This paper provides an extensive Monte Carlo comparison of several contempo
rary cointegration tests. Apart from the familiar Gaussian-based tests of J
ohansen, we also consider tests based on non-Gaussian quasi-likelihoods. Mo
reover, we compare the performance of these parametric tests with tests tha
t estimate the score function from the data using either kernel estimation
or semi-nonparametric density approximations. The comparison is completed w
ith a fully nonparametric cointegration test. In small samples, the overall
performance of the semi-nonparametric approach appears best in terms of si
ze and power. The main cost of the semi-nonparametric approach is the incre
ased computation time. In large samples and for heavily skewed or multimoda
l distributions, the kernel based adaptive method dominates. For near-Gauss
ian distributions, however, the semi-nonparametric approach is preferable a
gain.