Further evidence on the robustness of the Tobit estimator to heteroskedasticity

Abstract

When the error terms in a Tobit model are heteroskedastic, the MLE which assumes homoskedasticity is inconsistent. For the special case of a constant-term-only model, we investigate the size of the inconsistency. The inconsistency is greater the greater the heteroskedasticity and the greater the degree of censoring (i.e., the greater the number of limit observations). However, the inconsistency is much smaller than in the corresponding truncated-normal model considered by Hurd.