CAJ | 학술논문

本文讨论了含有随机效应的面板数据模型，通过引入条件Laplace先验，构造了一种新的贝叶斯Lasso分位回归法。与一般贝叶斯分位回归法不同，该方法能够更大程度地将模型中非重要的解释变量系数压缩至0，从而在估计系数的同时也起到变量选择的作用。利用积分恒等式，本文构造了一种易于实施的参数估计切片Gibbs抽样算法。模拟结果显示，模型含有较多变量时，新方法排除“噪声”变量的能力明显高于现有文献中的其他方法。本文最后对我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。
본문토론료함유수궤효응적면판수거모형，통과인입조건Laplace선험，구조료일충신적패협사Lasso분위회귀법。여일반패협사분위회귀법불동，해방법능구경대정도지장모형중비중요적해석변량계수압축지0，종이재고계계수적동시야기도변량선택적작용。이용적분항등식，본문구조료일충역우실시적삼수고계절편Gibbs추양산법。모의결과현시，모형함유교다변량시，신방법배제“조성”변량적능력명현고우현유문헌중적기타방법。본문최후대아국각지구다개굉관경제지표적면판수거진행건모분석，연시료신방법고계삼수여도선변량적능력。
This paper discusses the random effects panel data model and establi- shes a new Bayesian Lasso quantile regression method by using of conditional La- place prior. It is different with the general Bayesian quantile regression method that it can shrink the coefficients o{ non-important explanatory variables to zero, and select variables simultaneously. By using of integral identities, the paper constructs an easy slice Gibbs sampling algorithm to estimate parameters. Monte Carlo simula- tion indicates that the proposed method is obviously superior to other methods in literatures when excluding noise variables from a lot of explanatory variables. Finally, we study the panel data including several macroeconomie indica- tors of our country and demonstrate the new rameters and selecting variables. method＇s capability o{ estimating pa