摘要
文章给出了DF统计量在整值自回归模型INAR(1)下的极限分布,研究了DF检验在具有泊松扰动项的INAR(1)和具有几何分布扰动项的INAR(1)等模型下的可靠性。在两个模型下,分别给出了DF统计量的有限样本分布、检验的势,以及一些真实数据的例证。结果显示,在这两类典型离散值模型下,DF统计量的有限样本分布与标准正态分布差异较大;DF检验在离散值模型的检验势好于其在连续值模型下的势;但检验势在样本容量较小、近单位根时非常低,往往会导致伪检验。
This paper presents the limit distribution of DF statistics in integer-valued autoregressive model INAR(1) and studies the reliability of DF tests in INAR(1) and other models with a Poisson perturbation term and geometrically distributed perturbations. And then, under the two models, the paper respectively provides the finite sample distribution of DF statistics, the potential of test and some examples of real data. The results show that under the two typical discrete value models, the DF statistics differ greatly from the standard normal distribution in the finite sample distribution, and that the DF test potential in the discrete value model is better than in the continuous value model, but the test potential is very low when the sample size is small and close to the unit root, which often leads to pseudo-test.
作者
蔺富明
史代敏
Lin Fuming;Shi Daimin(School of Mathematics and Statistics,Sichuan University of Science&Engineering,Zigong Sichuan 643000,China;School of Statistics,Southwestern University of Finance and Economics,Chengdu 611130,China)
出处
《统计与决策》
CSSCI
北大核心
2020年第12期44-49,共6页
Statistics & Decision
基金
四川省高校重点实验室(桥梁无损检测与工程计算)开放基金项目(2018QZJ01)
四川轻化工大学人才引进项目(2019RC10)。
关键词
单位根检验
整值自回归
有限样本分布
unit root test
integer-valued AR(1)
finite sample distribution