Title: Inference in data with high-dimensional dependence structures
Authors: Damian Kozbur - University of Zurich (Switzerland) [presenting]
Christian Hansen - University of Chicago Booth School of Business (United States)
Jianfei Cao - University of Chicago Booth School of Business (United States)
Lucciano Villacorta - Central Bank of Chile (Chile)
Abstract: An inference approach is presented for dependent data in spatial applications. We consider a setting in which a high-dimensional parametric SAR model approximates the score process of a statistical model of interest. The method selects from among a large set of candidate spatial weight matrices which characterize dependence in the score process across observations. In a second step, we estimate standard errors that are robust to cross-sectional correlation structures implied by the selected spatial weight matrices. We show that the resulting procedure defines an inferential strategy which is robust against a flexible class of spatial dependence structures. We provide simulation evidence that shows the procedure outperforms conventional inference procedures.