This is How to Specify ARDL Models

Upon performing the bounds cointegration test, there are two (2) likely outcomes: either the variables are cointegrated or they are not. If the variables are not cointegrated, the next thing to do is to specify the short-run model, which is the autoregressive distributed lag (ARDL) model but if cointegration is the outcome, then the appropriate model to specify is the error or vector error correction model (ECM/VECM) as the case may be. This video details the different model specifications under the ARDL framework.
Follow up with soft-notes and updates from CrunchEconometrix:
Website: cruncheconometr...
Blog: cruncheconomet...
Forum: cruncheconometr...
Facebook: / cruncheconometrix
KZbin Custom URL: / cruncheconometrix
Stata Videos Playlist: • (Stata13):Estimate and...
EViews Videos Playlist: • (EViews10):Interpret V...

[@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.

[@muhammadarshad2734]

CrunchEconometrix I’ve benefited from your videos in a great deal. I would like to thank you for this. Your videos will have long-run causal effect on my understanding of econometrics! Btw, I’ve subscribed your Channel and will suggest the same to my peers as well. Stay blessed.

[@aremoadeleke1397]

I am well impressed by your pedagogical skills and depth of knowledge of applied econometrics. Despite that I teach econometrics at the most advanced level I still derive optimal academic inspiration from what you teach. I have since subscribed and I passionately appeal to others, including all my students, to subscribe. Well-done.

[@CrunchEconometrix]

@@muhammadarshad2734 Sorry for my belated response. I am so happy that you find my videos very helpful...and thanks for suggesting my KZbin Channel to your colleagues. May I know from where (location) you are reaching me?

[@CrunchEconometrix]

@@aremoadeleke1397 Sorry for my belated response. I am so happy and encouraged that you find my videos very helpful...and thanks for suggesting my KZbin Channel to your Students and colleagues. May I know from where (location) you are reaching me?

[@habtamulegese6084]

how can I get your ppt ? I hope you can send it to me through my email [e-mail]

[@adebayojjulius6056]

Dr Ngozi, I can't thank you enough for the great and wonderful knowledge u are imparting to students and d society, two days ago I presented my MSc thesis in economics and I was awarded an A which was also d highest among all other presenters. Not only that, I broke a record of such high grade in thesis at d department as the professor of economics confirmed. I couldn't have done this but I made ur channel my best friend, used it as a guide to analyse my data by myself with excellent interpretation of results. Thanks so much, God bless u. Plz continue d good work.

[@CrunchEconometrix]

Thanks so much for your encouraging feedback, Mr. Julius. Huge congratulations!!! 🥂🍾😍

[@LeCassiopeia]

Hi I am also working on a thesis right now. It would be interesting to see a well applied example of the ARDL model in a master thesis set up. Please reach out to me if you want to connect and share thesis or insights.

[@pedrocolangelo5844]

Ms. Adeleye, thank you for another great video. You have one of the best KZbin channels on the subject of Econometrics. Greetings from Brazil!

[@CrunchEconometrix]

Thanks, Pedro for your positive feedback. I am encouraged. Much love from Nigeria 🇳🇬 💕❤️

[@eswatinimyhome6859]

Honestly you are a life saver, wow, also the way you explain is so easy to follow. God bless.

[@CrunchEconometrix]

You're welcome! Thanks for your encouraging feedback...deeply appreciated!

[@infinity3668]

Thank you so much for your excellent videos. I am currently in my undergraduate studies , but I found your videos very helpful. A huge thanks Mommy, love from the Gambia.

[@CrunchEconometrix]

You're very welcome, Sir!

[@thekrejavu810]

This is the best channel ever. Very explanatory and informative.

[@CrunchEconometrix]

Hi Krejavu, I humbly accept the commendation. Deeply appreciated! May I know from where (location) you are reaching me?

[@1Nchimunya]

I learnt more about econometrics from your videos than in class...Thanks Crunch..cheering from Zambia 🙏🙏

[@CrunchEconometrix]

Thanks Nchimunya, for the encouraging feedback. Much love 😍 and gratitude from Nigeria 🇳🇬.

[@nnamdiohachenu7835]

Watching this took me back to being in your class in CU. I remembered how you would switch to explaining in pidgin to us when we are having a hard time understanding the topic being taught lol. Very informative video. Thank you ma.

[@CrunchEconometrix]

Hahahaha, Nnamdi. Thanks for taking me back to memory lane. Amazing experience 👌

[@funshoolu]

Excellent presentation

[@CrunchEconometrix]

Glad you liked it...thanks! 🥰🙏

[@anthonyosobase1999]

Good job ma. Thanks a zillion for uploading. Your explanations are quite clear in all your videos. Once again, Kudo.

[@CrunchEconometrix]

Thanks Anthony for the encouraging feedback. Deeply appreciated! May I know from where (location) you are reaching me?

[@anthonyosobase1999]

Lagos, Nigeria, Ma.

[@mishalkhaled8327]

thanks a lot for excellent video one of the best in KZbin

[@CrunchEconometrix]

Hahahaha, Jasem thanks for the encouragement and positive feedback. Deeply appreciated! 💕 😊

[@hassanaumar419]

Just came across your videos and they are amazing!!

[@CrunchEconometrix]

Thanks Hassana for the positive feedback. Deeply appreciated! Kindly share my Channel with your fellow students and academic networks because many of them are not aware that CrunchEconometrix exists...may I know from where you are reaching me?

[@MKH24-c3f]

Thank you. your analysis is very helpful.

[@CrunchEconometrix]

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

[@MKH24-c3f]

@@CrunchEconometrix from Bangladesh. But I am having phd in China, thanks

[@olufunkeajoje2285]

Filled with so much lessons.

[@CrunchEconometrix]

Thanks!

