Prediction in Linear Index Models with Endogenous Regressors

Abstract

In this article, we examine prediction in the context of linear index models when one or more of the regressors are endogenous. To facilitate both within-sample and out-of-sample predictions, Stata offers the postestimation command predict (see [R] predict). We believe that the usefulness of the predictions provided by this command is limited, especially if one is interested in out-of-sample predictions. We demonstrate our point using a probit model with continuous endogenous regressors, although it clearly generalizes readily to other linear index models. We subsequently provide a program that offers one possible implementation of a new command, ivpredict, that can be used to address this shortcoming of predict, and we then illustrate its use with an empirical example.