ARDL Approach to Cointegration

[mrrox]

Hi,

I am new to this forum, I would like to ask you if I can do ARDL approach to cointegration in Eviews.

If yes, where can I get instructions or text about how to apply this approach on data?

Thanks
mrrox

[EViews Glenn]

It's not a built-in feature but you should be able to estimate the ARDL just by entering the full specification in the least squares estimation dialog box. Depending on how many variables that are in the non-cointegrating portion of the specification and the dimension of the VAR, this can be kind of a pain. But it is straightforward.

I don't know of any examples.

[solarin]

the bound test can be easily done in eviews via:
go to Estimate and specify your Equation under LS-Least Squares(NLS and ARMA):
like d(DV) c dDV(-1 to -4) dfirstIV(-1 to -4) dsecondIV(-1 to -4) DV(-1) fisrtIV(-1) secondIV(-1)and so on
then go to View-- Wald-Coefficient Tests--Coefficient restriction and test for the inclusion of the level variables.
The F-statistic generated is similar to Microfit's. Try to check it out

[mrrox]

Many thanks Solarin,

I have some more questions regarding ARDL approach to cointegration in Eviews. If you are willing to answer, I will post them here.
Thank you,
mrrox

[solarin]

go ahead

[mrrox]

Great! many thanks,

I have used Microfit 4.0/4.1/5.0 to run ARDL approach to coint. All versions crash while estimating if you increase the lag to above 4. My data is monthly, thus I need to include at least 12 lags.

My variables are lnip lnrop lncpi lnex. Here is my model:

d(lnip) lnip(-1) lnrop(-1) lnex(-1) lncpi(-1) d(lnip(-1)) d(lnip(-2)) d(lnip(-3)) d(lnip(-4)) d(lnip(-5)) d(lnip(-6)) d(lnip(-7)) d(lnip(-8)) d(lnip(-9)) d(lnip(-10)) d(lnip(-11)) d(lnip(-12)) d(lnrop(-1)) d(lnrop(-2)) d(lnrop(-3)) d(lnrop(-4)) d(lnrop(-5)) d(lnrop(-6)) d(lnrop(-7)) d(lnrop(-8)) d(lnrop(-9)) d(lnrop(-10)) d(lnrop(-11)) d(lnrop(-12)) d(lnex(-1)) d(lnex(-2)) d(lnex(-3)) d(lnex(-4)) d(lnex(-5)) d(lnex(-6)) d(lnex(-7)) d(lnex(-8)) d(lnex(-9)) d(lnex(-10)) d(lnex(-11)) d(lnex(-12)) d(lncpi(-1)) d(lncpi(-2)) d(lncpi(-3)) d(lncpi(-4)) d(lncpi(-5)) d(lncpi(-6)) d(lncpi(-7)) d(lncpi(-8)) d(lncpi(-9)) d(lncpi(-10)) d(lncpi(-11)) d(lncpi(-12)) c

should I do diagnostic checks of this model residuals before Wald test?

I used Wald test to see if the lagged level variables are jointly significant, and they were

Then I wanted to run VECM. But for VECM I need ECT of the long run equation. This is where I got stuck. In microfit it is done automatically. Since I am doing it manually here, should I run this and save the residuals as ECT? lnip c lnrop lnex lncpi

Then run this: d(lnip) c ECT d(lnip(-1)) d(lnip(-2)) d(lnip(-3)) d(lnip(-4)) d(lnip(-5)) d(lnip(-6)) d(lnip(-7)) d(lnip(-8)) d(lnip(-9)) d(lnip(-10)) d(lnip(-11)) d(lnip(-12)) d(lnrop(-1)) d(lnrop(-2)) d(lnrop(-3)) d(lnrop(-4)) d(lnrop(-5)) d(lnrop(-6)) d(lnrop(-7)) d(lnrop(-8)) d(lnrop(-9)) d(lnrop(-10)) d(lnrop(-11)) d(lnrop(-12)) d(lnex(-1)) d(lnex(-2)) d(lnex(-3)) d(lnex(-4)) d(lnex(-5)) d(lnex(-6)) d(lnex(-7)) d(lnex(-8)) d(lnex(-9)) d(lnex(-10)) d(lnex(-11)) d(lnex(-12)) d(lncpi(-1)) d(lncpi(-2)) d(lncpi(-3)) d(lncpi(-4)) d(lncpi(-5)) d(lncpi(-6)) d(lncpi(-7)) d(lncpi(-8)) d(lncpi(-9)) d(lncpi(-10)) d(lncpi(-11)) d(lncpi(-12))

Should I do diagnostic checks of the residuals here too? The I do wald test on the differenced variables.

