固定效应模型空间相关性的Bootstrap LM-error检验  

Bootstrap LM-error Test of Spatial Correlation in Fixed Effects Model

作　　者：龙志和[1] 李文丽[2] 陈青青 

机构地区：[1]华南理工大学工商管理学院 [2]华南理工大学经济与贸易学院 [3]广东南方广播影视传媒集团有限公司

出　　处：《数量经济技术经济研究》2015年第8期149-160,共12页Journal of Quantitative & Technological Economics

基　　金：国家自然科学基金"空间面板数据模型LM检验的Bootstrap方法有效性研究"(71271088)的资助

摘　　要：基于Lee和Yu(2010)的正交转换消除固定效应,将FDB方法用于空间固定效应模型误差自相关的LM-error检验。在不同的误差结构、样本量、空间权重矩阵、序列相关系数和固定效应大小条件下,比较渐近LM-error检验和Bootstrap LM-error检验的水平扭曲和功效。蒙特卡洛模拟实验表明,当误差项为标准正态分布时,两者检验均具有较好的水平扭曲和功效表现。当误差项为异方差或者序列相关时,渐近LM-error检验存在严重的水平扭曲,而Bootstrap LM-error检验能够有效地校正其水平扭曲,且其检验功效与渐近LM-error检验功效近似相等,Bootstrap LM-error检验是更为理想的检验方法。In the paper, we apply FDB method for LM-error to test spatial error auto- correlation in fixed effects panel data model, which is based on orthogorual transformation method proposed by Lee and Yu （2010） to remove fixed effects. The size distortion and pow- er of asymptotic LM-error test and these of Bootstrap LM-error test are compared in different conditions, which include structures of error, sample size, spatial weight matrix, serial correlated coefficient and fixed effects level. The Monte Carlo experiments show that both the asymptotic test and Bootstrap test have good performance of size distortion and power when the error is normally distributed. When the error is in presence of heteroskedasticity or serial correlation, the asymptotic LM-error test will result in serious size distortion. However, in the same condition, Bootstrap LM-error test has better size per- formance. In addition, its power is asymptotically equal to asymptotic LM-error test. Thus, Bootstrap LM-error test is a better test method.

关 键 词：Bootstrap抽样 LM-error检验 空间误差 蒙特卡洛 

分 类 号：C01[社会学]