A1367
Title: Rank assisted network regression
Authors: Weining Wang - University of Bristol (United Kingdom) [presenting]
Abstract: The aim is to propose studying the interaction effects of social and spatial networks in the presence of a noisy adjacency matrix. First, evidence is provided that existing network datasets exhibit low-rank, sparse, and noisy structures, and this information is utilized to create a de-noised version of the network. The least absolute shrinkage and selection operator (LASSO) is employed in conjunction with nuclear norm penalization to simultaneously regularize the sparse and low-rank components. Two procedures are introduced: a two-step estimator, where the adjacency matrix is first de-noised before using it in regression analysis, and a one-step supervised generalized method of moments (GMM) estimator using proximal gradient methods for efficient computation. Results show that the estimation method performs favorably compared to GMM, especially when dense errors are present and networks are endogenous to measurement errors. Simulation exercises indicate that the method outperforms GMM by up to 30 in root mean squared error (RMSE) terms when noise is present in the network and maintains a significant advantage of approximately 40 on average with endogenous networks.
