Research On China’s Systemic Risk From A Global Perspective

Since the 2008 financial crisis,there has been increased attention paid to systemic risks by global economy.In recent years,the combination of the century-old changes and the covid-19 pandemic has brought the world into a new period of turbulence and change.To address the economic downturn caused by the pandemic,various stimulus policies have been introduced by countries,and friction among the major global economies is intensifying,putting huge pressure on global capital markets.In China,data shows that over the past five years,the country has dealt with over 120 billion yuan in non-performing loans,far exceeding the total for the previous 12 years.Nearly 5,000 P2 P lending institutions have stopped operations.Thus,measuring,warning,responding to,and resolving potential systemic risks is imperative.The profound lesson learned from the 2008 financial crisis is that there is a long cumulative process in the early stages of systemic risk.Therefore,based on the current economic situation,preventing and resolving systemic risks remains an important and critical task,both now and in the future.However,the measurement of systemic risk faces many difficulties,such as the lack of a timely warning mechanism and significant differences in measurement methods.As a result,the measurement and warning of systemic risk is also a widely discussed topic among scholars.The SRISK model proposed by Brownlees and Engle(2017)is a quantitative model for evaluating systemic risk in the financial system.It is based on asset price data and company capital structure information and measures the contribution of individual companies to systemic risk by calculating their capital shortfall during market downturns.By summing the SRISK values of individual companies,the overall risk of the system can be obtained.In addition,Brownlees and Engle(2017)found in their research that the model can effectively identify systemic risk companies and provide accurate systemic risk warnings.The SRISK model focuses on the negative externality of individual economies in the overall economic performance and has good performance in individual systemic risk measurement and overall systemic risk warning.Thus,it has an important position among all systemic risk measurement models and is widely adopted in academia.Systematic risk not only stems from internal companies within the system,but also from external economic environments and other market factors.However,traditional methods of systematic risk assessment are inherently flawed,such as difficulty in utilizing more data resources and rigid modeling approaches,thus making it difficult to consider external market indicators,such as investor sentiment.Based on this,the present study improves upon traditional systematic risk models,building upon the SRISK model proposed by Brownlees and Engle(2017).This paper selects 128 Chinese A-share listed financial companies from 2011 to 2021 as samples,introduces exogenous macroeconomic factors and investor sentiment constructed by text information to calculate marginal expected loss MES and systemic risk SRISK indicators of individual companies.Compared with the original SRISK model,the main improvements in this paper are as follows: 1.In the definition of systemic events,fixed threshold is changed to time-varying VaR.2.In terms of the measurement of time-varying VaR,the traditional GARCH model is improved to EGARCH(SENT)model,where SENT is an investor sentiment index constructed from the keywords related to the stock market obtained from Weibo platform.3.When estimating marginal expected loss MES and systemic risk SRISK under systemic events,the traditional GARCH-DCC model is improved into a machine learning model.The benefit of machine learning models is the ability to include exogenous economic factors in estimating systemic risk.4.This paper also extends the research perspective to the international market.By collecting the stock index data of 73 major countries in the world,this paper evaluates the contribution of major countries or regions to the systemic risk of the Chinese market,and uses the hierarchical relationship of machine learning model to embed the stock index information of major countries into the measurement of systemic risk.To explore the impact of different countries and regions on the systemic risk of the Chinese market and the transmission relationship.Finally,based on the dynamic correlation coefficients of stock indexes of major countries in the world,this paper draws a network diagram of major countries in the world to explore the source of systemic risk and the relationship between risk contagion.The main conclusions of this paper include: 1.Compared with the traditional fixed threshold definition of systemic events,the SRISK index calculated by the dynamic VaR index introduced into investor sentiment in this paper is more effective.2.Based on the calculated VaR-SENT indicators,MES and Srisks estimated by the machine learning method in this paper are stronger than the GARCH-DCC model in describing systemic risks.3.Compare the four machine learning models selected in this paper(SVR,Adaboost,ANN and LSTM),and the model performance is ranked as LSTM >ANN >Adaboost >SVR.4.This paper expands other company characteristics into the machine learning model and finds that introducing complex company characteristics does not further improve the performance of the machine learning model.5.By studying the distribution characteristics of systemic risk in various financial institutions,we find that the level of systemic risk: securities industry > Banking industry > Insurance.6.This paper uses the machine learning model to calculate the characteristic importance of 73 major countries and regions on the return rate of the Chinese mainland market,and finds that Chinese Hong Kong and the US market have the greatest impact on the Chinese mainland market.By using the multi-layer structure characteristics of the machine learning model,the fourlayer and three-layer network structures of the United States-Chinese Hong Kong--Chinese mainland--Chinese mainland companies and Chinese Hong Kong--Chinese mainland--Chinese mainland companies are embedded.In this paper,we find that in the ability to characterize systemic risks,the three-layer network structure > Four-layer network structure.Finally,based on the dynamic correlation coefficient obtained by GARCH-DCC model,this paper constructs the global systemic risk transmission network.According to the node strength information,this paper obtains that the United States,China and the United Kingdom have a significant impact on the global stock market.This network conforms to the "small world theorem".