[@naymanadia8625]

Very effective video. Thank you

[@CrunchEconometrix]

Thanks Nadia, for the positive feedback...may I know from where (location) you are reaching me?

[@orkeer]

Thank you so much for this video!

[@CrunchEconometrix]

U're welcome Orkeer...may I know from where you (location) are reaching me?

[@carl-henrycadet2297]

very well explained, very helpful

[@CrunchEconometrix]

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

[@carl-henrycadet2297]

@@CrunchEconometrix from Germany.

[@CrunchEconometrix]

@@carl-henrycadet2297 Awesome! I'll appreciate it if you can share the link to my KZbin Channel with your students and academic community in Germany 🇩🇪...thanks! ❤️ 🙏

[@mohammedalnour318]

Thank you very very very much for the explanation

[@CrunchEconometrix]

Thanks, Mohd for the positive feedback. Deeply appreciated! May I know from where (location) you are reaching me?

[@mohammedalnour318]

@@CrunchEconometrix I am a Sudanese, I study a PhD in Economics at Erciyes University, Republic of Turkey

[@jamyangnurboo5448]

Recently i saw your videos and it is very helpful thank you so much.ç

[@CrunchEconometrix]

Hi Jam, thanks for the positive feedback...may I know from where (location) you are reaching me?

[@jamyangnurboo5448]

@@CrunchEconometrix From India Mam.

[@CrunchEconometrix]

@@jamyangnurboo5448 Awesome! Please spread the word about my videos....thanks! 💕

[@jamyangnurboo5448]

@@CrunchEconometrix sure.

[@samsouma3958]

u re the best ....thanx crunch

[@CrunchEconometrix]

Hahahaha, Sam. I'm not the best only sharing the little I know with the global academic community. Thanks for the encouraging feedback. Deeply appreciated!

[@juliebasconcillo]

Hi! Thank you for the helpful tutorial video. I'm not sure I understood why you had to estimate three ARDL models where you jumbled the dependent variables.

[@CrunchEconometrix]

Hi Julie, thanks for the encouraging feedback. I gave clear explanations on why I did that. You may want to watch the clip again. Thanks

[@juliebasconcillo]

@@CrunchEconometrix oh my bad. I will watch the entire video. Thanks!

[@mateusfigueiredo9961]

Thank you very much!

[@CrunchEconometrix]

U're welcome 🙏

[@mynameisjoejeans]

Great video thank you. If one of my variables isn't significant, should I drop it from the equation I report?

[@CrunchEconometrix]

No. That would be result falsification. All coefficients from a regression must be reported

[@tesfayefikadu7058]

I like very much this video, Thank you

[@CrunchEconometrix]

Thanks Tesfaye, for the positive feedback. Please share my videos with your colleagues and may God bless you!

[@jolaoshooluwaseun1357]

Nice video tutorial but I'm a bit confuse about the first ARDL model equation. I didn't see the current value of the regressor before the lag value was included.

[@CrunchEconometrix]

Thanks, Jolaosho. If it's not specified, then it's an omission.

[@lucifergaming2229]

Hello Prof Bosede, i have questions related to ardl. My optimal lags are 2 for all variables. Now my questions are: 1) in ARDL long run and bound tests table, i have different values for each variable in the table of eviews 10. Suppose, VAR1(-1), D(VAR1(-1)), and D(VAR1). Now for short run results, which values among these for one variable i.e. VAR1 I should report in my short run table? 2) In my ECM regression: there are only C, D(VAR1) and CointEq(-1)*. Only one variable D(VAR!) is shown in the ECM regression. No other variables are there in the ECM regression as I have 3 other independent variables apart from VAR1. Why is that so? Why other variables are not in that regression? As i have seen in many tutorial, in ECM, other variables are present. I hope to see your kind answer. Thank you My second question is:

[@CrunchEconometrix]

Report all the results as shown from EViews. You have only the ECM showing in the SR part of the results because the variables appear to have zero lags.

[@yuqingwu1011]

perfect video, and a question: if the variable is stationary at first difference, why we still use the original variable, not using the variable at first difference in bounds test and ARDL model? I’m really confused about this

[@CrunchEconometrix]

Hi Yuqing, thanks for your encouraging feedback. Kindly read the Pesaran articles listed at the end of the video for detailed information about the ARDL model.

[@idakwojiblessing4475]

Good morning Prof, pls is there any video on Non linear ARDL? I would like to learn about it.

[@CrunchEconometrix]

No, I don't have any videos on that.

[@lombechangala2803]

Hello,great videos. can i use ARDL if the dependent variable is stationary at level.?

[@CrunchEconometrix]

You can use either ARDL or OLS.

[@popoolapraise3184]

Good day, what can I do If after user using the ARDL technique my variables do not have a long run relationship?

[@CrunchEconometrix]

Estimate the unrestricted ARDL. I explained this in the ARDL video "this is how to specify ARDL" please watch it. Thanks

[@ajayant.s604]

which lag method is need to use for ardl... Maxlag or varsoc optimum lag..

[@CrunchEconometrix]

Hi Ajayan, use either. I demonstrated both in my ARDL videos.

[@farhanfarzam4278]

Thank you dear ma'am for your insightful lessons. I would like to ask you. Are you familiar with Mr. Mohammad Musa Shafiq, I see his name as a Co-author in some of your papers. He was my friend and also he was my professor in University. Best of luck ma'am.

[@CrunchEconometrix]

Yes, I do. We co-wrote an article.

[@iwasamegbe6177]

thank you Ma for this video. Please, must i run the three variables as independent variables each even if they are not independent variables in my model?

[@CrunchEconometrix]

Iwasam, ARDL is a single equation model so, not really. Thanks.

