Feedback in panel data models

Much of the analysis panel data has been based on an assumption strict exogeneity. Distributions are specified for outcome variables conditional a latent individual effect and observed predictor at all dates, with future values assumed to have no distribution. The paper relaxes this in order allow lagged dependent and, more generally, feedback from current variables. Such would arise evaluation study if treatment variable is randomly assigned only previous outcomes. An information bound derived semiparametric regression model sequential moment restrictions, set increasing over time. then applied (scalar) multiplicative random effect. mean not restricted, so that can control various omitted This nonparametric component model. There transformation eliminates leads restrictions which function depends finite-dimensional parameter. simpler problem coincides original problem. form optimal instrumental derived. also considers identification problems when vector two or components.