(EViews10):ARDL Models (General-to-Specific)

[@CrunchEconometrix]

KZbin recently changed the way my content will be monetised. My channel now needs 1,000 subscribers. So it would be amazing if you show your support by both watching my videos and subscribing to my channel if you haven’t done so already. Monetising my videos allows me to invest back into the channel with some new equipment so this small gesture from you will be extremely huge for me. Many thanks for your support….CrunchEconometrix loves to teach, help me stay online.

[@haider821d]

Madam! my topic is the impact of inflation and unemployment on economic growth...... the previous literature is not clear i am too confused about the model for the topic ????????

[@haider821d]

Madam! my topic is the impact of inflation and unemployment on economic growth...... the previous literature is not clear i am too confused about the model for the topic ????????

[@hssouneabdelkarim626]

Done. Subs and likes already. Good luck .

[@anuragmaurya7853]

Ma'am could you plz mention the sequence of videos ... i found it difficult to follow the sequence for ardl bound test eviews videos.

[@chrisalwis1626]

your videos are extremly helpful especially for beginners. Because you explain very clearly.

[@CrunchEconometrix]

Thanks Chris, for the positive feedback and remarks, deeply appreciated!

[@SherMehta-l4p]

Why are the variables of the ARDL model in level forms. If they are non-stationary, does not one need to construct a first differenced ARDL model? Are these three variables I(0)? Because if they are, then the above specification is perfect. But if They are I(1), then how can one estimate an ARDL in level form of the variables in logs? Should it not be a first differenced ARDL model then?

[@CrunchEconometrix]

In addition to reading the Pesaran ARDL papers, you may want to watch my video on THIS IS HOW TO SPECIFY ARDL MODELS for better understanding. Thanks.

[@stephaniekimani3068]

Thank you for your wonderful videos. Do you have this explanation for stata?

[@CrunchEconometrix]

Not at all.

[@hossameldin4535]

Hi, thanks for those videos. It has been of great help to me. I have a question. How can I chouse the variables to apply on them wald or redundant variable test? Thank you Thank you Thank you

[@CrunchEconometrix]

Hi Hossam, thanks for the encouraging feedback. Kindly watch the video again and follow the steps. Thanks.

[@hossameldin4535]

@@CrunchEconometrix thanks for responding. You said there were 8 insignificant variables weighing more than 50% of the model variables, then you specified 5 out of them! According to what basis!! That's what I couldn't catch. Is it about the least significant variables or what should it be the criteria? Thanks again. Hope you're doing well these days.

[@CrunchEconometrix]

Hossam, but I explained what I was doing. I did not arbitrarily remove the variables. I took the time to explain.

[@sidaliseghiri1655]

Thank you very much for your videos

[@CrunchEconometrix]

Hi Sidali, thanks for the positive feedback. Deeply appreciated! Please may I know from where (location) you are reaching me?

[@kenechukwujohnpaul3661]

I suppose the outcome of this analysis will be that for a long run ardl model. how does one get the coefficients that will be used to interpret the result?

[@CrunchEconometrix]

Hi Kenechukwu, this video did not cover long-run analysis. It only shows how the removal of irrelevant and redundant regressors improves the result. The final outcome is then interpreted accordingly.

[@ahmadz113]

Thank you as always , regarding the Wald test , shall i put all non-significant variables in one shot ? ex. C(3)=0,C(5)=0,C(6)=0 . because when i run the wald test for each variable separately or with only one other variable , the variable will be redundant but when i run it in a bigger group it will be significant and i can't remove from the model . please advise

[@CrunchEconometrix]

Yes, you can do that.

[@nataliahoffmann9398]

Hi, I have a question. Why there is no variable LNIMP without any lag? I just consider it, because variables LNMVA and REXCH without lags are in the comparison.

[@CrunchEconometrix]

Natalia, your query is unclear. Rephrase. Thanks.

[@mrzadocknanyaro7584]

Hello , thank you for this video . You recommended I watch after my comment on other ardl model video. For the wald test , after ommiting the variables that were not significant(>0.05) I got the probability to be 0.000. does this mean I should just leave them and work with them?

[@CrunchEconometrix]

Hi Mr. Zadock, kindly follow the steps and explanation given in this video. Thanks.

