平滑转移空间随机前沿模型的贝叶斯估计  

The Bayesian Estimation of Smooth Transition Spatial Stochastic Frontier Model

作　　者：蒋青嬗 黄晶晶 马佳羽 Jiang Qingshan;Huang Jingjing;Ma Jiayu(School of Mathematics and Statistics,Guangdong University of Foreign Studies,Guangzhou 510006,China;School of Financial Mathematics and Statistics,Guangdong University of Finance,Guangzhou 510630,China)

机构地区：[1]广东外语外贸大学数学与统计学院,广州510006 [2]广东金融学院金融数学与统计学院,广州510630

出　　处：《统计与决策》2023年第3期43-48,共6页Statistics & Decision

基　　金：广东省自然科学基金资助项目(2019A1515110267);广东省哲学社会科学项目(GD19YYJ08);全国统计科学研究项目(2021LY042;2022LY035)。

摘　　要：经济数据常存在空间相关性,忽略空间相关性会引发内生性问题,导致相应估计量有偏且不一致。空间随机前沿模型在随机前沿模型的基础上考虑了生产单元的空间相关性,更利于效率测算。然而现有空间随机前沿模型的生产函数形式单一,适用性较差,实证分析存在局限性。文章在空间随机前沿模型中引入平滑转移效应,构建了平滑转移空间随机前沿模型,该模型同时考虑了空间相关性和个体异质性,适用性较佳。为丰富估计方法,同时采用极大似然方法和贝叶斯方法估计模型,其中极大似然估计的核心在于推导对数似然函数、对数似然函数的最优化以及使用JLMS法估计技术效率,贝叶斯估计的核心在于推导未知参数的后验分布及执行MCMC抽样。数值模拟结果显示:(1)极大似然估计和贝叶斯估计的估计精度均较高,其中贝叶斯估计的估计精度略高于极大似然估计;增加样本容量,贝叶斯估计和极大似然估计的估计精度更高。(2)若忽略空间效应或者平滑转移效应,则估计精度较低。Spatial correlation often exists in economic data.Ignoring spatial correlation will lead to endogenous problems,thus resulting in biased and inconsistent estimators.Spatial stochastic frontier model considers the spatial correlation of produc⁃tion units based on the stochastic frontier model,which is more conducive to efficiency measurement.However,the production function of the existing spatial stochastic frontier model has a single form and poor applicability,and the empirical analysis has limitations.This paper introduces the smooth transition effect into the spatial stochastic frontier model,and constructs the smooth transition spatial stochastic frontier model.The model takes into account both spatial correlation and individual heterogeneity,and has good applicability.In order to enrich the estimation methods,the maximum likelihood method and the Bayesian method are used to estimate the model.The core of the maximum likelihood estimation is to derive the logarithmic likelihood function,op⁃timize the logarithmic likelihood function and use JLMS method to estimate the technical efficiency;the core of Bayesian estima⁃tion is to derive the posterior distribution of unknown parameters and perform MCMC sampling.The numerical simulation results are as the follows:(1)The accuracy of maximum likelihood estimation and Bayesian estimation is higher,and the accuracy of Bayesian estimation is slightly higher than that of maximum likelihood estimation.With the increase of sample size,the accuracy of Bayesian estimation and maximum likelihood estimation is better.(2)If the spatial effect or smooth transition effect is ignored,the estimation accuracy is lower.

关 键 词：随机前沿模型 平滑转移 空间效应 极大似然估计 贝叶斯估计 

分 类 号：O212.8[理学—概率论与数理统计]