Improvement And Research Of Regime Switching Transformation Model

Financial data usually have structural mutations and state changes over time.The linear model may not be able to perform the variation trend of data well.Therefore,it is necessary to discuss other types of models.The regime switching transformation model is one of the important models to study this kind of problems,which can accurately describe the volatility characteristics of financial data under multiple mechanisms.The regime switching transformation model has been widely used in time series analysis,most of the models introduce two mechanism structures,and use the state transition function to determine the specific mechanism structure of the time series in the economic cycle.Hamilton introduced Markov process into auto-regressive model to build MS-AR model in 1989,and Kim made a more in-depth analysis of this model in 1994.Subsequently,many researchers introduce Markov transformation into more general models,such as regression model and wave model.However,these models have some limitations.For example,when setting the model,the residual error of each state is required to follow the normal distribution with zero mean value.Since the density function of financial data usually presents incorrect biased distribution,the estimation parameters of the model by traditional estimation methods may be distorted.Non-parametric method is a statistical method in the case of unknown population distribution,which can accurately predict the asymmetric data,and its operation efficiency is greatly improved compared with the traditional econometric model.In order to study the volatility of financial data under different mechanisms,the framework of STAR model and Markov model is discussed in detail,and the methods of model identification,parameter estimation and verification are given.The rationality of the model are discussed through empirical research.Therefore,the research on multi-mechanism model has theoretical significance and application value.Financial Risk is one of the important research issues in financial markets.Faced with the uncertainty and volatility of the financial system,the VaR(Value at Risk)model can be built to measure the future financial risks and provide a basis for financial risk managers to make decisions.However,during the construction of this model,it is usually assumed that the rate of return follows the normal distribution,and the financial data often do not meet this assumption,which brings errors to the estimation of the model.It is necessary to use the method of non-parametric statistics to improve the model,so that it can meet the needs of practical problems.This paper creatively proposes a VaR model estimation method based on non-parametric kernel density,which provides a new tool for studying the volatility of financial data.The thesis is divided into three aspects to study the financial risk model.Firstly,the STAR model is studied,and the construction form and parameter estimation method of STAR model are given.For financial data with high aggregation characteristics,this paper creatively presents a non-parametric STAR model estimation method.Secondly,in order to study the volatility characteristics of financial data,this paper tries to combine the STAR model with the GARCH model,obtain the STAR-GARCH model and give the corresponding estimation method.Aiming at the research of volatility of financial data under different mechanisms,this paper introduces Markov mechanism transformation model,and combines it with GARCH model to get MS-GARCH model,which can better predict the volatility trend of financial data.Finally,this paper improves the financial risk measurement model(VaR model),which can estimate the maximum loss that asset portfolio or economic value may suffer under the condition of given confidence level.Traditional VaR model requires portfolio return rate data to follow normal distribution,but in practice,it is difficult to achieve.This paper puts forward a new method based on non-parametric VaR value calculation method,and uses the method of non-parametric kernel density to estimate the distribution of the innovation,and then estimates VaR value on this basis,so as to obtain a more accurate estimate of stock return rate.The innovations of this paper are as follows:first of all,by constructing non-parametric and semi-parametric Markov models,we give the derivation methods of bandwidth,state transition probability and filtering probability of these models.In this paper,the model is used to model the CSI 300 stock index,and the state transition process of CSI 300 stock index is predicted accurately under the condition that the random disturbance term does not obey the normal distribution.Compared with the traditional method,the iteration rate of this method is greatly improved.Secondly,the VaR model based on non-parametric kernel density is constructed,and a new method of risk measurement for financial data is presented.Compared with the traditional VaR model prediction method,this method has lower requirements on data,and the predicted results are more accurate.Thus,it provides more powerful tools for financial risk managers to carry out practical work.At last,This paper discusses the framework of STAR model and Markov model in depth,gives the corresponding model identification,parameter estimation and verification methods,and combines these two types of models with GARCH model to form the STAR-GARCH model and Markov-GARCH model.This paper provides a new method to study the volatility of financial data.Along with the changes in the world economic structure,a large number of economic variables show nonlinear characteristics.Sometimes the traditional linear model can not adapt to the changes of the mechanism,and thus can not accurately describe the economic phenomenon.The use of nonlinear models to describe economic phenomena can better reflect the generation process of economic data.Through further study of STAR model and Markov model,this paper finds out an effective modeling method to deal with the data that has different kinds of structures,and combines it with non-parametric statistics theory,so that the model can describe the volatility characteristics of the data more accurately,which can help decision makers to develop a reasonable investment strategy.