[@TheDominock]

Thank you very much! Am I allowed to apply these steps to panel data too? (Considering I intend to use panel ARDL)

[@CrunchEconometrix]

Best for time series analysis not panel.

[@TheDominock]

@@CrunchEconometrix Thank you.

[@sylviaw4517]

hi， thanks a lot for your video, but I have a question, how could we choose the appropriate lag p,q for ARDL(p,q)? Many thanks!

[@CrunchEconometrix]

I showed how you can allow EViews choose those lags.

[@ruchigupta1418]

Ma'am we can create parsimonious model for ECM model alsoif coefficients are insignificant?

[@CrunchEconometrix]

You have parsimonious model from the ARDL and not the ECM.

[@finomics]

I loved watching this video. I have a question in my mind. if we find the same situation as you performed in this equation which coefficient should I report as a long-run coefficient? since we included lags of independent variables. the level form of the equation only includes the signal coefficient in the regression equation. kindly please reply

[@CrunchEconometrix]

Hi Fida, the G-to-S approach is performed on the "unrestricted ARDL model". After getting the optimal model it is expected that you give a uniform interpretation of the results w/o separating into short- or long-run.

[@zoyashah7826]

Mam I m still stuck with the interpretation of variables in ARDL model..Can you please help me as it is urgent.

[@CrunchEconometrix]

Hi Zoya, I always advise that video tutorials should be supported with readings. Kindly pick and article on ARDL approach, read the sections 3 and 4 for guidance. Regards.

[@zoyashah7826]

@@CrunchEconometrix mam I have already done my part..I have been stuck since 6 months..kindly provide your Email I'd.. so that I can explain you better with my problem

[@CrunchEconometrix]

Zoya, I refrain from personalized tutoring due to time constraints. I have guided you repeatedly on what to do. It is up to you to take the advice. Regards.

[@chrisalwis1626]

There are few videos on ARDL model. Which One I should watch. Pls. specify me.

[@CrunchEconometrix]

"Few videos"? I doubt if any KZbin Channel has my collection of ARDL videos. My advise is that you watch all. Thanks.

[@muhammadivanfadzil5941]

So what is the reason when you choose unrestricted constant in linear trend option? What makes it difference between the others ? Thank you so much for the video ma'am 🙏

[@CrunchEconometrix]

Hi Muhammad, I use it because it is the default option and serves my analytical purpose of having ONE intercept. You will have to read more about the difference among the 5 options. Please may I know from where (location) you are reaching me?

[@muhammadivanfadzil5941]

I'm from Indonesia.. i read about 5 deterministic cases, but none of them really explained the real stories behind it (just formula), not like what kind of variables or data in real economy use case 2 or 3.. i still don't get it..

[@muhammadivanfadzil5941]

Thanks for the answer, but why you used the unrestricted constant rather the restricted one? Do you have another assumptions in your data?

[@CrunchEconometrix]

@@muhammadivanfadzil5941 I'm sure you won't find resources on it and even if you they will be the econometric constructs that are too complicated to understand.

[@CrunchEconometrix]

@@muhammadivanfadzil5941 Please refer to my earlier response.

[@aiboudaziz8318]

please, on what basis we chose the coefficients for the wald test. thank you and good luck

[@CrunchEconometrix]

I gave clear explanations, Aiboud. You may need to watch the clip again. Regards.

[@zoyashah7826]

Mam can you tell me if my 3 variables are correlated with each other,how to resolve this prob.As the variables I have taken is important to the study

[@CrunchEconometrix]

Hi Zoya, kindly watch my video on MULTICOLLINEARITY. Well explained.

[@zoyashah7826]

Thank u mam

[@emirarefa]

Hello mam, thankyou for the clear explanation, it is always fun to hear your explanation. But I have a question, why C(12)rexch(-2) doesn't included in the wald test since it has p-value > 5%?

[@CrunchEconometrix]

Hi Emira, once I got my parsimonious model the rest coeffs do not really matter.

