple provinces is shorter.6.Research on the prevention and supervision of China's systemic financial risk.This paper holds that establishing and improving the macro-prudential policy framework and strengthening the collection,analysis and utilization of financial big data are not only two important aspects of strengthening macro-prudential supervision,but also the main means to prevent and supervise systemic financial risk in the future.This paper theoretically combs the basic framework of macro-prudential policy,summarizes the practical experiences of macro-prudential supervision based on big data in countries such as the United States,the United Kingdom and China,and obtains corresponding enlightenment.Finally,combined with the econometrical conclusions of this paper,put forward countermeasures and suggestions for preventing and supervising China's systemic financial risk in the future from the perspectives of macro-prudential supervision and big data,including: establishing a macro-prudential supervision big data platform conducive to guarding against systemic financial risk,actively studying and applying big data cutting-edge analytics to effectively monitor and warn systemic financial risk,improving China's macroprudential policy framework system in combination with the new situation of current financial development,focusing on strengthening the prudential supervision of systemically important institutions to guard against systemic financial risk contagion and preventing regional systemic financial risk contagion from evolving into global systemic financial risk.The innovations of this paper mainly include :(1)In terms of monitoring of systemic financial risk,this paper adopts Internet search big data to enrich the risk perception indicators of economic subjects from the perspective of behavioral finance,and expand the data and information sources for monitoring and analyzing systemic financial risk.By using mixed frequency dynamic factor model and effectively utilizing all kinds of high and low frequency data,the frequency of financial stress index is increased from the quarterly,monthly and weekly frequency of most existing literatures to the daily frequency,realizing the real-time and accurate monitoring of systemic financial risk.(2)In terms of early warning of systemic financial risk,this paper uses "the uncertainty of models" frontier concept,by integrating and building a variety of machine learning and deep learning models for early warning of systemic financial risk in China,and finds the optimal forecasting model by using methods of model performance evaluation,greatly improved the accuracy and effectiveness of systemic financial risk early warning in our country.The machine learning and deep learning model interpretation technique is introduced into the study of systemic financial risk,and the important factors that push up the level of systemic financial risk in China are scientifically inferred.(3)In terms of measuring the contagion of systemic financial risk in the financial system,this paper uses crawler technique to crawl Internet text information,and uses text mining,kernel principal component analysis to fuse with traditional structured financial data,which makes up for the lack of application of unstructured Internet text data in existing relevant studies.The introduction of Gaussian Graphical Model based on Bayesian estimation provides a new method for the study of the contagion of China's systemic financial risk.(4)In terms of the measurement of the contagion of regional systemic financial risk,this paper expands the latitude of establishing the index system of regional systemic financial risk and optimizes the index synthesis method.This paper comprehensively uses traditional and emerging indicators and models in big data network analysis to analyze the contagion effect of regional systemic financial risk from multiple perspectives and at multiple levels,improving research methods and enriching research perspectives.The introduction of SIRS model in scale-free network provides a new powerful tool for regional systemic financial risk research.