Usually cointegration models have dynamic, stochastic components as well as
deterministic components. This paper identifies relevant cointegration mod
els in terms of interpretability and similarity with respect to parameters
of-deterministic components. Similarity implies that interference on cointe
gration rank or common trends can be separated from inference on parameters
of deterministic components. The idea is that the functional form and ther
eby the interpretation of deterministic components is not questioned in con
nection with the rank test, but it can be tested subsequently. The paper fo
cuses on likelihood based inference in sector autoregressive models and a f
ew numerical simulations illustrate the role of similarity.