The Study On VaR Gaussian Process Autoregression Model Of Financial Data

Sound management and control of financial risks are of great significance to maintaining the stability of the global financial system and promoting the growth of the real economy.At present,the main measure index of financial risk in the world is VaR,which is a quantile of the distribution of the amount of certain risk loss.However,due to the characteristics of financial data,there is still no stable and effective method for the effective measurement of VaR.The traditional VaR measurement method is mainly divided into parametric method and nonparametric method.The parametric method makes certain assumptions about financial data,which is difficult to meet.The non-parametric method also makes limited use of data,which is easy to cause the phenomenon of "ghost effects".This paper starts from solving the defects of the traditional model,combines the gaussian process autoregression theory suitable for financial data processing,and establishes a new model that can accurately predict VaR,so as to solve the difficult problem of financial data modeling and enable financial institutions and relevant regulatory authorities to more accurately measure financial risks.This paper firstly summarizes the research status of financial risk management and gauss process theory.Secondly,it introduces the basic principles of three traditional VaR models: historical data simulation,normal distribution,monte carlo simulation and gaussian process autoregressive model.Thirdly,this paper introduces the source of the experimental data set,the calculation of the rate of return and the parameter selection of the model.Finally,the above models and data are used for modeling comparison.The empirical results show that the gaussian process regression method in this paper is the best way for the forecast of the VaR.Compared with three kinds of traditional measuring,gaussian process regression method can get more accurate VaR estimates,and the predictive results of the traditional method is less accurate.The actual data of the test results also show the actual data does not meet the requirements of traditional parameter method of hypothesis.Thus,the gaussian process autoregression method is better than the traditional VaR estimation method to some extent.