We propose a semiparametric first-difference estimator for panel censored-s
election models where the selection equation is of tobit type. The estimato
r allows the unit-specific term to be arbitrarily related to regressors. Th
e estimator minimizes a convex function and does not require any smoothing.
A simulation study is provided comparing our proposal with the estimators
of Wooldridge (Journal of Econometrics 68 (1995) 115) and Honore and Kyriaz
idou (Econometric Reviews (2000)). (C) 2001 Elsevier Science B.V. All right
s reserved. JEL classification: C24; C34.