时变系数广义空间滞后模型的贝叶斯估计

Bayesian Estimation of Generalized Spatial Lag Models with Time-varying Coefficients

传统空间计量模型采用单一不变系数表征单元间的空间关系,但随着样本中个体和时间的增大,不变系数模型难以准确反映空间关系的时变性,可能导致参数估计有偏。基于此,本文构建时变系数广义空间滞后模型,利用贝叶斯方法和MCMC抽样估计模型参数,并与不变系数模型作对比,最后应用于具体实例。数值模拟结果表明,时变系数模型参数估计的平均偏差和均方误根都小于不变系数模型,且均方误根随个体或时间的增大而减小。实例应用不仅重新测度了产业集聚对产业结构升级影响的空间时变效应,还再次证实了模型和方法的实用性。

Traditional spatial econometric models use a single invariant coefficient to characterize the spatial relationship between the units.However,with the increase of individuals and time in the sample,the invariant coefficient model cannot accurately reflect the time variability of the spatial relationship,which may lead to biased parameter estimates.Based on this,this paper constructs a generalized spatial lag model with time-varying coefficients,uses Bayesian method and MCMC sampling to estimate model parameters,compares it with invariant coefficient model,and finally applies it to specific examples.The numerical simulation results show that the mean deviation and root mean square error of the parameter estimates of the time-varying coefficient model are smaller than those of the invariant coefficient model,and the root mean square error decreases with the increase of individuals or time.The example application not only re-measures the timevarying effect of the impact of industrial agglomeration on the upgrading of industrial structure,but also confirms the practicability of the model and method.