Nonstationary Censored Regression

This paper considers the censored regression model under the assumption that the regressors are integrated. We show that Maximum Likelihood estimation is superconsistent and asymptotically mixed normal, implying that standard inference techniques remain valid, and that in general least squares estimation based on the positive observations only is superconsistent, but not mixed normal. An exception to this is the case of a single integrated regressor; in that case, least squares on the positive observations will be asymptotically mixed normal and asymptotically equivalent to the Tobit Maximum Likelihood estimator. We also derive a test for the null of Tobit cointegration, and apply this test to price floors of agricultural commodities.