摘要
In time series modeling, the residuals are often checked for white noise and normality. In practice, the useful tests are Ljung Box test. Mcleod Li test and Lin Mudholkar test. In this paper, we present a nonparametric approach for checking the residuals of time series models. This approach is based on the maximal correlation coefficient ρ 2 * between the residuals and time t . The basic idea is to use the bootstrap to form the null distribution of the statistic ρ 2 * under the null hypothesis H 0:ρ 2 * =0. For calculating ρ 2 * , we proposes a ρ algorithm, analogous to ACE procedure. Power study shows this approach is more powerful than Ljung Box test. Meanwhile, some numerical results and two examples are reported in this paper.
提出一种时间序列模型残差诊断捡验的非参数方法．基于残差与时间ｔ的最大相关系数ρ２＊，在零假设Ｈ０∶ρ２＊＝０下，用Ｂｏｏｔｓｔｒａｐ方法建立统计量ρ２＊的零分布。为计算ρ２＊，提出了一种与ＡＣＥ算法相似的ρ算法。有效性分析表明该方法比Ｌｊｕｎｇ－Ｂｏｘ检验更有效。同时还列出一计算机模拟结果和两个实例．
