【主讲】卯光宇 (北京交通大学)
【主题】Testing for Error Cross-Sectional Independence in a Two-Way Error Components Panel Data Model
【时间】2016年4月29日 (周五) 15：30-17：00
【地点】上海财经大学经济学院楼701室
【语言】英文
【摘要】This paper proposes a new test for the error cross-sectional independence in a two-wayerror components panel data model. The model is postulated to be based on large dimensional data sets, and hence both the cross-sectional dimension and the time-series dimension of the model are assumed to tend to in nity when related asymptotic theories are developed. Based on an existing statistic in the statistical literature under the raw data circumstance, we formulate an analogy test statistic using the within residuals of the model. It is shown that the resulting statistic needs bias correction to make valid inference. However, the leading term of the bias relies on unknown parameters. To implement feasible correction, we propose a method to estimate the leading term. It is theoretically proved that the feasible bias-corrected statistic can be employed to test the error cross-sectional independence. Additionally, several simulation experiments are also conducted to evaluate the nite-sample performance of the newly developed test. It is found that the test performs well under the null hypothesis irrespective of the relative magnitude between the cross-sectional dimension and the time-series dimension considered in this paper, and has power against several typical alternatives.
