Pei Yu

This paper studies semiparametric identification and estimation of dyadic (directed) link formation on social networks when nodes are endogenously sampled based on observed and unobserved characteristics. Using a special regressor conditionally independent of the unobserved effects and shocks in node payoffs, I transform the binary link outcome into a linear model with interactive fixed effects. The approach does not require structural modeling of sample selection or parametric distributional assumptions. I propose a two-step estimator with a kernel estimate of the special regressor's conditional density in the first step. Under suitable conditions, the estimator is consistent and asymptotically normal, but the incidental parameter bias for individual heterogeneity persists. I develop an analytical bias correction and estimate the variance with a modified jackknife. Monte Carlo simulations examine finite sample performance, and an empirical application studies village lending networks in India.

Point identification and closed-form estimation of the joint distribution of random coefficients and time-varying errors in fixed-T dynamic panels via an inverse Radon transform.

Estimation of social interaction models with endogenous treatment and network formation via a control function approach.