A Research On The System Network Of China Stock Market Based On Lasso VAR Model

Investors had suffered a lot in the exploring of systemic risk such as the financial crisis of America in 2008,the sovereign debt crisis of Greece in 2009 and the crash of the stock market of China in 2015.With globalization and the fast developing of global finance,the relation between financial institutions and listed companies has become more and more complicated,and this leads to a more vulnerable market,in which systemic risk can spread widely,and become more contagious and devastating.To recognize systemic risk and its contagion path is meaningful for governments to implement risk regulation which can make sure that financial markets develop more stably.In the current researches of systemic risk,traditional ways to measure systemic risk such as MES,CoVaR and pairwise VAR model,used by a lot of researchers had shown some meaningful results,and turn out to be effective to recognize systemically important institutions through measuring their systemic risk contribution to the whole system.In China,there are also some researches aimed to measure systemic risk with MES,Va R and CoVaR.However,there has been some limitations,and the most important one is that these measures do not come along with measure of the Granger causality between institutions among the system,and this leads to uncertain network construction of system,and uncertain relationship between systemic risk and market system network.Based on lag-1 multivariate VAR model,instead of using OLS solution,we use Lasso solution to do the regress and force the original model to be sparse,so that it can be applied into high-dimensional data field,and after using Lasso VAR model to estimate the network of sample firms,we bring the bias correction of Lasso VAR estimates.After that,we use FDR based on BH procedure to do the hypothesis testing aimed to judge whether VAR estimates are significant.First,we estimate Lasso VAR model on simulated data and contrast it with other measures of systemic risk such as MES,CoVaR and pairwise VAR model.Based on the performance of different measures in the recognizing of systemically important institutions and network connectivity.Next,we use monthly returns data of stocks of CSI300 from January,2012 to March,2018 to be the sample of empirical application,in which we estimate Lasso VAR model and pairwise model with regard to financial institutions,non-financial firms.With performance comparison of two models,we discuss about the application of our Lasso VAR model in different industries,and whether its accuracy is acceptable,and based on the results,we try to recognize systemic important institutions and measure the contagion of systemic risk,and in the meantime discuss the relationship between systemic risk and system network construction.The conclusions drawn from the study include:(1)Despite that Lasso will bring bias to the model estimation,Multivariate VAR model still have acceptable accuracy in recognizing the construction of system network with debiasing method,and Lasso VAR model can simplify the system network connectivity which is helpful to recognize systemically important institutions.(2)With being sparse,there are still a lot of strong connections in the system network,which can be a help of contagion of systemic risk.(3)Before the exploring of systemic risk,the system network may change significantly,it will becomes more complicated,and the average degree and average closeness of system network will increase significantly.(4)Lasso VAR model can effectively recognize the Granger causality between two different firms with consideration of the other firms in the system,which will lead to the result that their monthly returns show similar fluctuation at the same time.(5)The network degree and closeness of financial institutions and non-financial firms increased significantly during the crisis,the exploring of systemic risk will influence all parts of the stock market.The main innovations of this paper are as follows:(1)In this paper,we use Lasso to force weak relationships among firms in the network to zero,allowing us to take a true system-wide approach in estimating the multivariate VAR model with limited data,which can be useful in high-dimension data field.(2)Using Lasso will bring bias to the model,which can be potentially large in a finite-sample setting.Also,with a bias model the model estimates do not come with any measure of uncertainty.That is,we can not tell whether the estimates are significant.In this paper,we use a methodology proposed by other researchers to bias-correct the VAR estimates so as to draw statistical inference.(3)We apply the Lasso VAR model to the CSI300 which is typical,so as to estimate the network construction of part of the whole system,and draw some points about the recognizing of systemically important firms and the systemic risk contagion of stock market of China.(4)We use degree and closeness to measure the network construction,and we find out that these two indexes will increase significantly during the crisis.Due to the limited level of research,this paper still has some shortcomings:(1)The sample size is too small,we only use monthly returns data of stocks of CSI300 from January,2012 to March,2018.(2)We estimate the network construction only based on the simplest first order delay relationships without consideration of multi delay.(3)We only divide the sample into two part: financial institutions and nonfinancial firms.In fact,different industries may have differences in the network construction.