本文將Phillips and Han (2008)之一階差分估計式推展至有時間趨勢之一階自我 迴歸模型的估計與推論上。藉由簡易的去除時間趨勢過程, 一階差分估計式仍具 有漸近常態分配的性質; 同時, 據以建構之單根檢定相較於以雙重差分估計式為 基礎之單根檢定更有檢定力。本文提出的方法在固定效果動態追蹤資料模型下更 具應用價值。
This paper adapts the first-difference estimator of Phillips and Han (2008) to the estimation and inference in AR(1) models with trends. With a detrending procedure, the first-difference estimator remains applicable and is shown to retain the Gaussian asymptotics. A unit root test based on the estimator is more powerful than that based on the double-difference estimator. The proposed estimator is especially useful when applied to dynamic panels.