Cointegration Tests Using Instrumental Variables

This paper proposes new cointegration tests based on instrumental variable (IV) estimation. An important property of our tests is that the asymptotic distribution remains standard normal (or Chi-square) regardless of the number of regressors, differing deterministic terms, structural dummies, and inclusion of stationary covariates. Thus, our IV cointegration tests have the operational advantage that they do not depend on nuisance parameters. As such, we can incorporate stationary covariates into a model to enhance power without affecting the asymptotic distribution of the test. This is important because it alleviates the need to tabulate the critical values for every possible case or to bootstrap the critical values.