EViews: Unit Root Test, Cointegration Test and ARDL-ECM (Estimation and Interpretation)

Step by step process on how to estimate unit root test, bounds cointegration test and ARDL-ECM using EViews

[@saakamahamadu5440]

Great job done, I used your lectures to analyse my thesis. Keep it up!!

[@habibamohammedyimam7275]

Thanks a lot, i used your lecture to analyze data for my term paper

[@ajaegbuikenna6613]

Thanks sir. Uv actually saved me from spending more tons of cash just for the estimation and interpretation part of my ongoing MSc and future PhD research thesis.

[@favourokwuchukwu-uba8674]

what if the trend is statistically significant but the constant i.e the intercept is not? what happens then?

[@adamumuhdbello5058]

Nagode 🙏 the lecture is very helpful, God bless you sir.

[@nalabhembe3085]

Straight to the point indeed, thank you sir.

[@kfohocran]

great presentation. i have a question, any guidelines as to how to use the Bootstrap ARDL model

[@nalabhembe3085]

Strait to the point. Perfect thank you

[@adeyemoolusola3779]

Weldon Dr. your lectures on Eviews has widen my knowledge. However, I want to ask how I can identify Error Correction Form in Eview 13 and constant and trend when performing ARDL bound test. Thanks sir!

[@natashagbologah6345]

Thank you for the video, please is the Conditional error correction model in the long run Same as the ECM in the Short run?

[@THYWILLEje.I.]

Thank you very much, Sir

[@olaniyanezekiel9395]

This is really helpful. Thanks so much. But when presenting my unit root test table, which value do I include under my level and 1st difference out of 1,5 and 10 percent?

[@sherzadshahab9628]

Well done, great job, thanks a lot.

[@faithreigns4972]

Goodafternoon, please can you help with video on how to do PCA using eview 10. Thank you

[@aliyuyusuf8013]

Hi Dr., Can you show how to represent the tables on the main work because it seems it is not the complete table one will take to his work. Thank you. Aliyu Yusuf

[@faithreigns4972]

Great job. Very helpful. Please what eview model do you use here

[@obezipacademy]

EViews 10

[@angelicaantoine7688]

When interpreting the short run coefficient which lag must I take.or should it be a summary of the different lag

[@zoyashah7826]

Same ques I also want to ask

[@tolulopeawere7525]

So helpful, thanks

[@sherzadshahab9628]

Excellent job

[@otekanonso7059]

after conducting the ADF test, what else is done with the stationary data seeing that you didnt use it for any of the ARDL tests

[@lemondedesafa687]

Thank you. But please what can we do if the bounds test is inconclusive ??

[@user-ks9tx8bu6w]

great...,keep it up!!!

[@amalmahmoud6198]

great job can i know please what the upper table in the long run bound test stands for that include the variables in lag and difference

[@benchoukwahiba3017]

thank u very much sir , i have a question how i can check for short-run and long run causality from the ECM-ARDL model. can i investigate the short-run causality by the significance of the Wald tests of the differenced explanatory variables!!!!!

[@adeyinkairewole19]

good day, what if my variable is greater than 0.05 @C and trend, what does it mean and what should i do? thank you

[@emmanuelsenior1191]

Please the short run result includes lag result for some of the variables whilst others have ore than one lag reuslts, which one should i explain in my work

[@easytolearn28]

Great job .sir can you please tell us what can we do if our dependent variable is stationary in level? Whether we apply simple ardl or augmented ardl???

[@obezipacademy]

when dealing with time series data where the dependent variable is stationary at level (I(0)) and the independent variables are stationary at first difference (I(1)), the Autoregressive Distributed Lag (ARDL) approach is often considered suitable. The ARDL model can handle a mix of I(0) and I(1) variables, which is one of its main advantages, However, whether to use a simple ARDL model or an augmented ARDL model depends on the specific characteristics of your data, such as the presence of autocorrelation or other issues in the residuals.

