A1743
Title: Heteroskedasticity-robust GMM for hierarchical network models with random group effects
Authors: Jiankun Chen - University of International Business and Economics (China) [presenting]
Yanli Lin - University of Western Australia (Australia)
Yang Yang - Tianjin University (China)
Abstract: Identification and estimation are studied in a cross-sectional hierarchical network model in which individuals interact within groups while groups interact across groups. The model accommodates random group effects and unknown heteroskedasticity at the individual level. Although the stacked system can be written in a SAR-like form, the hierarchical structure creates a distinct econometric problem: after eliminating group effects by within-group demeaning, the transformed error covariance is no longer diagonal, so the conventional zero-diagonal quadratic-moment construction used in spatial GMM is not robust to heteroskedasticity. A corrected class of quadratic moments for the within-transformed equation is derived and paired with a second class of between-group quadratic moments based on group averages. The two classes of moments exploit different sources of variation and separately identify the within-group and inter-group network parameters. Consistency and asymptotic normality are established for robust and optimally weighted GMM estimators, and a feasible estimator for the covariance matrix of the linear and between-group quadratic moments is provided under unknown heteroskedasticity. Monte Carlo results show that the proposed procedure improves substantially on linear-moment estimators and remains robust under heteroskedasticity.
