Nonlinear minimization estimators in the presence of cointegrating relations

In this paper, we consider estimation of a long-run and a short-run parameter jointly in the presence of nonlinearities. The theory developed establishes limit behavior of minimization estimators of the long-run and short-run parameters jointly. Typically, if the long-run parameter that is present in a cointegrating relationship is estimated, its estimator will be superconsistent. Therefore, we may conjecture that the joint minimization estimation of both parameters jointly will result in the same limit distribution for the short-run parameter as if the long-run parameter was known. However, we show that unless a regularity condition holds, this intuition is false in general. This regularity condition, that clearly holds in the standard linear case, is identical to the condition for validity of a two-step Granger-Engle type procedure. Also, it is shown that if the cointegrated variables are measured in deviation from their averages, the standard asymptotic normality result (that one would obtain if the long-run parameter was known) holds. ∗I thank Jeff Wooldridge for suggesting this problem to me, James Davidson for pointing out the problem of including the cointegrated variables in deviation from their averages, and two anonymous referees for excellent comments and suggestions.