EViews.com

Hi there.

I'm going to run a regression but the time-series are non-stationary. Is it enough to include a time-index variable in the regression to de-trend the data? Or should I transform the data first so that it will be stationary, and then run the regression with the time-index variable?

Regards, Ondin.

Assuming that by "nonstationary" you mean "difference stationary," you do need to use the first difference. Including a trend won't do the trick.

Well, the data I've got show both upward and downward trending. The data is still non-stationary (according to Dickey-Fuller tests) when using first differences. Isn't there somekind of a trend model that could be used in this situation?

Using a trend doesn't adjust for the stochastic properties of something that fails the Dickey-Fuller test.

A couple of things you might consider.

1. First differennce the first-difference.
2. Take logs of the variables, if that would still make sense.

Thanks for that information, finally I'm sure about the trending variable, it had been bothering me for a few days! I've tried both 1. and 2. you mentioned in your earlier reply but that did not make ALL the series stationary. I'll check again on that to be 100% sure. Is there anything else I could do?

Regards, Ondin.

I managed to see that my dependent variable is I(2) and all the independent variables vare I(1). Would it be possible to use OLS with those I(2) and I(1) variables? Should I use some kind of AR(I)MAX model, ADL model or something else?

Help very much needed, thanks in advance.

Regards, Ondin.