[@yasmineel-hady4777]

Can You help me with a question? If all my variables are stationary at first difference NOT at level, SHOULD I REPLACE (ardl y x1 x2) with (ardl d.y d.x1 d.x2)? If there is no cointegration OR it doesn’t matter if I run it at level or at first difference, since my diagnostic tests are good when running ardl at level OR these diagnostic tests are invalid? Thanks in advance

[@CrunchEconometrix]

Hi Yasmine, my videos on ARDL are detailed and well-explained. You may want to follow the steps as shown. Thanks.

[@andliona9981]

You're tutorial is more helpful thank you so much. But I have a question how would I know if am supposed to use ARDL model in my objectives is there any criteria used? Could you please help me

[@CrunchEconometrix]

Hi AndL, thanks for the encouraging feedback. I mentioned in the clip that ARDL is a single-equation model. There are several publications that used the ARDL technique and some are listed at the end of the video (including my paper). Read them particularly the INTRODUCTION and Section to know how to tidy up your objectives with the methodology. I hope you find this feedback helpful. Thanks.

[@philipthompson8904]

Mum is it ok to use in the regression real and nominal series like you did in the explanation

[@CrunchEconometrix]

Hi Philip, "real" values are adjusted for inflation. So, I don't quite understand your query.

[@lagipeters7863]

What to look at in the unit root test to see whether it is stationary ot not?

[@CrunchEconometrix]

Lagi, watch my videos on stationarity tests. Thanks.

[@Luther124]

Thank you very much for the insightful video. I have a question and I'm hoping you can assist me. I recently read that an ARDL model requires one cointegrating relationship. I have one dependent variable and one independent variable (hence i have two equations). Does that mean if use the bounds test to detect cointegration in both equations the ARDL does not apply? Your assistance would be highly appreciated.

[@CrunchEconometrix]

Hi Luther, ARDL is a SINGLE EQUATION model as explained in the video. So, if you have one depvar and one indpvar then you have just ONE ARDL model to estimate and not two.

[@Luther124]

@@CrunchEconometrix Thank you very much for your response. Just to clarify, don’t I have to check cointegration when each variable in the model is treated as the dependent variable? Then if I detect cointegration in this case doesn’t that mean there is more than 1 long run relationship which goes against the assumption of the ARDL?

[@CrunchEconometrix]

Luther, you are confusing things. I have an ARDL video where I used each variable in the model as a depvar. I also explained why i did that. You may want to watch it for detailed understanding. Thanks.

[@Luther124]

@@CrunchEconometrix I think maybe I might be expressing what I’m trying to say the wrong way (forgive me please...I’m new to econometrics😀). So let me try it this way - in the case where my model consists of 2 vectors as variables and I treat all of them as dependent variables, and I find cointegration in both equations. I take it I have to specify a VECM not an ECM? Does this then mean I have to do a VAR model (instead of an ARDL model) in order to do a VECM? (apologies if I’m missing the points you made earlier)

[@CrunchEconometrix]

Luther, my response: ARDL: will give you 2 separate single equation models and corresponding cointegrating relationships. VAR: will give you a system of equations with ONE cointegrating relationship normalized on the depvar listed first in the system. I have detailed videos on these.

[@aprilmaecadungog77]

Hi, is it okay to run ARDL if all my variables are stationary at all levels? I only have 2 variables, one independent and one dependent. with 30 years observation. Thanks.

[@CrunchEconometrix]

April, use OLS if both series are I(0) stationary.

[@aprilmaecadungog77]

@@CrunchEconometrix Thanks, CrunchEconometrix! Do you have a video guide and steps for this method?

[@CrunchEconometrix]

Yes, April. Kindly search through my Channel for "Cointegration" you see those for I(0) series.

[@owusuopokumensah9796]

Hello Prof. For ARDL model, please suppose you do your panel unit root test and realise that your variables are stationary at first difference, do you have to generate a series to convert them to first difference before you perform your regression or you perform the regression on the original data but eviews will conduct the regression but will consider them at first difference. Please help.

[@CrunchEconometrix]

Hi Owusu, I have several videos on ARDL kindly watch them to see what I did. Thanks

[@markmashilana2621]

Hi Doc, talking about the ECM, on one of your videos you mentioned that if the ECT is positive one can proceed with the interpretation (meaning it doesn't matter whether its positive or negative). How do you harmonise that with what you're saying on this video. Thank you.

[@CrunchEconometrix]

Hi Mark, this video provides the basic understanding of the ARDL model. When you analyze your data you MAY end up with a positive ECT. So, interpret your results accordingly. Thanks

[@markmashilana2621]

@@CrunchEconometrix Is still considered explosive though 🤔, and would the results be valid?. Thank you once more.

[@CrunchEconometrix]

Mark, getting a positive ECT is a function of several factors: data, scope, variables, the economy under investigation etc. You can EITHER go online to get articles with positive ECT to support your results OR you re-estimate your model by changing variables, lags etc which may amount to results manipulation...but you decide on that.

[@maheshehu7943]

Good day Bosede!. Nice watching your youtube videos. they are quite informative and rich in contents, I'm running an ARDL model with one (1) Dependent variable and four (4) independent variables, using annual time series. Now, if I perform estimation techniques of Unit root tests (ADF, and PP), cointegration test, long run coeffeifficient and short run coefficients tests, granger causality test and finally perform some checks by way of diagonistics. Do i need to do anything further again? Also, i'm looking at 3 different countries and i wouldnt like to do a panel ARDL. could you pls. give me your advice on this?

[@CrunchEconometrix]

Hi Shehu, you are on point on all counts. Do a comparative time series analysis and not panel analysis. Thanks for watching and the positive feedback, deeply appreciated!

