面板数据空间误差分量模型的空间相关性检验

将截面误差分量模型(spatial error components, SEC)扩展至面板数据, 推导其联合检验、边际检验及条件检验, 并通过Monte Carlo模拟实验证明: 当随机效应存在时, 条件检验更为有效; 当随机效应不存在时, 边际检验更为有效; 空间权重矩阵的选取与随机效应是否存在相关, 但未标准化的空间权重矩阵更适合检验空间相关性; 此外, 更大的N或T使得检验更为有效. 研究同时发现, 当真实数据生成过程为面板数据SEC模型时, 传统空间经济计量模型中的Moran I、LM-Error及LM-Lag检验均失效.

Spatial correlation tests of panel data spatial error components model

This paper extends cross section spatial error components (SEC) models to panel data, we derivate joint tests, marginal tests and conditional tests, and using Monte Carlo simulation experiments, we prove that, when random effect (RE) exists, conditional tests are more effective, while RE does not exists, marginal tests are more effective; the choice of spatial weight matrix is related with whether RE exists, but non-standard weight matrix is more suitable; what's more, greater N or T will make tests more effective. At the same time, we find that when the real data garnering process is panel data SEC model, traditional spatial tests as Moran I, LM-Error and LM-Lag have poor performance.