• 论文 •

整数值时间序列模型单位根检验问题研究

• 出版日期:2016-08-15 发布日期:2016-08-11

Research on Unit Root Test for Integer-valued Time Series Models

Wang Zeyu et al.

• Online:2016-08-15 Published:2016-08-11

Abstract:

The unit root test research about the integer-valued time series is just getting started, compared with the non-integer-valued time series. In this paper, the Monte Carlo simulation would be ushered to check the DF statistic and the statistic in INAR (1) models with unit root process. Based on the research, DF statistic asymptotically conforms to the standard normal distribution, meanwhile the actual distribution of this statistic has been impacted by the sample size and the mean of the disturbance term in the finite sample. In addition, the DF statistic does not have the property of any level distortion. That is, the DF can well control the probability of type I error. Because of the data generation feature, the statistic’s probability of committing type I error is zero. Furthermore, the test powers of DF statistic and statistic are influenced by the sample size, autoregressive coefficient and the mean of the error term. In most cases, the test power of statistic is much better than the DF statistic.