Lastly, is the whole procedure I described correct?
Your help is greatly appreciated!
mrrox

[DJS]

Hi there

I am trying to code up a programme to run Pesaran's ARDL model in Eviews. My plan is to bootstrap the confidence intervals that are needed for the long run equation.

To check my results ive been looking at output using Microfit. Im having trouble figuring out how to include a constant in the error correction representation of the ARDL model when using Microfit 5. As far as I can tell from the manual and playing around with microfit, you cannot specify an unrestricted intercept (ie the constant only appears in the long run equation, but not in the Error correction model). For example, after selecting ARDL univariate modeling and including a constant (& INPT), under option 3 ‘display error correction model’ how can one force the inclusion of an intercept in the error correction specification?

I guess I could just use Microfit to generate the long run coefficients (since this avoids me having to bootstrap the standard errors) and then combine the coefficients to run the error correction representation in Eviews instead (since I cant figure out how to get Microfit to include a constant in the error correction model).

I know this is a Eviews forum, but anyone know how to include a constant in the error correction model in Microfit 5?

Thanks..
dan

[solarin]

I am not sure if there is any specific justification to use 12 lags for monthly data.

On the ECT, I think the procedure is right.

But will suggest you go back to microfit and use it ( with different lag specification) until you get to ecm, then lift it (ect term beneath the result, then go to Process to make it a variable) and manually do the vecm and granger causality in eviews (which provides F-statistics unlike microfit)

Great! many thanks,

I have used Microfit 4.0/4.1/5.0 to run ARDL approach to coint. All versions crash while estimating if you increase the lag to above 4. My data is monthly, thus I need to include at least 12 lags.

My variables are lnip lnrop lncpi lnex. Here is my model:

d(lnip) lnip(-1) lnrop(-1) lnex(-1) lncpi(-1) d(lnip(-1)) d(lnip(-2)) d(lnip(-3)) d(lnip(-4)) d(lnip(-5)) d(lnip(-6)) d(lnip(-7)) d(lnip(-8)) d(lnip(-9)) d(lnip(-10)) d(lnip(-11)) d(lnip(-12)) d(lnrop(-1)) d(lnrop(-2)) d(lnrop(-3)) d(lnrop(-4)) d(lnrop(-5)) d(lnrop(-6)) d(lnrop(-7)) d(lnrop(-8)) d(lnrop(-9)) d(lnrop(-10)) d(lnrop(-11)) d(lnrop(-12)) d(lnex(-1)) d(lnex(-2)) d(lnex(-3)) d(lnex(-4)) d(lnex(-5)) d(lnex(-6)) d(lnex(-7)) d(lnex(-8)) d(lnex(-9)) d(lnex(-10)) d(lnex(-11)) d(lnex(-12)) d(lncpi(-1)) d(lncpi(-2)) d(lncpi(-3)) d(lncpi(-4)) d(lncpi(-5)) d(lncpi(-6)) d(lncpi(-7)) d(lncpi(-8)) d(lncpi(-9)) d(lncpi(-10)) d(lncpi(-11)) d(lncpi(-12)) c

should I do diagnostic checks of this model residuals before Wald test?

I used Wald test to see if the lagged level variables are jointly significant, and they were

Then I wanted to run VECM. But for VECM I need ECT of the long run equation. This is where I got stuck. In microfit it is done automatically. Since I am doing it manually here, should I run this and save the residuals as ECT? lnip c lnrop lnex lncpi

