Instrumental variable estimation for duration data

In this article we focus on duration data with an endogenous variable for which an instrument is available. In duration analysis the covariates and/or the effect of the covariates may vary over time. Another complication of duration data is that they are usually heavy censored. The hazard rate is invariant to censoring. Therefore, a natural choice is to model the hazard rate instead of the mean. We develop an Instrumental Variable estimation procedure for the Generalized Accelerated Failure Time (GAFT) model. The GAFT model is a duration data model that encompasses two competing approaches to such data; the (Mixed) Proportional Hazard (MPH) model and the Accelerated Failure Time (AFT) model. We discuss the large sample properties of this Instrumental Variable Linear Rank (IVLR) estimation based on counting process theory. We show that choosing the right weight function in the IVLR can improve its efficiency. We discuss the implementation of the estimator and apply it to the Illinois re-employment bonus experiment.