Estimators for Panel Duration Data with Endogenous Censoring and Endogenous Regressors

My paper proposes a new method to estimate duration models. In particular, it shows that the integrated hazard yields moment functions that solve the following, recognized estimation problems: 1. Van den Berg notes in the forthcoming Handbook of Econometrics that we need panel data for robust inference, but lack estimation methods for censored data. The integrated hazard provides a consistent estimator that converges at rate root N and thereby solves this estimation problem. 2. Honore (1993) argues that regressors can be endogenous in the sense that they may depend on the length of previous spells. The integrated hazard yields the first duration estimator that is consistent for endogenous regressors; it thereby allows distinguishing between state dependence and heterogeneity. 3. Chamberlain (1985) notes that all the fixed effect estimators require multiple spells for all observations; the integrated hazard method only needs multiple spells for some observations. This generalization dramatically increases the number of datasets that can be used for duration analysis. Moreover, the concept of the integrated hazard can be used to estimate the baseline hazard nonparametrically while allowing for individual specific effects. The paper also derives a new semiparametric matching estimator. By demonstrating that the integrated hazard solves recognized estimation problems I show that it is a very powerful tool for estimation.