Estimation of a model for matched panel data with high-dimensional two-way unobserved heterogeneity.

Abstract

We consider a model for matched data with two types of unobserved effects: a random effect related to the main observational unit and a random or fixed effect related to a secondary unit to which the main unit is matched. In typical applications, e.g., on registry data, there is a curse of dimensionality which we propose to mitigate using an iterative feasible GLS approach on variables subjected to the Helmert transformation. Control functions allow for correlation between the explanatory variables and the random effects. This approach is illustrated by a wage equation with unobserved individual- and firm-specific effects and an endogenous years-of-schooling variable.