It is well known that unit root tests and non-cointegration tests depend on
the deterministic elements like: constants, trends, breaks, outliers, segm
ented trends, etc., that are present under the null hypothesis and maybe al
so under the alternative hypothesis. This is a serious inconvenience for em
pirical work since one could arbitrarily influence the result of the unit r
oot test by a convenient selection of the deterministic elements that are i
ncluded in the regression test. In this paper eve analyze if those problems
could be reduced by forming the cointegration tests on extended error corr
ection models. The analysis is done based on Monte Carlo simulation experim
ents allowing for several structural breaks in the data generating process.