[@swarnachari2703]

Hi

[@swarnachari2703]

Hi shehu...u did analysis ..got results

[@maheshehu7943]

Swarna Chari good day. Yes I did the analysis but need to work on it again

[@maheshehu7943]

CrunchEconometrix Hi aniseed! Please, I have done my estimates and analysis of the results. The variables turned out to be in mixed stationary. That means based on the result of stationarity obtained the ARDL approach is suitable. Pls. My question is could it be possible to go further and perform VAR and CECM Models based on what I have? Because on the countries I got one of the countries two variables are stationary at level and 3 variables are stationary at first difference. While the other two countries one variable each are stationary at level and the other 4 variables of the two countries are stationary at 1st difference . Regards. Shehu

[@saraijacobs6966]

What do you do if your f statistic is only greater than the upper bound for 3 critical values and not 4? Does that mean we accept a long run relationship among variables in the model or not?

[@CrunchEconometrix]

I don't understand what you mean by "3 critical values and not 4".

[@NusratJahan-eb6ox]

Thank you for your helpful video. In my model the coefficient of adjustment term came out with a positive sign. How can I interpret this?

[@CrunchEconometrix]

Hi Nusrat, you give the opposite interpretation to a negative sign.

[@benjamingawa169]

Hello Professor! I would like to estimate the ardl model in panel. However, after the stationarity test that I received with stata 15, I blocked for the choice of lags of the variables. I used the command to see in the attached screenshot. When I use three variables, I have the result (capture 1) but more than three variables the stata software notes me the collinearity error (capture 2). However, my ardl model has twelve explanatory variables (screenshot 3 and 4). My teacher's problem comes down to the choice of lags of the variables in the panel with stata. In the expectation of a favorable outcome, please, receive Mr. Professor, the expression of my highest considerations. Respectfully Benjamin GAWA

[@CrunchEconometrix]

Your queries are confusing. Unable to understand what you need to do. Kindly recast. Thanks.

[@abidemisomoye]

Hello Dr. Ngozi. Thank you for the great work you are doing. We are so grateful. Please I want to ask. Is the ARDL method specified at the level or at its differenced form? Thank you.

[@CrunchEconometrix]

Level. It is the ECM representation that is specified in 1st diff.

[@abidemisomoye]

@@CrunchEconometrix Thank you so much. It's clear.

[@thebullet8538]

Gracias ❤❤

[@CrunchEconometrix]

U're welcome! 🥰🙏

[@corinaniculaescu8659]

What can be done if the variables are not stationary at levels, nor after the first (or second difference)? Thank you!

[@CrunchEconometrix]

Hi Corina, discard and use alternative proxy variable.

[@aaronnzou4537]

You can consider the Toda Yamamoto criterion

[@shintalestari6634]

How to determine the lag of each variable in ardl model? I still confused to enter the minimum and max lag in eviews. Thank you very much

[@CrunchEconometrix]

Shinta, kindly watch my EViews-based ARDL videos. I recall I showed the procedure in some of them.

[@karnikaabhattacharyya407]

Ma'am, the original paper of Pesaran 2001 has considered the difference operator from lag 0 in case of independent variable but you have considered it from lag one. Why is that?

[@CrunchEconometrix]

I'm glad you pointed out that, Kamikaa. I made a typo. It should be t-i and not t-1.

[@karnikaabhattacharyya407]

No ma'am while running the short run ARDL model, you estimated the independent variables by taking difference operator from lag 1 and not lag 0. For example, you run the model by specifying d(x1(-1)) and so on.. you didnot start with d(x1) for the independent variables. Thats what I wanted to know. Thank you.

[@Waleedkhancooldevil93]

Nice video. Can i take the log of the variables such as gdp growth, gdp per capita and gross domestic income?

[@CrunchEconometrix]

Hi Waleed, take log of the last 2 variables.... GDP growth is already the log-difference of GDP.

[@sarahahmedchawsheen5455]

hi, thanks for your nice clarification. Would you do a video about panel data heterogeneity test of Hsiao in Eviews, please?

[@CrunchEconometrix]

Thanks, Sarah. Suggestion is well noted.

[@DjakaliaOUATTARA-s6z]

Hi, Thank you very much for all the teachings. It's always a pleasure for me to listen to you. However, I'm using a non-linear ARDL model in a panel. Can you help me with the stata code or a guide to estimate ? Also, I would like to know if it's possible to use a dummy variable. Thanks

[@CrunchEconometrix]

Hi Djakalia, thanks for your encouraging feedback. Deeply appreciated 🙏. As you would have observed, I do not have videos on NARDL. I am yet to use the technique and understand its assumptions. So, you may want to check out other online resources.

[@muhammadmurtaza3870]

Hi Ma. thanks for the great video. I have a question that in first you show the ARDL model in which the dependent variable is without difference operator and after the bounds test with difference operator. What is the difference? If we need a differenced dependent variable, it means that we cannot apply ARDL when our dependent variable is level stationary. what do you say?

[@CrunchEconometrix]

Hi Muhd, you are mixing this up. Watch the video again to see the difference between the econometric specifications of the ARDL, Bounds test and the ECM. They are not the same. I will also encourage you to read the listed references at the end of the video and articles that used the model for more understanding. May I know from where (location) you are reaching me?

[@muhammadmurtaza3870]

@@CrunchEconometrix greetings from Pakistan.

[@CrunchEconometrix]

@@muhammadmurtaza3870 Awesome Muhd! 💕 I've come across so many Pakistanis. Good-spirited and loving people. Kindly share my KZbin Channel link with your students and academic community. Thanks 😊

[@mjahircu]

I do not understand the meaning of lnmva you considered as dependent variable and what types of cointegration you wanted to find out among dependent and the independent variables. please a text file of your video. Thank you very much in advance for your reply. Actually i am working with the governance index, GDP, remittance inflows, export, exchange rate, inflation and their cointegration with the GDP growth of Bangladesh. please give me a ARDL model for my work if possible.

