高階社會網路模型的一般化動差估計與模型平均

Abstract

本論文首先討論一般化動差法如何估計高階空間迴歸模型，為了估計效率而使用最佳化權重矩陣並列出其極限分配。接下來探討模型平均的權重決定的準則以及準則的大樣本性質，概念參考自Feng, Liu, Yao and Zhao (2021)的非線性迴歸模型(nonlinear regression model)的模型平均，討論在一般化動差法估計以及樣本外誤差的基準下，準則的函數以及它的大樣本性質。最後，從蒙地卡羅模擬中可以得知我們的模型平均方法與LASSO在估計以及預測的表現不相上下。

This paper first discusses how the generalized method of moments (GMM) is used to estimate high-order spatial regression models, employing the optimal weight matrix for estimation efficiency and listing its limiting distribution. Next, it explores the criteria for determining model averaging weights and the asymptotic properties of these criteria, drawing on the concept from Feng, Liu, Yao, and Zhao (2021) for model averaging in nonlinear regression models. The discussion focuses on the criterion function and its asymptotic properties under GMM estimation and out-of-sample error benchmarks. Finally, Monte Carlo simulations demonstrate that our model averaging method performs on par with LASSO in terms of estimation and prediction.