A0543
Title: Estimating spillovers using panel data with a factor structure
Authors: Hao Dong - Southern Methodist University (United States) [presenting]
Qiwei Yao - London School of Economics (UK)
Abstract: The estimation of spillover effects using panel data is considered. In contrast with existing approaches, we propose a new method taking advantage of the latent factor structure. Precisely, the cross-sectional variations of the variables that generate spillovers are assumed to be driven by a finite number of factors. Such a latent factor structure implies constraints on spillovers and can be used to improve the performance of existing estimator like the Lasso. As the components of the constraints may not be observed, we propose an estimates based constrained Lasso (Elasso) by substituting in corresponding estimators. When all factors are strong, Elasso asymptotically shares the same error bound with the infeasible constrained Lasso (Classo) where the constraints are assumed to be fully observable, which is known to have sharper error bound than the Lasso. In the presence of weak factors, the error bound of the Elasso is not as sharp as that of the Classo, but still can be sharper than that of the Lasso when the number of factors is small. Simulation results demonstrate our findings.
