Non-linear transformation tim-series models and causal analysis

本研究提供了一种表征多变量间依赖结构的方法。以平稳ARMA模型为研究对象，提出了一种对部分因果测度进行统计估计和检验的数值可行方法。该方法虽然主要针对平稳ARMA模型，但在协整时间序列中具有广泛的适用性。本研究在大规模蒙特卡罗实验的基础上，对所开发的相互依赖测度的插件估计方法和模型参数的三步最大惠特尔似然估计方法进行了小样本性能仿真研究。该研究还为美国宏观经济和金融经济的实证研究做出了贡献。特别是对实际GDP、M2、CPI的期限蔓延与增长率之间的因果结构进行了深入研究，为相关领域的文献增添了新的见解。

This research provides an approach to characterize the dependency structure between multivariate variables. Focused on the stationary ARMA model, the research produced a feasible way of numerically feasible method of conducting statistical estimation and testing of partial causal measures. Although focused on the stationary ARMA model, the approach has wide applicability in cointegrated time-series. The research conducted simulation study to evaluate the small-sample performance of the developed plug-in estimation method for the interdependence measures on the basis of large-scale Monte Carlo experiments and also the three-step maximum Whittle likelihood estimation procedure for the model parameters. The research also contributed in an empirical study on the U.S. macro and financial economy. In particular, intensive study was made on the causal structure among the term spread and growth rates of real GDP, M2 and CPI, adding a new insight to the literature of the allied field.