Stochastic Frontier Models with Errors in Variables: A GMM Approach

當隨機邊界模型逐漸被應用到不同的研究領域時，如何處理變數衡量誤差的問題，成?亟待解決的課題。本文針對具有變數衡量誤差問題的隨機邊界模型，提出一組GMM估計式，以求得模型參數的不偏估計值。此GMM估計式是以模型的高階動差?建構基礎，且是特別針對具有截斷式常態分配的隨機邊界模型所設計，而此類模型正是目前實證應用最廣的模型。蒙地卡羅的模擬結果顯示，該估計式在觀察數?500的樣本中，即有良好的表現。As the use of stochastic frontier (SF) models increases in the various fields of economics and finance, the need to address the problem of measurement errors in variables becomes urgent. In this paper, we propose a generalized method of moment (GMM) estimator for a stochastic frontier model to correct the measurement error problem. The estimator differs from the method of moment (MoM) estimator of Chen and Wang (2004) in two important ways: (1) The GMM estimator is proposed for a SF model with a truncated-normal random variable, which is a more flexible and empirically-popular model than the half-normal SF model targeted by the MoM estimator. (2) The GMM estimator uses more moment conditions and is more efficient. Simulation results show that the GMM estimator performs quite well for data with a reasonable sample size