[@easytolearn28]

@@obezipacademy thanks for your reply.in my researchwork i applied ardl and all other diagnostic test results are also perfect.but issue is arise why you apply ardl when dependent is stationary on level, even it is mix of both 1(0) or 1(1).can you please guide me with e references .thanks again

[@christiankofiduah8832]

what is the difference between Conditional Error correction regression and error correction model. Which one is the short run result to interpret.

[@idakwojiblessing4475]

Well done sir, this is quite helpful but i have a challenge, i am trying hard to check my bound test. I followed every step you went but after the ADF, i tried to check if constant and @trend are significant to know whether to add it to ADF, it showed syntax error. I kept trying over and over again but it seems not improving, what do i do pls?

[@obezipacademy]

Carefully cross check the spelling or labeling of the variables... specifications of the equation etc and look at mine again carefully

[@obiaifeanyichukwu]

Obia favour thank you sir 💯💯

[@hezb1996]

Hi Sir, to use ARDL isn’t the dependent variable meant to be integrated by order I(1) ?

[@tomsingh2399]

Thank you for helpful video. One question though, what it mean if my CointEq(-1)* is negative but more than -1? For example -2.394868

[@obezipacademy]

It shows overconvergence

[@peterkehindemogaji3744]

You have done well.

[@victoraninwagu1972]

Great explanation, Dr. Please what should I do when my Bounds F-statistic value is greater than the lower and less than the Upper value at 5% significant? Can I explain that it exhibit a long-run relationship?!

[@obezipacademy]

if the F-statistic falls between the lower and upper bound, the test is inconclusive. In this case, we cannot make a definitive statement about whether or not there is a cointegration relationship among the variables. This is often seen as a limitation of the bounds test. Further testing or additional information would be necessary to make a definitive conclusion. This could involve, for instance, adding more lags to the model and retest. If it persists, then you the researcher is at liberty to go either way, which is to accept or reject that there's cointegration

[@victoraninwagu1972]

@@obezipacademy thanks

[@graceheavens584]

when regressing the variables on its constant and trend and you find out that only trend is statistically significant, what do you do? Because there is no option to choose only trend when running the unit root test. The data for this regression is for 8years. I dont know if it has anything to do with anything

[@emmanuelsenior1191]

Again, please do you have a video on Zivot Andrews breakpoint test

[@emmanuelokeme6272]

Is there a difference between ARDL and ECM form of data analysis?

[@farukonibudo7844]

Hello....if the bounds test fall between the lower and upper bound..whats the way forward

[@tosin_davidson]

Hello Dr.,can I log the variables and carry on with this methodology?

[@DanielSanchez-jh4kd]

Hello Dr. I hope you are very well, your video is very good. I have some questions if you can help me, I'm also working on my final degree project. 1-The sample that I have to estimate the ARDL model has 29 observations and I want to use 4 variables to explain. And let's say, is this viable at first sight? 2- When I estimate the ARDL model, there is an option that says "cointegration graph", is this graph the result of the entire model or only of the long-term variables? because I'm more interested in estimating the long-run relationships but I need to use the graphical form. Thank you very much in advance!

[@victoriaolutofunmi7420]

Thanks alot sir, this video is very simple to understand. My questions are: 1. In performing unit root, there is no single option for a case when only trend is significant, then what should I do? 2. I eventually selected option of none in 1above but when performing Bound Test, I couldn't get a result rather it says "singular matrix" what causes this?

[@victoriaolutofunmi7420]

Also some times it says "log of non positive number".

