统计研究 ›› 2011, Vol. 28 ›› Issue (12): 83-88.

• 论文 • 上一篇    下一篇

面板数据模型的截面相关检验研究

韩本三等   

  • 出版日期:2011-12-15 发布日期:2011-12-15

Cross-sectional Dependence Test for Panel Data Models

  • Online:2011-12-15 Published:2011-12-15

摘要: 相关系数的绝对值形式可以很好的避免Pesaran(2004)的CD统计量中异向相关性相互抵消的情况,相应得到一个新的检验面板数据模型扰动项截面相关的统计量。蒙特卡洛模拟显示,无论是在因子模型下还是在空间移动平均模型下,新提出的统计量水平扭曲(size distortion)检验和功效(power)检验表现较好。通过模拟还发现当存在序列相关的扰动项时,先将扰动项进行去序列相关处理可以有效地避免序列相关导致的水平扭曲,并且不会降低统计量的功效。

关键词: 面板模型, 截面相关, 序列相关, CD检验, LM 检验

Abstract: Pesaran’s CD statistics (2004) would tend to zero with positive correlation coefficients and negative ones interacting. Thus a new statistics is proposed to test cross-sectional dependence for panel data models replacing of the coefficients with absolute value forms. Monte Carlo simulations show no size distortion and good power in common factor models and spatial moving average models. And we also find that when the disturbance terms are serially correlated, the statistics proposed here presents no size distortion and the same power after dealing with the serial correlation.

Key words: Panel Model, Cross-sectional dependence, Serial correlation, CD Test, LM Test