Optimal model averaging for multivariate regression models
Optimal model averaging for multivariate regression models
复制标题
DOI：
10.1016/j.jmva.2021.104858
复制
发表时间：
2021-11
期刊：
影响因子：
--
通讯作者：
Jun Liao;Alan T. K. Wan;Shuyuan He;Guohua Zou
中科院分区：
文献类型：
--
作者：
Jun Liao;Alan T. K. Wan;Shuyuan He;Guohua Zou
在本文中，频率论模型平均被认为是在一个多元回归模型的背景下。我们提出了一个权重选择标准的基础上的插件对应的二次风险的模型平均估计，涉及一个近似的分布的二次型的比例的F分布。我们建立了一个渐近理论的结果模型的平均估计的一般和限制的权重集，并推导出收敛速度的模型权重的二次风险为基础的最优权重。模拟研究和基于中国第六次全国人口普查的应用说明了我们方法的优点。
In this paper, frequentist model averaging is considered in the context of a multivariate multiple regression model. We propose a weight choice criterion based on a plug-in counterpart of the quadratic risk of the model average estimator that involves an approximation of the distribution of a ratio of quadratic forms by an F distribution. We establish an asymptotic theory for the resultant model average estimator for both the general and restricted weight sets, and derive the convergence rate of the model weights to the quadratic risk-based optimal weights. The merits of our approach are illustrated by a simulation study and an application based on from the Sixth National Population Census of China.