A Study On The Specification Of STAR-GARCH Model

In the past 30 years,linear models have been widely used in the field of economic research,which theory tends to be mature.However,due to the limitation of linear model,it is hardly to describe the special economic change situation.With the rapid development of economy,the structure of macroeconomic and financial time series has changed,and the empirical data imply that most of the important data structures show nonlinear characteristics.In order to describe the economic process accurately,the nonlinear model has gradually become a popular research topic in econometric theory.The smooth transition autoregressive(STAR)model has become one of the most widely used nonlinear models because it can better describe the phenomenon of "mean reciprocating".In the discussion of economic fluctuation,the generalized conditional heteroscedasticity(GARCH)model is very popular.At the same time,the STAR-GARCH model is a very useful tool for describing the nonlinear characteristics of economic series.In the past research,the test of the stability of stochastic process under the STAR-GARCH model is immature.Therefore,the research on the STAR-GARCH model has some theoretical significance.This paper integrates and combs the existing theoretical results of the STAR-GARCH model.Based on the study of the overall specification of the model,this paper focuses on the actual power of the test and the empirical size of the unit root test statistics of the time series under the STAR-GARCH model.According to the ESTAR-GARCH model and the LSTAR-GARCH model,this paper introduced the Wild bootstrap method to revise the KSS and tlstar statistics and constructs two new unit root test statistic KSSWB and tLSTARWB statistics for the existence of conditional variance of stochastic term under smooth transition autoregressive process.The modified STAR-GARCH model is used to analyze the dynamic characteristics of RMB exchange rate since 2015,and to make an effective prediction of RMB exchange rate by using the derived model.The innovations of this thesis include:(1)This paper compares the actual test efficacy and Test level of DF,KSS and KSSWB and Df,tlstar and TLSTARWB unit root test statistics by Monte Carlo simulation respectively.Results show that KSSWB and TLSTARWB have better actual test efficacy in different sample sizes and different degrees of heteroscedasticity,and there is no obvious distortion in their experience level compared with two other test statistics.This makes the relevant theory of STAR-GARCH model more completely.(2)This paper applies the LSTAR-GARCH model to the study of the nonlinear dynamic characteristics of RMB exchange rate since 2015.The results show that the RMB exchange rate has obvious nonlinear fluctuation characteristics since the exchange rate reform of the China,and the LSTAR-GARCH model can depict this kind of nonlinear change pattern,compared with the traditional AR model,the LSTAR-GARCH model has accurate prediction ability and more likely to recognize outliers.