I have a data set of 32 obs (yearly) and all the variables are I(1). I checked for co-integration but the variables are not co-integrated.
Which modelling is appropriate in this case? Can't I use ARDL model for this purpose?
Your suggestions and quick responses are needed. Please clarify the issue (that is most important) and if possible send links for the same (secondary).
You can still use ARDL. The reason is that you want the reviewer to know that you are aware of it, that you checked for cointegration and then go to interpret the first differences because your FTR decision.That is,
Your cointegration test asked: H0 : long terms are zero (no coint)
Ha : long terms are not zero (coint)
You failed to reject (FTR) that long terms are zero so they can be treated as zero.
If all vars are I(1) then you can use both johnson and ARDL ,after checking lon run relationship. If long run relationship not exists, discuss the given results. This is good to give actuall results and conclusins than to play wd data
Sofia Anwar .@@
He stated that there is no cointegration and you still are talking about Johansen estimator with 32 observations; That will not make sense either!!
Thank you Chuck and Sofia for your response.
Let me just ask you one more thing that I gone through the ARDL and bounds test also and AIC criteria is telling that have ARDL (1, 0). Should I interpret this and come to conclusions???
Thank you in advance for your response.
Simply, If you want a model individually for each variable without considering other variables you can use ARIMA models; however, If you want a model for each variable including other variables in the model you can use ARDL models.
@Brian I have used Johansen's co-integration test as well as Bounds test for that purpose.
Then you may need to only consider a short run analysis with variables in differences
It makes sense to estimate an ARDL and this allows the referee to see ALSO that tecm is insignificant. Simply estimating the I(0) terms attracts a question and possible rejection. Yes, Arize (2017) unders that Bounds F-test is easily affected by the included I(0) terms, therefore, one can find no cointegration easily(Type 2 error).
You should use ARDL Bond test. Using F test value you can determine whether to report both long & short term coefficient or short term only. See the following YouTube video https://youtu.be/qovhIG_WzoE
I would suggest use a simple reduced form VAR since your variables are integrated of the same order but they are not co-integrated. This is one condition of estimating a reduced form VAR. you can check (Amisano and Giannini, 1997) and (McCarthy, (2000),
Thanku @Harris Sir for your valuable suggestion. I will go through these research papers.
- Badges
- Science topic