Statistical estimation of dynamic Tobit models

Abstract

This paper considers the estimation of the parameters of the dynamic linear models where the dependent variable is discrete and truncated to the right of a known constant. Due to censoring, some dependent variables cannot be observed. Two approaches are considered: Newton-Raphson algorithm is used for the Poisson regression dynamic Tobit model and the SML-GHK simulator is used for the estimation of dynamic Tobit model whose lagged dependent variable is latent. Both approaches give similar prediction results in terms of the average relative prediction error.