社會互動模型的一般化動差估計法和模型平均

Abstract

在一階社會互動模型下，Kelejian and Prucha (1998) 提出一般空間二階最小平方回歸估計法 (general spatial 2SLS) 而Lee (2003) 建議使用更有效率的工具變數，稱為最佳工具變數 (best IV)。本文延伸Lee (2003) 的最佳工具變數至高階社會互動模型 (high-order social interaction model)，並在一般空間二階最小平方回歸估計法計算漸進分配。為了增加高階模型的估計效率，本文再應用Liu (2015) 的最小平方回歸模型平均法 (least square model averaging) 至本文的高階社會互動模型。藉由最小平方回歸模型平均估計法中的局部誤差假設 (local misspecification assumption)，本文得以比較一般空間二階最小平方回歸估計法與最小平方回歸模型平均估計法的估計效率。

Kelejian and Prucha(1998) proposed a general spatial 2SLS estimating framework and Lee (2003) suggested the use of best instrument variables with first-order spatial models. This paper extends such best IV to a high-order social interactions model and derives the asymptotic distribution of the 2SLS estimators. Further, this paper applies the least square model averaging framework proposed by Liu (2015) to improve the estimating performance of the high-order model. Under the local misspecification assumption of the least square model averaging framework, this paper compares the estimating efficiency between the 2SLS estimators and model averaging estimators.