[@CrunchEconometrix]

Hi Mohd, you may have to watch the clip again for a better understanding. Also, go through the references listed at the end of the video.

[@MohamedEsmailAlazraq]

Thank you

[@CrunchEconometrix]

U're very welcome, Mohd. May I know from where (location) you are reaching me?

[@tomaseduardochungaseminari5605]

Another question in the case that I only want to operate under a returning, I have 4 variables but the returned one in my 3 models that I will do is the same, what should I apply and how would I use ecm and ardl, one of them is not cointegrated in the bound test .

[@CrunchEconometrix]

I'm having some difficulty understanding your questions.

[@kenechukwunwisienyi6262]

Hello Dr, I am currently working on my PhD dissertation. My Supervisor directed me to disaggregate my dependent variable (FDI) into 3 different sectoral allocations (FDI for Primary sector, manufacturing sector and oil and gas sector respectively). this will make my work have 3 dependent variables. Please is there a cointegration test that can run such an analysis with 3 dependent variables? I don't want to run it separately. Your response will be appreciated.

[@CrunchEconometrix]

Hi Kenechukwu, the unit root test guides you on the appropriate cointegration test to deploy.

[@kenechukwunwisienyi6262]

@@CrunchEconometrix ok Dr. Thanks. So if I got you correctly, what it implies is that any of the cointegration tests can be deployed on an analysis with more than 1 dependent variable. It all depends on where your unit root test points you at, right?

[@CrunchEconometrix]

Yes. Without testing for a unit root you can't determine the cointegration method to use.

[@kenechukwunwisienyi6262]

@@CrunchEconometrix Ok. Thanks Dr.

[@missstar8532]

Greetings Ngozi, l am looking at cointegration between macroeconomic variables, microstructure variables and Bitcoin price using the ARDL bound test because l have a mixture of I(1) and I(0) variables. My results show the presence of cointegration in my series, hence l carried out ECM to test for both the long run and short-run relationship but the results are indicating a long-run relationship in some of the results whereas for the other variables which are not significant the results are not showing the short-run results. How can l get the short-run relationship for my model? I have ten variables and three of these variables are not significant. I want to test for the short-run relationship of those three variables but my model is only showing short-run relationship for only two variables which are significant in terms of a long-run relationship. How can l obtain short-run results for all the variables and why is my model only giving me a short-run relationship for two variables out of ten variables that l have? thank you

[@CrunchEconometrix]

Hi girl, I'll pass your query over. TLDR. Sorry.

[@울산도시]

thank you for valuable lecture. I hope it can solve my model of which consists of mixture of i(0) and i(1) with cointegration

[@CrunchEconometrix]

It will.

[@fuzhufeifei]

Dear and respected Ngozi, After ARDL bond tests, if variables are cointegrated, we will use the error corrected model, not VECM model. Am I right? But in this video, Ngozi wrote long-run (VECM) (in video 2:37). Is it a mistake or actually is true? Despite this small thing in the slide, in fact, I heard that Ngozi mentioned ECM model as well. But for the slides, here is VECM model. From what I learnt so far, I think that it is a mistake. VECM model should not be here. However, I firmly believe that Ngozi knows better than me. Please forgive me if I said anything inappropriate.

[@fuzhufeifei]

Dear Ngozi, I realized that it might be true to apply VECM model after bond tests as I have watched to the end :) But in order to perform VECM model, we have to make sure that the series is all I(1). I used to think that VECM and ARDL will never be connected with each other. This is because if we choose to use ARDL, it shows that our variables are not strictly I(0) and I(1).

[@CrunchEconometrix]

Not a mistake. Listen again to what I said. "VECM or ECM as the case may be"...

[@fuzhufeifei]

@@CrunchEconometrix Got it, dear Ngozi. Indeed, it is not a mistake. I have to study more to comprehend all those models. Thanks very much for all the support, patience, kindness and guidance. May you always be blessed.

[@CrunchEconometrix]

@@fuzhufeifei Amen and you too, Cherry....I still make mistakes too, nobody knows it all :)!

[@Salma-of8hn]

Great video! Can you please provide a reference/source for the claim that Johansen is no longer valid for series with a combination of I(0) and I(1) variables? I need to cite this for my undergraduate thesis. Thank you in advance!

[@CrunchEconometrix]

Hi Salma, there are references listed at the end of the clip. Thanks.

[@Fm11111]

Thank you so much for these videos it has helped me a lot on my research. After ARDL test and found that there is no cointegration, do we proceed with testing the granger causality as we did in VAR? and what would the command be for that in Stata? thank you so much

[@CrunchEconometrix]

Fonny Muliyadi Yes, you can test for causality. All dofiles used in my tutorials are on my website. The link is at the end of the video...and thanks for the kind remarks!👍🏽

[@chikokamanga6859]

Thanks so muuuchh

[@CrunchEconometrix]

U're welcome, Chiko! 💕 😊. May I know from where (location) you are reaching me?

[@Biotechnist]

how can I cite the modelling procedure which you've described? is there any references?

[@CrunchEconometrix]

Hi Bio, you can cite Pesaran, Shin and Smith (1999, 2001) and published paper that used ARDL.

[@TheEnyoy]

In last video you performed a bounds cointegration test on ARDL model, so it is not possible to perform cointegration test before taking ARDL or VECM?

[@CrunchEconometrix]

I don't understand your question...you may need to watch the video on Bounds test on the explanations.

[@Dr.Ateeb-Economist]

what is lnmva entering eq 2 and 3 with t-1? shouldn't that be t-i? Very helpful. Thanks - awaiting response!

