摘要
提出了将自助法(bootstrap)的不同变体应用于平滑转移自回归模型的线性检验,并通过蒙特卡罗实验分别考察其在误差项独立同分布和存在序列相关时的有限样本性质.重点研究了非线性参数和序列相关系数对检验水平和功效的影响.实验结果表明,基于自助法的线性检验在各样本容量下都具有更高的功效,并且可以很好地纠正基于极限分布理论的LST统计量的水平扭曲.本文还详细介绍快速了两阶段自助法(FDB)的基本思想和实现方法,模拟实验证明它比基本自助法具有更好的稳健性和收敛性.
This paper applies Different variants of bootstrap methods to the linearity test of smooth transition autoregressive model and studies their finite sample properties under independent identical distribution and serial correlation through Monte Carlo experiments. The paper also puts emphasis on the impact of nonlinear parameters and correlation coefficient on the size and power properties of bootstrap linearity test. Experiment results show that for all sample size the bootstrap linearity test has better power properties and can correct size distortion of LST statistic based on limit distribution theory very well. Furthermore, the paper still proposes the basic idea and implementation framework of fast double bootstrap（ FBD）, and it is shown with experiment that the FBD is robust and has better convergent property than simple bootstrap method.
出处
《系统工程学报》
CSCD
北大核心
2010年第2期177-184,共8页
Journal of Systems Engineering
关键词
平滑转移自回归模型
线性检验
自助法
smooth transition autoregressive model（STAR）
linearity test
bootstrap method
