Output Unit Root ADF Test

[nikieb]

Dear all,

So far the forum has helped me a lot, but now I have a few questions myself.
1
I am testing for unit root for with ADF and PP test. At level and first difference, and I have taken the natural logarithm of two variables, lm1 and lm3. I have tested for unit root before, and now that I am testing it again with the exact same data, the result differs from last time. I have attached the workfile with group 1 and group 2, and both include lm1 and lm3. Group 1 includes the 'new' data, and group 2 the 'old'.

Can you help me and explain why the p-values of the unit roots are different? I can't solve it myself. The data are exactly the same, and the method used as well. Automatic lag selection for adf.

2
Eventually I will forecast inflation with four models, which I will compare using RMSE.
Myavailable data is van 1996:1 tot 2013:4.
My subsample is 1996:1 2011:4
So I forecast 1,4,8 quarters from 2012:1 to 2013:4 to get the RMSE.
The out-of-sample forecast will be from 2014:1 to 2015:4.

For the out-of-sample forecast, should I use the actual data of the 1,4,8 quarters from 2012:1 to 2013:4 or continue with the forecasted data van 2012:1 tot 2013:4 as a basis for 2014:1 to 2015:4?


3
Is the following correct?
For comparison of the various models (AR, ARIMA, VAR, VEC), using RMSE, all models and variables should be tested in the same way, i.e. d(log(cpi)) d(log(m1)) etc. All in first difference or in log or..

Thank you in advance,

Nikie

[nikieb]

One more additional question:

Why is it that if i create a variable in natural logarithm first,that the unit root test (adf & pp) gives different results from when I have the variable in a workfile, convert the workfile to natural log, and then test for unit root?

Again, the data is the same?

Thank you

[EViews Gareth]

With regards to your first question, both groups give incidental results.