[@obezipacademy]

For question 1, use trend and intercept option. For questions 2, reduce ur lag length

[@victoriaolutofunmi7420]

Thanks, I did as advised and I was able to obtain my result. Thank you once again

[@alfaprofwengeracademy5099]

My question is that '" my CointEq(-1) value is -1.240196". How will I interpret the speed of adjustment

[@obezipacademy]

An error correction term of -1.24 would suggest that the model is overshooting its equilibrium value, and it corrects significantly and rapidly, since the magnitude of the error correction term is fairly large. It also mean that if the system deviates from the equilibrium relationship between the independent and dependent variable, it will adjust by approximately 124%.Keep in mind that the sign of the coefficient is negative, which implies a stable ECM. This is because when the system is above the equilibrium (the error is positive), the negative sign ensures the adjustments are made downwards. Similarly, when the system is below equilibrium (the error is negative), the adjustments are made upwards. in the next time period.

[@victoraninwagu1972]

Hello Sir. Please, is it advisable to go on with ARDL when the F-bound test is less than the lower and upper bound?

[@obezipacademy]

Simply estimate the first difference ARDL. Here's the video: kzbin.info/www/bejne/eInReaqcr7WJjqM

[@mapaderunemmanuel8639]

Very good....i would need you to personally coach me pls

[@androidit5698]

Thanks a lot for these tutorials. Can you please answer my question; my unit root test reveals that 2 of my variables are stationary at the 2nd difference such that I~(2), one is dependent and the other is independent. While the rest of the variables are level or 1st difference stationary. Can I use the ARDL test?

[@obezipacademy]

ARDL does not accommodate I(2) variables

[@jonahjesse4631]

Jonah Jesse Thank you sir

[@blissezekiel6485]

Esther Ezekiel Thank you sir

[@davisumoru6879]

How do i get the data you used?

[@nkosinathimama8081]

Thank you very much Dr

[@alfaprofwengeracademy5099]

Thank you for the good job

[@drypht]

Thank you so much, just try to be slow in your speech

[@ammahboss3465]

excellent

[@KiraZ-yf9bh]

when I try to test for cointegration using Bound test on E-views, it fails: it says "singular matrix" , what wrong with that?! It would be nice if you make a video on this topic! I am greatful to you for else!

[@obezipacademy]

It simply shows that there's multicollinearity among your independent variables. Reduce the lags on the ARDL and rerun

[@matiullahamarkhil3267]

Hi sir I am using Eview 9 but after running the ARDL model for wald test there is no ECT in eview 9 what is the solution?

[@obezipacademy]

For eviews 9, under coefficient diagnostics, select cointegration and long run form to view ECM

[@joyunigwe3257]

❤❤ thanks sir

[@helinasiripi6820]

Please in reporting the results of the estimation, which specific one should I report since the results bring that of say lag 1 to lag 4 estimates of the variables. Also what if the cointEq(-1) is greater than 1 however it's negative and statistically significant, what should I do? Thank you

[@obezipacademy]

When reporting, you're actually expected to report the complete result with all the lags, even though you might not need to interpret all. If the ect is greater than 1, it shows that there is over convergence in the model. You can actually resolve that by reducing the lags and rerun. If it's still same way, kindly proceed with ur reports, but capture it as over convergence

[@helinasiripi6820]

@@obezipacademy thank you

[@mydress116]

very knowledgeable video, i have one Q, the paper am following has a series stationary at 1st difference, but when i estimate same series it is stationary at 2nd difference. Now what should i do, Regards,

[@richapatel8892]

Perform Toda- Yamamoto test

[@UmeshRR]

Why didn't you perform diagnostic checking?

[@Waleedkhancooldevil93]

Thank u sir. What to do if the trend is significant and constant is insignificant in the ardl estimation?

[@user-qi7dp2ee3p]

trend significant means ur data is a unit root, which means not stationary. further, means. variance and covariance are not constant. while, when the trend is not significant means ur data is not a unit root, which means stationary at level.

[@faithayuba729]

Faith Ayuba Thank you sir

[@yusauaudu9044]

Please, what do I do if there's long run equilibrium between the dependent and independently variables? Thank you!

