时变空间权重矩阵面板数据模型稳健LM检验有效性研究

空间权重矩阵是描述个体间空间关系的重要工具，通常基于个体间的地理距离构造不随时间而改变的空间权重矩阵。然而，当个体间的空间关系源自经济/社会/贸易距离或人口流动性/气候等特征时，空间权重矩阵本质上可能将随时间而改变。由此，本研究提出时变空间权重矩阵面板数据模型的稳健LM检验。大量Monte Carlo模拟结果显示：从检验水平和功效角度来看，基于误设的非时变空间权重矩阵的稳健LM检验存在较大偏差，但是基于时变空间权重矩阵的稳健LM检验能够有效地识别面板数据中的空间关系类型。尤其是，在时间较长和个体较多等情况下，时变空间权重矩阵的稳健LM检验功效更高。

Validity of Robust LM Tests in Panel Data Models with Time Varying Spatial Weights Matrices

As an important tool to represent spatial correlation among observations, time invariant spatial weights matrices are generally constructed using geographic distance between observations. However, when spatial correlation depends on economic/social/trade distance, migration and climate condition, spatial weights matrices will be substantially changed over time. This paper proposes robust LM tests in panel data models with time varying spatial weights matrices. Extensive Monte Carlo simulations indicate that in the view of size and power, robust LM tests with mis-specified time invariant spatial weights matrices cause large bias, and robust LM tests with time varying spatial weights matrices could effectively identify types of spatial correlation. The power of robust LM tests with time varying spatial weights matrices is much higher than that with mis-specified time invariant spatial weights matrices, especially for the large cross-sectional dimension and for the large time dimension.