An alternative procedure to test for cointegration in STAR models

Abstract

This paper proposes an alternative procedure to test for cointegration in smooth transition autoregressive (STAR) models. We consider the exponential STAR (ESTAR) and double logistic STAR (D-LSTAR) models. The proposed tests are t -tests with a null hypothesis of no cointegration and an alternative hypothesis of cointegration with STAR adjustment. The procedure introduced in this paper employs a grid search to compute a test statistic. Monte Carlo simulations demonstrate that as compared to other tests, the proposed approach has better power when the persistence of the process toward equilibrium and the sample size increase.