A multivariate autoregressive distributed lag unit root test

This paper introduces a new unit root test based on the multivariate ARDL framework. This new test yields higher power properties compared to the existing multivariate unit root tests based on Covariate Augmented Dickey-Fuller (CADF) as well as a few commonly used univariate unit root tests. The main advantage of the new test over the CADF test is its consideration of possible cointegration relationships between the variable of interest and the explanatory variables in the process of testing for unit roots while the latter does not allow cointegration. Several sets of experiments for size and power are conducted to check the reliability and robustness of the test. The experiments also discover that univariate unit root tests face serious size distortions in testing a cointegrated process. An empirical example using the proposed multivariate ARDL unit root test is demonstrated in this paper.