國立東華大學應用數學系

專    題    演    講

       主講人：冼芻蕘 教授(清華大學經濟學系)

       時  間： 108 年 12月 13日 (星期五)  15:10-17:00

地  點：理工一館 A324

講  題：On selecting valid instruments for structural vector autoregression

Abstract

With the prevalence of the so-called big data", structural models/equations are often estimated with high-dimensional instruments. Notable research papers include Belloni, Chen, Chernozhukov and Hansen (2012); and Kang, Zhang, Cai and Small (2016). The former assumes all instruments are valid and considers an effcient estimator; while the latter proposes some confidence sets of the structural parameters, and investigates their properties under various assumptions on the number of valid instruments. In this paper, we adopt and modify the OGA-HDIC algorithm proposed by Ing (2019) and search for valid instruments out of some high-dimensional potential instruments. Unlike Lasso, this algorithm is arguably more suitable for time-series data. We close this paper with (i) Some comparisons with the high-dimensional urbin-Wu-Hausman (DWH) test proposed by Guo, Kang, Cai and Small (2018); (ii) Some Monte-Carlo simulations.

KEYWORDS: High-dimensional potential instruments; irrepresentable assumption; OGA-HDIC algorithm; sparsity; structural models/equations; valid instruments.

Post

※※※   歡      迎      參      加   ※※※se1081213