Research On Forecasting Methods Based On Threshold Cointegrated Systems

For some nonstationary time series process, on the whole regime, there are no linear cointegrating relationships among them, but there exists nonlinear cointegrating relationships. The author deals with the problem of nonlinear cointegration by using the concept of threshold cointegration, divides the whole regime into several sub regimes by testing, and carries on the linear cointegration analysis on each sub regime to solve the problem of nonlinear cointegrating. In context, the author do lots of research on the testing of threshold cointegrating, the parameter estimation of threshold vector equilibrium model and its application in economic forecasting. Beside, the author especially discussed nonlinear cointegrated systems with structural changes, which is one kind of threshold cointegrating. When the threshold parameters are defined as time variables, threshold cointegrating become nonlinear cointegrated systems with structural changes. Because of the changing in politics, economics, culture, technology and so on, there widely exist nonlinear cointegration systems with structural changes. The superiority of the threshold cointegration on forecasting the nonlinear cointegration systems with structural changes was verified by analyzing the 180-share index of Shanghai Stock Exchange and volume of transaction. This paper also explains the rolling forecast methods, and proposes a method to evaluate the models.