面板数据的自适应Lasso分位回归方法研究

如何在对参数进行估计的同时自动选择重要解释变量,一直是面板数据分位回归模型中讨论的热点问题之一。通过构造一种含多重随机效应的贝叶斯分层分位回归模型,在假定固定效应系数先验服从一种新的条件Laplace分布的基础上,给出了模型参数估计的Gibbs抽样算法。考虑到不同重要程度的解释变量权重系数压缩程度应该不同,所构造的先验信息具有自适应性的特点,能够准确地对模型中重要解释变量进行自动选取,且设计的切片Gibbs抽样算法能够快速有效地解决模型中各个参数的后验均值估计问题。模拟结果显示,新方法在参数估计精确度和变量选择准确度上均优于现有文献的常用方法。通过对中国各地区多个宏观经济指标的面板数据进行建模分析,演示了新方法估计参数与挑选变量的能力。

Study on Adaptive Lasso Quantile Regression for Panel Data Models

How to do parameter estimation and variable selection simultaneously is a hot issue in the study of quantile regression for panel data models .On the base of the assumption that the fixed effect coefficients are subject to a novel conditional Laplace prior ,the paper constructs a hierarchical Bayesian quantile regression model and gives the Gibbs sample algorithm for the unknow n parameter estimation .In consideration of different explain variables should have different shrinkage degree ,the proposed prior has the property of adaptivity ,w hich could select the important explain variables in the model automatically . Furthermore ,the slice Gibbs sample algorithm that the paper proposed is able to estimate the posteriori mean estimation of unknown parameter quickly and efficiently .Monte Carlo simulation study indicates that the proposed method is obviously superior to the existing methods in literatures on the accuracy of parameter estimation and variable selection .Finally ,the paper gives a research of modeling the panel data including several macroeconomic indicators of our country and demonstrates the new method＆#39;s capability of estimating parameters and doing variable selection .