Determening whether a series has an intercept/trend/none

[SSotirov]

Hi all,

I need to run unit root tests on a number of processes, and while for most it is fairly obvious that I should choose the option trend and intercept when running the tests, there is one which I am not sure about.

The image is attached to the post. I would be very grateful if someone can tell me what testing strategy to apply and how to determine it in the future. Thank you!

[trubador]

Visual inspection gives the sense that the series has structural breaks, which might induce nonstationarity.
Although unit root tests may be useful, I believe they are overratted. As long as you build a theoretically valid time-series model that passes all the usual diagnostics, you do not have to worry about the results of unit root tests.

[SSotirov]

Thank you for your reply, trubador!

I am actually doing a residual analysis trying to find cointegration, so I need to test for unit root and rejcet it in order to conclude that there is indeed cointegration. Actually all three testing strategies reject the null hypothesis of unit root, but at different confidence level. I need to know which one to include in my paper :)

[trubador]

In general, I suggest the use of Johansen procedure for the test of cointegration, but if you must, then it would be wise to report the results of all three unit root tests.
Also, you might find Gregory-Hansen Cointegration Test useful.