ARDL/ECM Question

[igor]

To better understanding of this issue, please, see Johansen cointegration procedure, 5 cases of deterministic trend specification. Here is approximately the same situation. Case 2: constant restricted into cointegration space, this means loosely that variables in the model do not demonstrate trending behavior. Constant should be only in the long-run equation, but not in the short-run. Case 3: unrestricted constant, this means that variables in the model can have trends. Constant should be included only in the short run equation (Case 4 is also about trending behavior).

ARDL is estimated by OLS too, but long-run equations is estimated here together with short-run dynamics, so it is more appropriate, since in static long-run we can face with so called common factor restriction, in-short, we have to allow for dynamics while estimating long-run eqation.

As I understand ARDL bounds test is so popular because of a possibility to include both I(1) and I(0) variables. But you should take into account that cointegration is possible between I(1) variables and I(0) variable is only help to establish this relationship more clearly in several cases. By the way, Johansen multivariate cointegration procedure is also permits to include I(0) into cointegration space. This leads to additional cointegration vector since I(0) is cointegrated with themselves.

Finally, most papers that used PSS approach is total trash in terms of econometric theory and implementation of PSS method. My last recommendation in this topic: if you want to use PSS bounds test approach properly, then proceed with Eviews (or do it manually in the same manner). If your goal is to produce something self-professed, then use video from internet.

[EnjoFaes]

To better understanding of this issue, please, see Johansen cointegration procedure, 5 cases of deterministic trend specification. Here is approximately the same situation. Case 2: constant restricted into cointegration space, this means loosely that variables in the model do not demonstrate trending behavior. Constant should be only in the long-run equation, but not in the short-run. Case 3: unrestricted constant, this means that variables in the model can have trends. Constant should be included only in the short run equation (Case 4 is also about trending behavior).

ARDL is estimated by OLS too, but long-run equations is estimated here together with short-run dynamics, so it is more appropriate, since in static long-run we can face with so called common factor restriction, in-short, we have to allow for dynamics while estimating long-run eqation.

As I understand ARDL bounds test is so popular because of a possibility to include both I(1) and I(0) variables. But you should take into account that cointegration is possible between I(1) variables and I(0) variable is only help to establish this relationship more clearly in several cases. By the way, Johansen multivariate cointegration procedure is also permits to include I(0) into cointegration space. This leads to additional cointegration vector since I(0) is cointegrated with themselves.

Finally, most papers that used PSS approach is total trash in terms of econometric theory and implementation of PSS method. My last recommendation in this topic: if you want to use PSS bounds test approach properly, then proceed with Eviews (or do it manually in the same manner). If your goal is to produce something self-professed, then use video from internet.

Thank you for this crystal clear advice. I finally understand part of it. As Eviews team has also mentioned on the blogs: choosing the right model is art and science, it is sometimes dubious..

for example: if only 1 or 2 of 7 variables is trend stationary. Because with trending behavior, you mean trend stationary right? Does it automatically force us to case 3? Or because of the fact that most variables are not trend stationary along the ADF & PP tests: we use case 2 with a constant in the long run levels equation.

About your remark about the I(1) and I(0) variables: Thank you. I also feel that most papers are really trash using ARDL, so I feel I am sticking with the Eviews method and use the ECM by eviews.

[igor]

Because with trending behavior, you mean trend stationary right?

I mean, in the first place, variables with stochastic trend, i.e. I(1) variables. Then you need case 3 or 4 (if trend is significant in the long-run).

[EnjoFaes]

Because with trending behavior, you mean trend stationary right?

I mean, in the first place, variables with stochastic trend, i.e. I(1) variables. Then you need case 3 or 4 (if trend is significant in the long-run).

But if you also have I(0) variables you still use case 3 or 4 then? Because if variables do not show trending behavior in case 2, how do you explain eviews using the case 2 option for trending variables like consumption or GDP in their examples & blogs.

[michaelk]

Hi Igor,

I am not sure if your comments are helpful. This is a community for helping each other, not putting people down.

I was asking about the 2 methods of ARDL estimation that seem to be floating around.

1) The Eviews ARDL process
2) The OLS manual process (where you create your own long run model through OLS methods.

Now my thesis supervisor, who has been teaching econometrics for 30 years, says the 2nd method is not correct for various reasons.
I asked eviews staff and they directed me towards the 1st method, which is what I am using, but nothing about the OLS process.

So my question was around OLS. If you knew the answer, you could have said "No, do not use the simple OLS long run model because".....or you could have said "this OLS method is ok to use because of....and this paper uses it...".

Telling me to read econometrics diligently and quoting Pesaran et al (2001) does not help. That paper does not discuss the 2 methods of ARDL estimation that I have mentioned.

These forums are for questions and answers, like a class room.
I really hope you do not teach econometrics and say to students "If you cannot understand this and about trend specification, you have to learn econometrics diligently".

If you cant help, don't reply

Thank you.

[EViews Gareth]

Actually, you asked a question in the EViews estimation forum, which is a forum dedicated to questions solely about EViews estimation methods, and not general econometric questions, or questions on which estimation method is best etc....

Since this specific forum is absolutely not for asking questions about the non-EViews related OLS method, answering with "you need to look elsewhere" seems quite appropriate.

Having said that, Igor appears to have been quite helpful despite the inappropriate question.

[michaelk]

Eviews Graeth,

100% of my question was about estimation methods, nothing more.
Which method to use, 1) or 2).

Thank you.

[EViews Gareth]

Which has nothing to do with using EViews.

[michaelk]

I apologize,

Using ARDL or OLS on eviews is not possible.

Thank you.