以貝氏方法估計空間自我回歸模型中之未知網絡

Abstract

我們在空間自我回歸模型(Spatial Autoregressive Model)的框架下引入了貝氏方法來估計其中的未知網絡。本文以指數隨機圖模型(Exponential Random Graph Model)作爲貝氏方法下的先驗分佈。這種作法有兩個主要優點:首先，這種方法是常見高維度統計方法(High-dimensional Methods)的自然延伸;其次，這種方法使計量經濟學家有很大的靈活度將先驗知識或信念納入網絡形成過程中。通過模擬研究，我們展示了這種方法能夠良好地估計未知網絡連接和網絡的高階特徵。

We introduced a Bayesian method for estimating unknown networks in the context of Spatial Autoregressive Model models by introducing a network formation model, Exponential Random Graph Model in this paper, as a prior distribution. This method brings two main advantages: first, this method is a natural extension to common high-dimensional methods; second, this method enables the econometrician to incorporate prior knowledge or belief about the network formation process with great flexibility. Via simulation studies, we demonstrated this approach is practical in recovering unknown networks ties and higher-order characteristics of the network.