[@CrunchEconometrix]

Hi Syed, it's a typo should have been t - i. Thanks for the observation. May I know from where (location) you are reaching me?

[@bbouchra1000]

Thanks so much. Please Prof, what can we conclude when we get an error correcting term significantly negative and with a value of -2

[@CrunchEconometrix]

U're welcome. It shows the speed of reversion to long-run equilibrium is 200%. Redo your analysis again. Change your control variables.

[@bbouchra1000]

@@CrunchEconometrix thank you so much for your help

[@ishasinha947]

Beautiful explanation!! Ma'am, I had a doubt about the interpretation of the ECT result. My ECT is -2.111 and significant at 1% can i say that this high adjustment coefficient (in absolute term) indicates a faster adjustment process and the results imply that 211% disequilibrium of the previous year’s shocks are corrected back to the long-run equilibrium in the current year or is it 2% disequilibrium of the previous year’s shocks are corrected back to the long-run equilibrium in the current year?? Thank you so much

[@CrunchEconometrix]

Hi Isha, re-estimate your model. No one will take your results seriously with ECM of -2.11.

[@ishasinha947]

@@CrunchEconometrix Ma'am I re-ran the model by dropping some of the variables and now ECM is -1.26. Is it an acceptable value? Can I proceed with this value? Or should ECM typically range between 0 and -1?

[@CrunchEconometrix]

Isha, this can be better but acceptable because the value is greater than -2. There are articles with such ECM. Search and use them to support your outcome.

[@joselopes4588]

Dear Dr. Ngozi, I used a panel ardl model to study the relationship between economic growth and trade openness. My PMG estimation confirms a positive relationship between these variables. However, when I analyse the PMG heterogeneous estimates (by countries), I realize that in six of the fourteen variables this relationsip is negative. What should it mean? Is it correct?

[@CrunchEconometrix]

Hi Jose, the countries are different so, the results are expected. You need to read the literature extensively in addition to understanding the economic dynamics of these countries to bring out plausible reasons.

[@forexmaster6142]

Its all nice. Do you have any video regarding Markov switching model?

[@CrunchEconometrix]

Oh not at all...I've heard so much about it, I may have to look into it.

[@sasweet8183]

i want to use the bootstrap ardl of cointegration.kindly guide me what will be the steps for running?

[@CrunchEconometrix]

Hi Sa, I have no idea on how to do this. I'll have to read it up.

[@sasweet8183]

thanks for ur response

[@mrs.smaheswaranathan7698]

When I run the ARDL modi and find the lower value and upper value i received (estat ectest ) "There must be at least twice as many observations than coefficients". Why this one? How Can I solve this problem?

[@CrunchEconometrix]

Hi Mrs. Maheswaranathan, it implies you have few observation. So, increase the time span of your data. May I know from where (location) you are reaching me?

[@John-zf4or]

Thank you very much for the lesson. Can we run a ARDL model with Python, because I cannot find any library or method support ARDL model yet (I already build ARDL model with R-Studio but I want to build on Python with my graduated thesis). It will be great if you can make a video with Python or show me the way to do it. Many thanks ^_^

[@CrunchEconometrix]

Thanks for the positive feedback, Vinh...but I have no idea about Python. You may want to check other online resources. Regards.

[@Fm11111]

hi thank you for your videos, they are really helpful! Can we still perform bounds test ARDL when we only have one dependent and one independent variable? thank you!

[@CrunchEconometrix]

Yes Fonny, you can....and thanks for the kind words on my videos. Humbly appreciated! Please tell others too...thanks!

[@zoyashah7826]

Mam I am finding it difficult to interpret the result of short run dynamics under ARDL model. I have chosen 3 lags for each of the variable and the estimation shows 3 different coefficients of each variables. So my question is that which coefficient to include while preparing table of short run dynamics for my research paper?? Thanks in advance mam

[@CrunchEconometrix]

Hi Zoya, you can either watch my other ARDL videos OR get any ARDL-based article and adapt their interpretations.

[@apica1234]

Professor Hope that you are doing well. Please guide me. This is Apica from India. PhD scholar. I ran the entire ARDL model but ONE equation(out of three) could not pass the residual diagnostic tests. Please guide me , what could be the problem. Thank you and Regards

[@CrunchEconometrix]

Please note that you don't have to estimate all equations by using each variable as a dependent variable. Not necessary. I only used that approach to explain a vector case. In the event that a model fails a residual test, estimate using higher-order lags.

[@apica1234]

@@CrunchEconometrix Thank you , Professor.

[@pawankumarsingh2114]

great

[@CrunchEconometrix]

Thanks Pawan!

[@nikmatsyukur167]

assalamualaikum Prof, Good morning. If we only want to examine the effect of, for example imp and rexch to mva, do we need to do the ardl/ecm for other 2 equations? Thank you in advance.

[@CrunchEconometrix]

Hi Nikmat, not at all. Simply estimate the desired model. I only showed that you can also create a vector of equations in ARDL models.

[@nikmatsyukur167]

@@CrunchEconometrix ouhh. I see, thank you very much. I'm watch this video as prerequisite for other videos. Very interesting and informative channel. Thank you again

[@tomaseduardochungaseminari5605]

One question, if the variables are cointegrated in the model, I no longer use differences to transform them into stationary if they are not, because being cointegrated they are already stationary with each other.

[@CrunchEconometrix]

What are you talking about, please?

[@tomaseduardochungaseminari5605]

@@CrunchEconometrix I'm sorry, I didn't know how to explain, what can I do if when I submit my variables to stationarity tests I have two of them of second order and the other are I0 and I1, which model I can run. Second, the Granger causality test was performed with the variables at their level or transformed into differences if they are not stationary in level.

