Estimating a Censored Dynamic Panel Data Model with an Application to Earnings Dynamics

Abstract

This paper proposes a method for estimating a censored panel data model with a lagged latent dependent variable and individual-specific fixed effects. The main insight is to trim observations in such a way that a certain symmetry, which was destroyed by censoring, is restored. Based on the restored symmetry, orthogonality conditions are constructed and GMM estimation is implemented. The estimation method is used to study earnings dynamics, using matched data from the Current Population Survey and Social Security Administration (CPS- SSA) Earnings Record for a sample of men who were born in 1930-39 and living in the South dining the period of 1957-73. The SSA earnings are top-coded at the maximum social security taxable level. Although linear GMM estimation yields no difference in earnings dynamics by race, the earnings process for white men appears to be more persistent than that for black men (conditional on individual heterogeneity) after censoring is taken into account.