基于线性化技术的删失变点回归模型估计

Estimation of Censored Regression Model With Change Point via Linearization Technique

文章基于线性化技术估计删失变点回归模型中的变点位置及模型参数,克服了传统格点搜索法的收敛速度过慢、变点估计的实际意义不强等缺陷。由于变点存在导致目标函数在变点处不可导的困难,因此线性化技术通过泰勒展开将变点位置转变为模型参数,利用传统删失回归模型加以估计,简便易行且估计结果的收敛速度较快。数值模拟表明估计结果具有良好的有效性和稳健性,慈善捐款的实证分析也验证了所提方法的可行性。

This paper is based on a censored regression model with change point via linearization technique to overcome the slow convergence speed of traditional lattice search method, weak practical significance of change point estimation and other shortcomings. Due to the difficulty that the objective function is not differentiable at the change point, the linearization technique uses Taylor expansion to convert the change point into model parameters. It is simple and easy to use traditional censored regression model for estimation, but the convergence speed of the estimation result is faster. Numerical simulations show that the estimation has good validity and robustness, and empirical research on charitable donations also verifies the validity of the method.