It is shown that the fully modiﬁed ordinary least squares (FM-OLS) estimator of a unit root in time series regression is T 3 /2 -consistent. Relative to FM-OLS, therefore, the least squares and maximum likelihood estimators are inﬁnitely deﬁcient asymptotically. Simulations show that this dominance of FM-OLS persists even in small samples.
Phillips, Peter C.B., "Hyper-Consistent Estimation of a Unit Root in Time Series Regression" (1992). Cowles Foundation Discussion Papers. 1283.