- Toulouse School of Economics
Wednesday, February 8, 2017
Polo Santa Marta, Via Cantarane 24, Room 1.59
This paper considers models of social interaction when the underlying networks are unobserved but sparse and there are endogenous, contextual, and correlated effects. We accommodate prior knowledge on the sparsity pattern (group sparsity, known existing or nonexisting links) and restrictions on the parameters. We provide results on identification, rates of convergence, model selection, and inference for the parameters and linear functionals in the high-dimensional paradigm. Inference is robust to identification and uniform over large classes of sparse identifiable parameters and data generating processes. Some results hold in finite samples. For computational convenience, we only rely on convex programs.