[@disnaniharepola755]

Hi, I have 16 independent variables, I want to run the ARDL model, can I do it in for all variables at once , Please advice me

[@CrunchEconometrix]

Hi Disna, with 16 expl vars, your model will definitely collapse. Reduce to 4 or at the most 5. Always remember that parsimonious models give better predictions than over-parameterised models. So, reduce them to something very minimal. I often don't go beyond 4 variables.

[@chelseajacob7272]

Good evening Dr, Please when you have more one ECM in ARDL estimation which of the ECM represents the speed of adjustment coefficient. That if you have ECM(-1), ECM(-2)........ ECM(-n). Thank you.

[@CrunchEconometrix]

Hi Chelsea, you can have just ONE ECT in a model. Kindly watch my ARDL videos for better understanding. Kind regards.

[@oyedemijeremiah2623]

If one of my variable is integrate of order 2 which model will I use

[@CrunchEconometrix]

Toda-Yamamoto technique.

[@swarnachari2703]

Ma'am few variables are stationary at ,2nd difference ...I cannot run ardl in that case?

[@CrunchEconometrix]

Hi Swama, no you cannot. Please may I know from where (location) you are reaching me?

[@swarnachari2703]

India .. Bangalore maam

[@danielmutuku111]

CrunchEconometrix so if that the case what do i do from there? For my case my dependent variable is of order 2 while the independent variables are of order one. Daniel from kenya

[@jeteshaina6220]

Hello Thank you for all,I use ardl and ravallion model transformation by timmer in my school work ,so I want to calcule index of market connection (ICM)but I don't know how can I do that what commad ou regration command a can use in stata Please a need help

[@CrunchEconometrix]

Hi Jete, unfortunately I have no idea what you are referring to. You may need to check other online resources for more information. Please may I know from where (location) you are reaching me?

[@jeteshaina6220]

@@CrunchEconometrix all right ,I came from Mozambique thank you

[@CrunchEconometrix]

@@jeteshaina6220 Awesome! I'll appreciate it if you can share the link to my KZbin Channel with your students and academic community in Mozambique 🇲🇿. Thanks 😊

[@tinotendamakova5563]

can i do the bound test if i have 4 betas. I have 4 independent variables and i also have to include the independent variables

[@CrunchEconometrix]

Hi Tino, your query is a bit confusing....."I have 4 independent variables and i also have to include the independent variables". Please recast.

[@tinotendamakova5563]

I am estimating the relationship between fdi and economic growth. My independent variables are fdi, human, capital, institutions and consumption. I want to run an ardl. But I see that for cointegration tests you use b1=b2=b3=0. Will it be ok if I have b1=b2=b3=b4=b5=0 for my cointegration tests

[@CrunchEconometrix]

@@tinotendamakova5563 Yes.

[@tinotendamakova5563]

Thank you

[@omduttdixit3433]

Can u suggest some reading materials on ardl modal and it's interpretation ?

[@CrunchEconometrix]

Hi Dixit, search my paper online Adeleye et al (2018) Financial reforms and credit growth in Nigeria.

[@theamaxingnature24_7]

Hi Professor! I'm a beginner at ARDL model. I find this video of you very useful. Thank you! I've E-views 8 installed in my PC. Can you please tell me, if I can run ARDL with this version? wishing you BEST!

[@CrunchEconometrix]

Hi girl, thanks for watching my video and for the kind words. Humbly taken. Yeah, you should be able to run ARDL with EViews8. Try it out and see...

[@theamaxingnature24_7]

Oh! thanks a bunch. I was too worried about this. I hope to be more benefited from your knowledge in near future too.

[@CrunchEconometrix]

@@theamaxingnature24_7 That'll be gr8, and I'll keep doing well-informed videos for the global academic community. Please share my videos with your academic community and colleagues too...gracias! 💕 😊

[@nikmatsyukur167]

Helo Dr. when we want to check for the cointegration, we need to do the bound test first, then decide if is it cointegrated or not. if not, then we specify ARDL model. So my question is, as I watched your video about the bound test, you first regress the series using ARDL model before check for the long run relationship, so, what is the difference between ARDL that we run, before do the bound test and ARDL after we check the cointegrated, which is for short-run relationship? Thank you in advance.

[@CrunchEconometrix]

Nikmat, it's the same ARDL algorithm. That's the way the estimation procedure is configured. The 2nd ARDL (restricted model) generates the long-run estimates once cointegration is confirmed.

[@nikmatsyukur167]

@@CrunchEconometrix So, after there is cointegration, its mean that there is long run relationship and we need to estimate long run model (ECM). So, for estimate ECM, what I understand, we just combine the ARDL regression's result + the result for bound test which is ECT. Is it like that?

[@CrunchEconometrix]

@@nikmatsyukur167 Watch the ARDL-ECM videos, and you'll have less questions.

[@nikmatsyukur167]

@@CrunchEconometrix thank you for everything Dr.

[@Rose-Inotila]

What if one of the variables is integrated of order 2, what method do you use to specify?

[@CrunchEconometrix]

Hi Rose, you use the Toda Yamamoto procedure. But since I don't know how to do that, once I have an I(2) series, I drop it and replace with a close proxy.

[@jamyangnurboo5448]

Mam i have a panel data of N=5 ,t= 36. Autocorrelation is there in my data, all my variables are I (0) or I (1). Should i go for ARDL?

[@CrunchEconometrix]

Use panel ARDL. Watch my videos on this procedure. They'll show you how to go about it.

[@mishalkhaled8327]

if my variables are stationary at first difference and second difference also can I apply ARDL model?

[@CrunchEconometrix]

No you can't. Use the Toda-Yamamoto procedure. You can do an online search to know more about it. Thanks.