Then run this: d(lnip) c ECT d(lnip(-1)) d(lnip(-2)) d(lnip(-3)) d(lnip(-4)) d(lnip(-5)) d(lnip(-6)) d(lnip(-7)) d(lnip(-8)) d(lnip(-9)) d(lnip(-10)) d(lnip(-11)) d(lnip(-12)) d(lnrop(-1)) d(lnrop(-2)) d(lnrop(-3)) d(lnrop(-4)) d(lnrop(-5)) d(lnrop(-6)) d(lnrop(-7)) d(lnrop(-8)) d(lnrop(-9)) d(lnrop(-10)) d(lnrop(-11)) d(lnrop(-12)) d(lnex(-1)) d(lnex(-2)) d(lnex(-3)) d(lnex(-4)) d(lnex(-5)) d(lnex(-6)) d(lnex(-7)) d(lnex(-8)) d(lnex(-9)) d(lnex(-10)) d(lnex(-11)) d(lnex(-12)) d(lncpi(-1)) d(lncpi(-2)) d(lncpi(-3)) d(lncpi(-4)) d(lncpi(-5)) d(lncpi(-6)) d(lncpi(-7)) d(lncpi(-8)) d(lncpi(-9)) d(lncpi(-10)) d(lncpi(-11)) d(lncpi(-12))

Should I do diagnostic checks of the residuals here too? The I do wald test on the differenced variables.

Lastly, is the whole procedure I described correct?
Your help is greatly appreciated!
mrrox

[mrrox]

Hello Solarin,

Many thanks for the hints. I copied the ECM equation from the short run coefficients estimation menu pasted it to the "Process"and made it a variable. I will now run the VECM and Granger causality tests on Eviews.

Here is the video I made for others on how to estimate ARDL model http://www.youtube.com/watch?v=d9E8BKsocis&feature=plcp

I will post more questions in case I have to, hopefully you do not mind replying.

Best Regards,
mrrox

[bvguizar]

Dear MMROX,

I saw your video on youtube is great but I have some doubts.In articles authors express ARDL model by using just lag values of independent variables plus differences of them. But in your model, you use t values of explanatory variables instead of t-1. Why? And how can I establish by using lag values of explanatory variables in microfit? I have the general form of my model in le85q le85p lgp lgdp le85nst le85nafv, I am using these variables to estimate the demand of ethanol in dependance of its own price, gasoline price, gdp as a proxy for income, number of e85 stations and number of alternative fuel vehicles available. You could say my mode is of the form y y(-1) x(-1) z(-1) d(y(-1)) d(x(-1)) d(z(-1))]. Also, I wanted to add dummy variables to express seasonality on the months where corn has shortage production. Can you please advise? Best. Blanca

[mrrox]

Hi Blanca,

Which paper are you referring to?

I come across some papers who do not include t period values rather t-1 period values are used. But in Pesaran et a. (2001) I could see the t period values in their ARDL equation. Which software are you using? Also see Pesaran & Pesaran (1997) Microfit manual.

Hope this helps.
mrrox

[bvguizar]

Hi

The article I mean is Ethanol demand under the flex-fuel technology regime in Brazil Luciano Charlita de Freitas, Shinji Kaneko. Energy Economics 33 (2011) 1146–1154.

I run the ardl model in microfit but it gives me really weird results. I did the Wald test, the F-stat value came between the bounds so I test for Johansen procedure where I got that there was cointegration. After that I ran the ARDL in microfit but it in the long run coefficients says that nothing is significant and all is in t values. Also the ECM values are saying that my t values are not significant.

Is there any way I can run the model just with t-1 values like you do it on eviews?

THIS IS THE EQUATION I USED le85q le85p lgp lgdp le85nst le85nafv BUT I WANT TO RUN LIKE THIS

d(le85q) c t le85p(-1) lgp(-1) lgdp(-1) le85nst(-1) le85nafv(-1)
just like in the article I mentioned (equation 1) (find article attached as well)

I have a meeting with my supervisor tomorrow and I am really :( Please find attached document

Any help will be very very much appreciated

[bvguizar]

Hi,

I have found the way of using lagged values in microfit. However, the results I get are kind of weird. Could you please take a look at them?
I created the equation:
dLE85Q LE85P1 LGP1 LGDP1 LE85NST1 LE85NAFV1

where "1" at the end of each variable makes it a lagged variable and "d" stands for differences.

I am not sure which parameters must be taken into account at the moment of interpret the results of if what I am doing is wrong. I hope you can give me some feedback.

Also, do you know if you can use second differences in ARDL models?

Best

Blanca

[mrrox]

Hi,
Your model suffers from the presence of heteroskedasticity.

Can you upload part of your data, not all. but include all your variables. I will run a model and upload a video on youtube tonight. or write instructions in a word doc.

[bvguizar]

q