Estimating Unobserved Cross-Sectional Network Links with Large-T Panel Data

We propose a simple three-step method to estimate the cross-sectional interdependence in a customary network econometric model with an unobserved network and a large T. First, we estimate the reduced-form model to construct instruments for the endogenous peer outcomes. Second, we use two-stage least squares to estimate the direct and contextual effects of the exogenous regressors, as well as the endogenous peer effect. Third, we estimate the network weight matrix. Monte Carlo results for small- to medium-sized samples indicate that the estimators of the parameters of interest have small biases and root mean squared errors. We apply the approach to estimate the network structure of the 100 largest listed depository institutions worldwide using daily data over 2011-2020. This reveals the detailed structure of interbank network links and identifies influential banks in the data, of which some but not all are Global Systemically Important Banks designated by the Financial Stability Board.