基于线性插值的贝叶斯Lasso分位数回归及应用

Bayesian Lasso Quantile Regression and Its Application Based on Linear Interpolation

在贝叶斯Lasso分位数回归中,样本似然函数的计算和后验分布的抽样通常难以处理。针对这一问题,文章采用一种基于线性插值的似然函数计算方法,并结合拉普拉斯先验分布,设计出一种新的对后验分布进行抽样的算法。数值模拟结果表明了该方法具有较好的适应性和参数估计准确性。

In the Bayesian Lasso quantile regression, it is difficult to deal with the likelihood function calculation and posterior distribution sampling. To solve this problem, this paper adopts a likelihood function calculating method based on linear interpolation, and combines with Laplacian prior distribution to design a new sampling method for posterior distribution. Numerical simu- lation result shows that the proposed method has a better adaptability and higher parameter estimation accuracy.