This paper aims to detect structural break points in latent networks in a panel data setting. We consider panel models where the outcome of a unit depends on the outcomes and characteristics of other units. The latent network structure induces high-dimensional parameters and interactive outcomes generate endogeneity. Our goal is to detect breaks in high-dimensional network parameters associated with endogenous variables. We propose a two-step penalized nonlinear least squares approach to estimate the break points based on reduced forms, and show that the resulting estimator achieves superconsistency. This property allows us to estimate, and make inferences on, network and slope parameters as if the true break points were known. An empirical application illustrates the proposed methodology.
COLLABORATION
Strengthening global labour economics research and policy engagement | The Luxembourg Institute of Socio-Economic Research (LISER) is proud to announce that the IZA Network, one of the world’s foremost communities in labour economics, will join LISER as its new institutional home starting January 1, 2026.
