inclusion of deterministic term into cointegration

[upani]

Dear All,

I have doubt regarding the adding deterministic components in to the model. At which stage, it has to be checked whether we need intercept or trend. Whether we have to check the deterministic component while doing unit root test or we have to include both of these while performing cointegration. Some times we get that the data set (with first difference is stationary for both intercept and trend as well as none), what we have to infer from unit root test before proceeding for cointegration test in this case as unit root test is not conclusive regarding the addition of the deterministic term.

I will be grateful to you if you clarify my doubt.

With sincere regards,
Upananda

[TomDoan]

In general, you pick that based upon what your prior knowledge of the series. See the discussion in section 1-4 of the CATS Manual. Because of the interaction between the unit roots and the deterministic components, the labeling of the choice isn't "obvious" (for instance, DET=CONSTANT means the series (potentially) have trends), but once you understand it, it usually isn't that difficult.

There's also a chapter in Juselius' The Cointegrated VAR Model on choice of deterministic components including (possibly) dummies.