This paper finds three estimators of the largest autoregressive root that are √ T consistent and asymptotically normal even in the local-to-unity framework. The point of departure is that the estimators are based on moments of data that are stationary when evaluated at the true parameter vector. Critical values from the normal distribution can be used for hypothesis testing, irrespective of the treatment of the deterministic terms. Simulations show that the estimates are approximately mean

more »

... sed and the t test has good size in the parameter region when the least squares estimates usually yield distorted inference. JEL Classification: C1