Ponente:  Javier Hualde (UPNA)

Abstract: Cointegration, as introduced by Engle and Granger (1987), is pivotal in analyzing long-run equilibrium relationships among economic variables. While traditional cointegration models have been effective in handling mixed-order integrated variables, they can lead to misleading conclusions from an economic equilibrium perspective if trend stationary observables are involved. This type of variables lead to the socalled trivial cointegration, which might falsely suggest a long-run relationship where none exists. Testing for nontrivial cointegration is possible using standard methods (like ADF, KPSS or Johansen’s approach), but this necessarily requires a sequential method involving either pre-testing or post-testing, and, moreover, it typically leads to an inconsistent test. This paper proposes a direct and consistent test for nontrivial cointegration in a bivariate setting motivated by the different behavior of the sample correlation between the observables under various cointegration scenarios. Unlike residual-based methods, our proposal is invariant to the ordering of the variables and it is also robust to different initial conditions and deterministic components. Our testing approach is compared with standard methods by means of a Monte Carlo experiment, and we include the analysis of an empirical application to the term structure of nominal interest rates for the United States.