[@emirarefa]

@@CrunchEconometrix thank you for the answer mam

[@TeamBlauEnt]

Thank you very much for your videos. They are very helpful! I have i question regarding the short run coefficients. Is it possible in the final Error correction model just to add each lag up into one coefficient for each variable and use the walt-test p-value? My interpretation then would be: the an x percent increase in the exogenous variable leads to y percent increase/decrease in the endogenous variable after t periods, ceteris paribus. Is it correct? Best regards from Germany :) Mirco

[@CrunchEconometrix]

No. It is wrong to add up the lags. Interpret each coefficient separately. Thanks for watching and sharing!

[@spinebuster9490]

Dear Prof, I am not clear on how you selected the variables for the Wald test. Since 8 were not significant, why did you subject only C(2), C(6), C(7), C(8), and C(11) to the Wald test? Thank you.

[@CrunchEconometrix]

If you observe, the pvalues for these five are quite high. So, I initially started with them.

[@spinebuster9490]

oh, ok. Makes sense now. Thank you. Best tutorial 💪💪💪.

[@kenechukwujohnpaul3661]

@@CrunchEconometrix You just cleared me here. I was having the same reservations. Thanks

[@hssouneabdelkarim626]

Please i would like to know why you choose constant and not trend ? is that related that this variable is stationnary with constant and not in trend or wat exactlly ! please i need your answer ! And your videos are very helpfull my God bless you

[@CrunchEconometrix]

Hi Hssoune, thanks for your query and the positive feedback on my videos. Deeply appreciated. I often use the default because the series is usually first-difference stationary at constant, otherwise, I include the trend. May I know from where (location) you are reaching me?

[@hssouneabdelkarim626]

@@CrunchEconometrix the problem is that when some variables are stationnary in contant and other in trend or none (DS or TS). its a problem to know what u choose in ARDL model that include all variables. I am trying to use also default ... and your videos helped me a lot to success estimations. You deserve support and evrt for your efforts . i am from morocco. Please dont stop and keep going ...

[@CrunchEconometrix]

@@hssouneabdelkarim626 I understand what you mean but it is not a problem. If you have a combination of DS and TS you can still use ARDL which is a single equation model. Just make sure that your unit root table shows the true status of each series.

[@francisowino6985]

Wonderful video. Is this applicable in Stata? Unfortunately i dont have the Eviews Software. Regards.

[@CrunchEconometrix]

Not sure if possible in Stata.

[@surendraist]

A good explanation is given on this tutorial only suggestion little be slow speaking requires, Thank you.

[@CrunchEconometrix]

Thanks Surenda, I will do my best!

[@amirgharaie2448]

Hi prof thank you for your great explanations I have a question: why did u consider 4lags to run this model but in the other videos you consider just 1 lag ? Which one is correct? As you said before because of the number of data (34 obs) we can not use 4lags for running ARDL model I'll appreciate if you make it clear Best regards

[@CrunchEconometrix]

Hi Amir, I said it in the video. 4 lags are you simply to explain the G-2-S approach.

[@zoyashah7826]

Madam I am finding most of the variables insignificant.I have one question that if my r2 overestimates in the likelihood estimation then will I be able to use the significant values in my paper??

[@CrunchEconometrix]

Hi Zoya, the simple unspoken rule is that at least 50% of your coefficients must be statistically significant for your research to be believable.

[@zoyashah7826]

Mam after applying log my most of the variables are now coming significant..But the Error correction model is coming -2.what to do..pls reply

[@CrunchEconometrix]

Hi Zoya, ECM of -2 is not good. Change control variables and re-estimate the model.

[@zoyashah7826]

Thank you I will do that and let you know the final result.

[@MoonsUniverse140]

Hello Dr Ngozi, do we need to do the Wald test for cointegration or bound test is enough

[@CrunchEconometrix]

Hi Mohd, the bounds test is enough.

[@MoonsUniverse140]

@@CrunchEconometrix thanks for your feedback, what is the use of Wald test, what is it used for

[@CrunchEconometrix]

@@MoonsUniverse140 Are you implying that I used the Wald test without explaining its essence?

[@MoonsUniverse140]

@@CrunchEconometrix no Dr Ngozi, I got confused because of different explanations so I wanted to have your expert opinion

[@CrunchEconometrix]

@@MoonsUniverse140 I'm not an expert. No one is an expert. We are limited to the extent of what we know. Kindly watch the clip again. It is well-explained. Thanks

[@victoriaisere109]

what do i do when my bounds test show no cointegration? does it mean no long run relationship? do i just do only the short run model i.e. ARDL

[@CrunchEconometrix]

Yes Victoria, it means just that. Hence, estimate ARDL model and not the ECM.

