In recent economic studies, Tobit models (censored regression models) in which the dependent variables cannot be negative are widely used. Although the Tobit models are usually estimated by the maximum likelihood method, the estimator is inconsistent under heteroscedasticity and nonnormality of the error terms.
Powell [1984] proposed a modified least absolute deviations estimator which is strongly consistent under heteroscedasticity and nonnormality. One of the major problems of Powell's estimator is its computational difficulty. Since the minimand is the complicated form, we cannot use standard methods to caluculate Powell's estimator.
In this paper, I consider a new algorithm and evaluate Powell's estimator by the Monte Carlo experiments.