Finance Systemic Risk Measurement Based On Seemingly Unrelated Regression Models

The maintenance of financial market stability and the avoidance of large-scale financial system risk events constitute one of the most critical dimensions confronting our regulatory authorities.In a glance at economic market research methods,it is difficult for conventional linear correlation studies to accommodate the increasing complexity of economic markets.As such,research on statistical methods that can handle high-dimensional time-varying data has gained increasing importance.Through the graphical model structure analysis of the residual covariance array of the seemingly uncorrelated regression model,this paper presents the sparse SUR model and designs the MCMC algorithm.On the basis of consolidated conditional independence,the method can effectively address the structured problem of high-dimensional data.Numerical simulations demonstrate that the parameter estimation of the new method outperforms the traditional SUR model given the small sample scenario.Eventually,the theory is applied to the Fama-French five-factor model of the U.S.securities sector index to measure the risk of each factor on the stock sector index.It is evident from numerical simulations that the GSUR model provides superior estimation as compared to the conventional SUR model,in particular for small sample sizes.The results of GSUR-based variable selection reveal that the Fama-French five factors are highly applicable to the U.S.stock sector index,with significant effects of market factor,size factor,book-to-market ratio factor,earnings factor and investment factor.In conjunction with the Markov regime switching graphical model(MSGSUR),a graphical model approach based on regime switching is developed,whereby the algorithm is designed through a Monte Carlo simulation method.Furthermore,the method is implemented to measure dynamic systematic risk in China's Shenzhen stock market.As evidenced by the empirical results,systematic risk is time-varying and there are two significantly distinct graph structures for the industry index residuals,representing varying high and low conditional correlations and a relatively high probability of state continuity.