中文摘要：

固定效应和随机效应同时选择是面板数据模型研究中的重要问题之一。本文通过分别对固定效应和随机效应引入条件Laplace先验，提出了一种新的贝叶斯双惩罚分位回归法。该方法不仅能对模型中重要解释变量进行自动选择，而且充分考虑到个体随机波动对解释变量系数估计带来的偏差。通过对方差分量的惩罚压缩，减少了模型中未知参数的个数，提高了模型自由度。MonteCarlo模拟及实证分析显示，所提出的方法不仅能准确估计出固定效应系数，而且能精确地捕捉到个体随机效应的波动。

英文摘要：

It is an important issue to select fixed and random effects simultaneously for panel data mod- els. This paper proposes a new Bayesian double penalized quantile regression method by introducing the conditional Laplace prior both for fixed and random effect parameters. This method can not only select the important explanatory variables in the model automatically but also give a full consideration to the biases of parameter estimation for explanatory variables which are produced by individual ran- dom fluctuations. By applying shrinkage to the variance components, the number of unknown param- eters in the model is reduced, thus the model＇s freedom degree is enhanced greatly. Monte Carlo sim- ulation and empirical study indicate that the proposed method can accurately estimate the fixed effect parameters and catch the exact fluctuation of individual random effects.