[@victoriaisere109]

@@CrunchEconometrix thats very difficult to defend in seminar presentations. whats the minimum sample size for an ARDL model

[@CrunchEconometrix]

@@victoriaisere109 Read Pesaran and Shin and several tons of ARDL papers. References at the end of the video. 30 is the minimum sample size. You can confirm that from any econometrics textbook.

[@ehoumanjacquesallou9803]

Hello teacher, really appreciate your videos. I'm now working on ARDL model and in my variables I put dummy variable. Can you please explain to me how to run ardl model with dummy variable. Thank you

[@CrunchEconometrix]

EHOUMAN JACQUES Allou HI Allou, obtain the lags of the DV and that of the interaction terms and put in the model. Run the analysis with the obtained lags and another with 0 lags... then use the model with the best estimates. Most times, the DVs have 0 lags....hope this helps. By the way, what software are you using?

[@ehoumanjacquesallou9803]

CrunchEconometrix I'm using eviews 9

[@ehoumanjacquesallou9803]

Hello Teacher, i have another questions : the lag of my DV is 0 my question is should i use the lag of DV in different equation or put lag the dependant variable also in the DV?. for example this is my equation when LnINFLA is my dependant variable my long run equation is : lninfla c lninfla(-1) lninfla(-2) lnhdi(-1) lnhdi(-2) fdi(-1) fdi(-2) lngovspend(-1) lngovspend(-2) lngini(-1) lngini(-2) lnopen(-1) lnopen(-2) dummy??? and my shor run equation is d(lninfla) c d(lninfla(-1)) d(lninfla(-2)) d(lnhdi(-1)) d(lnhdi(-2)) d(fdi(-1)) d(fdi(-2)) d(lngovspend(-1)) d(lngovspend(-2)) d(lngini(-1)) d(lngini(-2)) d(lnopen(-1)) d(lnopen(-2)) dummy??? ecm(-1) . the optimum lag of inflation is 2 . so should i put dummy (-1) dummy(2) anr run the model or put is lag 0 in all the equation ? and also should i make another model and use dummy as depend variable? Thank you so much

[@CrunchEconometrix]

hi Allou, if the lag of the dummy variable is 0, then use it like that. Most times, the lag for dummy variables range between 0 and 1....in my opinion, depending on the research question, the dummy variable can also be a dependent variable.

[@CrunchEconometrix]

....remember to include a lag of the dummy variable if you are using it as a dependent variable. This is because an autoregressive model requires that the lag of the dependent be included as regressors.

[@ukuk9162]

Many thanks for your help

[@CrunchEconometrix]

No worries, Aziz... u're welcome.

[@mercyonyango4325]

what is the next step after getting the parsimonious equation

[@CrunchEconometrix]

Hi Mercy, you interpret your results and perform diagnostics accordingly. May I know from where (location) you are reaching me?

[@mercyonyango4325]

@@CrunchEconometrix from Kenya. Thank you for the reply. My question was since i have changed from ARDL to OLS when getting the specific model(parsimonious equation) , do I now do Johansen test for cointegration or go back to the original ARDL model and do bound tests for cointegration?

[@CrunchEconometrix]

@@mercyonyango4325 You should know by now that JCT is not applicable to ARDL models. Besides, if you use optimal lags and avoid multcollinearity, over-parameterization will not occur.

[@dr.sureshmago9211]

Ma'am, I have panel data of three variables (FDI, GDP and Exports), all in log form. One is stationary at level and two stationary at first difference. When I apply ARDL on these variables then eviews is showing error of "Near singular Matrix". Ma'am please help me to come out of this problem.

[@CrunchEconometrix]

Hi Suresh, you have a panel data but this thread is about time series ARDL... Pls post appropriately.

[@dr.sureshmago9211]

@@CrunchEconometrix Hello Ma'am, you mean, we can't apply ARDL on panel data? Please guide me, I need this for my Ph.D work.

[@CrunchEconometrix]

My response does not say that. You posted wrongly.

[@CrunchEconometrix]

You should have posted it on the panel ARDL videos.

