Subgraph Network Random Effects Error Components Models: Specification and Testing

Abstract This paper develops a subgraph random effects error components model for network data linear regression where the unit of observation is node. In particular, it allows link and triangle specific components, which serve as basal modeling effects. It then evaluates potential ignoring in estimation coefficients’ variance-covariance matrix. also proposes consistent estimators variance using quadratic forms Lagrange Multiplier tests evaluating appropriate networks. Monte Carlo simulations show that have good performance finite samples. applies proposed to Call interbank market Argentina.