非线性协整关系的存在性研究

Research on the existence of nonlinearly cointegrated relationship

运用神经网络技术 ,提出了一种估计和检验非线性长记忆时间序列之间的协整关系的方法 ,并通过仿真试验说明了所提方法的实用价值 .同时 ,根据马氏过程的遍历理论 ,对网络输出残差的平稳条件进行了研究 ,证明了网络输出平稳残差的充分条件 。

In this paper we develop a new estimating and testing procedure on the cointegrated relationship among nonlinear long memory time series by the means of neural network technique. Monte Carlo simulation is also presented to illustrate the application value of the new method. Furthermore, based on the ergodic theory of Markov process we analyze the stationary distribution of network residual series, and prove the efficient conditions under which the unique asymptotic stationary distribution exists.