trend stationarity

[engur]

I am pretty sure that my data has trend. However after detrending, it is still non-stationary. Also, when I take the first difference of my original data(not detrended), it becomes stationary. So should I do both or is my data difference(first degree) stationary ? thanks a lot for replies in advance!!!!

[trubador]

Data may have stochastic trend after all. In that case, you can try HP filter or frequency filters to remove the trend. Of course, what you should do depends on your research question. If your sole purpose is decomposing the series of interest, then filtering may be enough. However, if you'll continue with the modeling then differencing alone would do just fine. There are also nice models that can handle non-stationary series, but again it depends on the research question.

[engur]

Thanks a lot for your reply.. can I conclude that if a data is difference-stationary, I can start modelling ? Also because it is hard to get an answer in this forum, I want to ask a irrelevant question:

I am trying to run a VECM model. Should I put the variables in levels or first-difference would be ok ? I first estimated a VAR model, but I found cointegrated equations. So, thats why I am doing a VECM model. Should I directly put the stationary data,which I used for VAR model, for VECM or use the initial(level) data ? Thanks for your help!!

[trubador]

Also because it is hard to get an answer in this forum, I want to ask a irrelevant question:

It is really not fair to say that. As of this moment, total number of members are approaching 10K and the total number of topics has already exceeded it. There are almost 40K posts are available, which makes the forum itself as an important "supplementary" source. But the users (especially the newcomers) should keep in mind that there is a high chance that the query they are about to make might have answered before. Therefore, it is always a good idea to do a search in the forum before posting a question. Not to mention that no forum can be a substitute for a well-written textbook. And luckily, we have plenty of good textbooks in the field of econometrics.

I am trying to run a VECM model. Should I put the variables in levels or first-difference would be ok ? I first estimated a VAR model, but I found cointegrated equations. So, thats why I am doing a VECM model. Should I directly put the stationary data,which I used for VAR model, for VECM or use the initial(level) data ?

As you might guess, this question was asked (and answered) numerous times before. As textbooks devote separate chapters for VEC modeling, you might read and understand the method before trying to implement it. Long story short; you do not have to make the data stationary, since VEC model takes account for it.

[engur]

Thanks for the advice as I managed to find answers to some of my questions in the forum. However, I would be very grateful if you could help me one more time. I am looking for effects of oil prices on some macroeconomic variables such as industrial production index and inflation. I take the variables as monthly percentage change, as I am interested only in monthly changes. So, as expected my variables turn out to be stationary because I am doing something similar to first-differencing before putting them into eviews. All my variables are I(0), so does it make sense to do cointegration test? When I do the Johansen cointegration test, it concludes that there are 6 cointegrating equations. Should I do a VAR model without looking at cointegration test or do a VECM model? As I have learned, in order to do a cointegration test, the variables should be non-stationary in levels, but stationary in their first differences. Thanks so much for your help and advice on using this forum again..

[trubador]

Checking for cointegration is important, which you cannot do on I(0) variables. If the variables are I(1) and are cointegrated, then you should go with the VEC model. If the variables are not cointegrated, then you can either use them in levels or in first differences (see the following discussion: http://forums.eviews.com/viewtopic.php? ... 275#p39564)

[engur]

Sorry to post here again.. I am really short of time and could not find an answer to my question throughout the forum because it is a really basic one. I want to put dummy variable in a VAR model and I have to check the significance of this dummy variable for each of the variables in VAR model. Is it enough to just regress every variable on dummy the variable I have created and look at the p-values ? or should I estimate a VAR with the dummy variable and look at the p-values of it for each variable ? Thanks a lot for your help..