[@dr.sureshmago9211]

@@CrunchEconometrix Ma'am please help me to come out of this problem. I want to examine relationship among three variables: FDI, Export and GDP for the panel data of 12 countries, for the period of 39 years. Please suggest me, Can i apply ARDL if one variable is I(0)and two are I(1).

[@ukuk9162]

Good afternoon, I have question not related to this vedio please in order to estimate money demand what is appropriate aproch I should use real GDP and real M2 and ipc or I use nominal GDP and nominal M2 ?and why .

[@CrunchEconometrix]

Use variables in "real" terms to accommodate inflation. I actually thought I responded to this. My sincere apologies 🙏

[@zoyashah7826]

Mam my data is of annual 28 yrs so will I be able to apply ardl or I have to increase the sample size

[@CrunchEconometrix]

Hi Zoya, T > 30years is preferable. But I have come across studies with T < 30. So, if your results are good with T = 28, why not?

[@zoyashah7826]

No I am getting insignificant results with 28 years of data..should I increase the sample size more than 30??

[@zoyashah7826]

And please tell that in eviews student version 11..there is no option of add-ins or to test structural break..What to do in that situation??

[@Dr.IsaacJacobOmosimua]

Thanks thanks

[@CrunchEconometrix]

You are welcome, Isaac😊

[@elvinsindala1927]

I cant see some of your past videos being referred to.

[@CrunchEconometrix]

Hi Elvin, all my videos are listed. Kindly check through again. Thanks.

[@eboviwali9610]

Nice one ma

[@CrunchEconometrix]

Thanks Ebovi for the kind remarks!...simply tell your colleagues to stay focused and watch my clips relating to VAR and ARDL models, they'll surely get 99% of their worries cleared away (lol)

[@CrunchEconometrix]

Thanks Ebovi for the kind remarks!...simply tell your colleagues to stay focused and watch my clips relating to VAR and ARDL models, they'll surely get 99% of their worries cleared away (lol)...and remember to send your data to [e-mail]

[@CrunchEconometrix]

Yeah, u're right. But that will be when they subscribe to take advantage of the info available here. Until then...I'll keep sharing the videos. Thanks as always!

[@CrunchEconometrix]

Azam Thanks a great deal for the support. Honestly, I've received similar from gestures from other subscribers. You can help by sharing my videos and links whenever you see the opportunity even on Hossain and EconoMetrics platforms...and whenever I find another welcoming Group I'll let you know. Thanks!

[@ukuk9162]

I want to say beyond the effect of inflation why use real data instead nominal are there economic explanation?

[@CrunchEconometrix]

Advisable to use real variables to control for inflation.

[@haider821d]

plz madam answer me?????? can inflation and unemployment explain economic growth in the model

[@CrunchEconometrix]

@@haider821d Yes they can only you have to read more about how.

[@CrunchEconometrix]

@@haider821d Yes they can only you have to read more about how.

[@haider821d]

CrunchEconometrix Thank you Madam! But there is no relevenat literature on the impact of inflation and unemployment on economic growth.....

[@alisajid2400]

Thank you........

[@CrunchEconometrix]

U're welcome, Ali 😊. Please may I know from where (location) you are reaching me?

[@alisajid2400]

@@CrunchEconometrix From Pakistan...... I was stuck in my ARDL model as it did not show any cointegration and most of the variables were insignificant at level... so I was looking for scientific reasoning to validate my claim.....e.g FDI(-1) is significant but FDI is not keeping Stock market development as a dependent variable....... My prof told me that FDI takes some time to to take its effects on stock market...but to prove it scientifically was difficult for me....you video proved blessing for me....

[@CrunchEconometrix]

@@alisajid2400 Wow! I'm so glad you found what you needed to improve your study. Kindly assist to share the link to my KZbin Channel with your friends and academic community in Pakistan 🇵🇰. They will find the content helpful too 😊.

[@alisajid2400]

@@CrunchEconometrix One question...my F-Statistic from Wald test and variable redundancy test are not coming same...Wald test has 72% while redundancy test is 78%...is this normal? or i should take only one approch?

[@CrunchEconometrix]

@@alisajid2400 I don't quite understand your query. Can you be a little more specific?