固定效应时异系数广义空间自回归模型的估计与模拟

Estimation and Simulation of Time-varying Coefficients General Spatial Autoregressive Model with Fixed Effects

空间面板数据模型常呈现时异特征,现实经济现象中的空间关联多带有时异特性。基于此,本文构建固定效应时异系数广义空间自回归模型,首先采用拟极大似然(QML)方法估计模型并证明参数估计量的渐近性,其次依据贝叶斯(Bayes)公式推出参数后验分布并设计MCMC抽样,最后基于数值模拟比较两种方法在有限样本下的模拟情况,结合具体实例对比分析两种方法的实际估计效果。结果发现:一方面,两种方法的参数模拟均方误差都表现出随样本个体数目的增大而减小,表明增加观测个体数目能显著降低参数模拟偏差。另一方面,Bayes估计的均方误差都小于QML估计,说明Bayes估计比QML估计更可靠。

Spatial panel data models often exhibit time-varying characteristics,and the spatial correlations of real-world economic phenomena often have time-varying characteristics.Based on this,this paper constructs a time-varying coefficients generalized spatial autoregressive model with fixed-effect.Firstly,the quasi-maximum likelihood(QML)method is used to estimate parameters of the model and prove the asymptotic properties of the parameter estimators.Secondly,it infers the posteriori distribution of parameters from the Bayes for mula and devises the MCMC sampling.Finally,it is based on numerical simulation comparing the simulation results of the two methods in the case of limited samples,and comparatively analyzes the actual estimation results of the two methods combined with specific examples.The results show that the mean square error of the parameter simulation in both methods decreases with the increasing numbers of sample individuals on the one hand,which indicates that increasing the number of observed individuals can significantly reduce the parameter simulation deviation.On the other hand,the mean square error of Bayes estimation is smaller than QML estimation,indicating that Bayes estimation is more reliable than QML estimation.