A Simple Cointegrating Rank Test Without Vector Autoregression

This paper proposes a fully nonparametric test for cointegrating rank which does not require estimation of a vector autoregressive model. The test exploits the fact that the degeneracy in the moment matrix of the variables with mixed integration order corresponds to the notion of cointegration. With an appropriate standardization, the test statistics are shown to have a nuisance parameter free limiting distribution and to be consistent under reasonable conditions. Monte Carlo experiments also suggest that the performance of the test is satisfactory with a moderate sample size. The proposed tests are applied to the stochastic growth model using the U.S. aggregate data.