Research On Setting Nonlinear Time Series Model

The linear model, whether the theory development or the model setting, has been relatively mature. But with the development of economic, more and more economic phenomenon shows the nonlinear characteristics. People pay more attention to the nonlinear model setting and theory research. However, any kinds of model have their specific existing significance and application value. It's not more complex better. Therefore, it's quite important to set a good model when we modeling. There are many kinds of nonlinear model,but the modeling method is the same as the linear model, following the same rules. Last century, Hendry has proposed a modeling method named "From the general to special", which is also suitable for the nonlinear modeling. The econometric modeling is based on the economic theory, but the method only effects on the analysis of the economic relationships, not directly roles in the model setting. A series rigorous mathematical reasoning and statistical reference base on setting model correctly. Therefore, this paper sets from the modeling and discusses the application value. It's very important and valuable.The nonlinear time series model has many types, this paper only research on one kind of the model-Smooth transition autoregressive model (STAR). Firstly, we introduce the purpose and significance of the study. and the status of the nonlinear time series model, especially the STAR model. On the basis, we make emphasis on the form of the STAR model and the method of estimation and inspection. This paper set two angles to research the setting of model:one is pay attention to the reliability of the nonlinear test by using computer simulation; the other is using the economic analysis to examine the result of the STAR model.In the research of reliability of the nonlinear test, this paper used two ways of data generated process:one is the STAR model with exponential transition function:and the other is the autoregressive model (AR). Then we could check the reliability of the traditional nonlinear test and the improved method with Akaike and Schwartz information criterion in different samples. In the empirical aspect, This paper mainly used the real exchange rate of RMB to research the modeling of STAR. It's not only check the application of the purchasing power parity in China, but also provide some useful information for the nonlinear modeling.The innovation of this paper manly displays in using the Monte Carlo method to compare the reliability of different inspection methods. Meanwhile, we check out the influence to the methods when the residual series has heteroscedastic and serial correlation. It provides the theoretical support for the application of nonstationary STAR model. At the same time, the study applies the model to the real economic, combines theory and practice, starts with the purpose of modeling and establishes the start with the purpose of modeling.