部分线性可加分位数回归模型的NG估计量

NG Estimators of Partially Linear Additive Quantile Regression Model

文章将NG(Nonnegative Garrote)方法应用到部分线性可加分位数回归模型。首先应用B-样条基函数对非参数函数部分进行逼近,将部分线性分位数回归模型转化为"线性分位数回归模型",然后基于分位数回归模型的Mallows-型的信息准则,提出了"线性分位数回归模型"基于Mallows-型的信息准则的NG估计量;蒙特卡洛模拟证明了所提出的NG估计量表现良好。最后,将NG估计量应用到实际例子的分析中,结果显示变量选择效果较好。

This paper applies NG(Nonnegative Garrote) method to the partially linear additive quantile regression model.Firstly, the paper uses the B-spline basis function to approximate the non-parametric part of the function and converts the partially linear quantile regression model into"linear quantile regression model", and then refers to the Mallows-type information criterion of quantile regression model to propose the NG estimator of linear quantile regression model based on Mallows-type information criterion;Monte Carlo simulation shows that the proposed NG estimator performs well. Finally, the NG estimator is applied to the analysis of practical examples, and the results show that the variable selection effect is good.