Research On Specification Test Of Non-stationary Data Modeling

Model specification plays an important role in econometrics and is the core of econometric modeling.In the process of econometric modeling,correct model specification is the premise of research.If the model is set correctly,the estimation and test results of the model will be meaningful.Once the model is set incorrectly,that is,the model is set inconsistently with the data generation process,using models to study real economic problems will be meaningless.Therefore,if we can test the specification of the model before drawing a conclusion,we can improve the application value of the model.The research of specification test for stationary data modeling is mature.However,with the acceleration of reform and opening up,the process of economic operation has become very complex,and most of data in real life are non-stationary.Therefore,it is urgent to test the specification of models in non-stationary data.Firstly,the basic theory of non-stationary data modeling is discussed,and the modeling methods of different data generation processes are given.The non-parametric model is analyzed,and the different estimation methods and the selection of smoothing parameters are analyzed.Secondly,I study the test of co-integration relationship among non-stationary data in this paper,and give a good diagnosis method for the false regression model in non-stationary data.Finally,based on the parameter model specification test method for non-stationary data,I extend it to the specification test for semi-parametric models,further expanding the research scope of non-stationary data modeling.In the empirical aspect,the research results of the paper are applied to the study of the relationship between foreign economy and inflation.The specification test method of non-stationary data modeling is used to diagnose the econometric model between inflation and its influencing factors,which provides a theoretical basis for the study of real economy.Through specification test research on non-stationary data modeling,I came to some valuable conclusions in this paper,and its innovations are mainly reflected in the following aspects.Firstly,using Monte Carlo stochastic simulation method,the asymptotic distribution of local DW statistics proposed by Phillips is explored,and the rejection domain and Partial Critical Value Table of its test statistics is obtained.Secondly,a non-parametric false regression identification method is proposed by exploring the variation of goodness of fit with smoothing parameters using non-parametric method.Thirdly,I extend the method of parameter model specification test under non-stationary data to the specification test of semi-parametric model,which lays a foundation for further study of semi-parametric model.