[@mishalkhaled8327]

Thanks a lot. You mean it should be stationary at the level or first difference, but not stationary all of 3 differences. My case they are not stationary at the level they became stationary and first level and also they are stationary at the second level. Do you a YouTupe about Toda-Yamamoto. @@CrunchEconometrix

[@CrunchEconometrix]

@@mishalkhaled8327 You can use I(0) and I(1) series but not I(2)...no videos yet on the T-Y procedure.

[@mishalkhaled8327]

waw thanks a lot for your fast response I really appreciate your reply. @@CrunchEconometrix

[@CrunchEconometrix]

@@mishalkhaled8327 No worries, dear...keep watching, keep sharing. Tell the world about my YT Channel. That is the only favour I'm asking. Thanks!

[@amanpreetkaur1830]

Respected madam please please guide whether ardl is any useful if both my dependent and independent variables are statuonary at I(0) and what does it indicate? Thankyou for making such wonderful videos!!

[@CrunchEconometrix]

Hi Amanpreet, if all the variables are I(0) it implies no short-run fluctuations in the model so estimate a long-run model using OLS...and the compliment about my videos is humbly taken! Thanks. May I know from where (location) you are reaching me?

[@arrienevillanueva7551]

Hello. Excellent videos. Ive been taking notes based on your videos for a year now. I am currently doing my thesis. Please enlighten me. I have 4 variables, 1 dependent variable & 3 independent variables. My variables are integrated of order 1. Also, based on the johansen cointegration test, there is only 1 cointegrating relationship between my series. I only estimated 1 ARdL model. What should be my next step after that? Thank you in advance.

[@CrunchEconometrix]

Arriene, Johansen Cointegration = VAR technique and Bounds cointegration = ARDL. You decide which way you want to go and watch my videos on the selected procedure.

[@solomonamare8626]

what a channel!!? so amazing and explicitly informative. I am new for the ARDL model, but now I capture so many things while I follow your channel. Go ahead! When constructing the ARDL model, are we supposed to use the first difference of the dependent variable or its level form??

[@CrunchEconometrix]

Hi Solomon, thanks for the positive feedback. Deeply appreciated. Use the level form if estimation is via the ARDL algorithm and the 1st difference if by OLS (watch my video on it). In the simplest form ARDL accommodates non-stationary series (see Pesaran & Shin, 1994, 1999) but we test for stationarity to ascertain that the series not not I(2). May I know from where (location) you are reaching me?

[@solomonamare8626]

@@CrunchEconometrix Thanks so much for your quick response! I am an Ethiopian, studying MSc Economics in Istanbul, Turkey. I want to have a strong connection with you to learn more. I have seen almost all of your videos focusing on ARDL models. I have some questions about the model. Can I have your email, please? If yes, I can state all of my questions and will be learn more.

[@CrunchEconometrix]

@@solomonamare8626 Please state your questions here but make them brief. I often pass on too long queries. Thanks for watching and I'll appreciate it if you can share the link to my KZbin Channel with your friends and academic community in Ethiopia 🇪🇹 and Turkey 🇹🇷 for awareness...thanks 😊

[@gakidema3487]

Hello Professor I have some confusion regarding bounds co-integration test. As i understood from your lecture, if our model has four independent variable and one dependent variable, are we suppose to perform the bounds test by taking each variable as dependent variable or is it ok to perform bounds test for one equation. Since i will be looking for relationship between my dependent variable and other independent variables.

[@CrunchEconometrix]

Not at all. I only did that to show that a vector of equations can be created from an ARDL model. Go ahead and estimate your single-equation model which ARDL is designed for.

[@md.mahfujurrahman864]

baby..can you please provide on basic and spatial econometrics

[@CrunchEconometrix]

mahfujur rahman I don't know what that is....but if I find the time to understand the procedure, I'll do a video on it.

[@md.mahfujurrahman864]

..I am eagerly waiting for you fb mail [e-mail]

[@MultiMRVIP]

Peace and mercy of God The model talks about the effects of technology and communications on economic growth embodied in GDP The duration is not long 11 years Can I estimate this model with this number of observations note independent variables 5 but that has only three significant effect after estimating the model and you do the usual tests for the health of the model and everything was ok I got quadratic data for some variables and converted the other quadrant by the Eviwes program You made an excellent estimate and result but did not perform the sleep tests on the original data from stabilized at the first level The question here is whether the silence should be chosen for the variables of the study note that the sample became 44 views Or, you have adopted the first form or used the model ardl thank you very much PhD student Said Saleh from Palestinian

[@CrunchEconometrix]

Hello Saied, kindly simplify your query for better understanding. Thanks.

[@MultiMRVIP]

CrunchEconometrix The appropriate model for sample 11 views

[@CrunchEconometrix]

"sample 11 views" what does that mean?

[@seanh19954]

If my variables when logged transformed are non- stationary, but 1st differences are all stationary, for the ardl model in stata, do I use the 1st differences or the original log-transformed variables? And could you please check your email! I'd like to get your paypal to say thank you for your help

[@CrunchEconometrix]

Glad to hear that you've sorted it out. Well done.

[@CrunchEconometrix]

I appreciate your sincerity and gratitude. The only way to pay me is to share my KZbin Channel link with your students, friends and academic community on social media for awareness. They'll learn some useful tips and skills too. My goal is to teach the world how econometrics can be easy to understand. Help me to accomplish this goal, will you?

[@seanh19954]

@@CrunchEconometrix if my ardl model is Heteroscedasticity, how do i get around this?

[@CrunchEconometrix]

@@seanh19954 Oftentimes, log transformation works or changing control variables. Some people may suggest GLS (you may need to read this up).

[@lusungunkhonjera7281]

看不懂