[@obezipacademy]

If there is long run relationship, estimate ECM

[@prince-uba]

If the order of integration of variables where X is I(o), Y is I(1) & Z is I(2), what is the Appropriate Model for estimating the relationship among the variables.

[@obezipacademy]

No single equation technique for now can handle I(2) variables unless multivariate technique such as Toda and Yamamoto non-granger causality model. If you are using a single equation model and u come across an I(2), variable, kindly find an alternative variable for it

[@phd.eriolamariuscharlotade5634]

@Obezip Universal Statisticals (OBUS), if it doest past the bound test, what should we do?

[@obezipacademy]

Estimate first differenced ARDL

[@chikajoan-qi8tl]

Sir can you please explain how the maximum lags are picked when running the unit root test and ARDL

[@obezipacademy]

You can start with the default of 4 given. There is no literature for a particular number of lags to be used

[@emmanuelblessingoshua1768]

Thanks alot 💕💕

[@alejandroiribas6375]

Thanks very much for your video!! I would like to know where does the term CointEq(-1) come from. Is it the ols lagged estimated redidual series from the long run equation? Thanks in advance

[@obezipacademy]

OLS lagged estimated residual series

[@alejandroiribas6375]

@@obezipacademy Thanks very much for your quick response. I still have some doubts regarding the estimation of the ECT. I've seen that some researchers use ols lagged residuals form UECM equation (Tugcu 2013) while others use residuals from levels equation (as in Engle & Granger approach). Do you know any reference that might shed light to this question?

[@obezipacademy]

@@alejandroiribas6375 If you're estimating Engle and Granger ECM, your residual for estimating ECM, comes from OLS levels equation (without lag), but if you are estimating ARDL-ECM, your residual for estimating ARDL-ECM comes from one period lagged OLS

[@alejandroiribas6375]

@@obezipacademy Thanks very much!

[@victoraninwagu1972]

Greetings Dr. When running for my ARDL, I found that the constant was greater than 0.05, which I restricted before running the ECM, but my challenge now is that the F statistics of the ECM don't show on the result again. Can I go ahead and interpret without F statistics?

[@obezipacademy]

You can report your result like that. But If you want to add the f-statistics, you can manually estimate it and add to your result using Wald coefficient restrictions

[@victoraninwagu1972]

@@obezipacademy thanks

[@okechukwunnaemeka189]

Good evening Sir Please Sir after doing my unit root test, my variables were stationed at 1st and 2nd difference. So, subjecting my variables to ARDL it's showing Non singularity matrix. Please sir what should I do. I have logged my variables and still the same thing. Please what Should I do.

[@obezipacademy]

Reduce the lags and rerun

[@mercyiliya3841]

Thank you Sir

[@yusauaudu9044]

Hi Dr., please, I am trying to run a unit root test on Nigeria's RGDP between 1985-2019 using the ADF on E-View. The constant and trend are significant, however, the series is not stationary at level, first and second. Furthermore, I log and differentiate the log the series yet the series isn't still stationary. Please, what else can I do? Thank you!

[@obezipacademy]

What method of analysis did you propose to use?

[@yusauaudu9044]

I am using an OLS method for the multiple linear regression but need to test for series stationarity using the ADF test to avoid spurious results.

[@obezipacademy]

OK. Convert the variable to "rates or percentage" and retest. Do the conversion before moving it to eviews. If it doesn't pass unit test at levels or first difference this time, then you may have to find an alternative variable

[@yusauaudu9044]

Okay! Thank you, sir.

[@danielmogashoa1964]

Thanks

[@virajkumar4984]

Hello sir, I have 6 independent variables. But only 4 independent variables are appearing in my Short run ARDL result (Error correction results)..what should i do? Please help

[@obezipacademy]

Increase the lag length and rerun it

[@virajkumar4984]

@@obezipacademy Thank you so much sir

[@RN-jr2qe]

@@obezipacademy the variabel that didn't appear in ECM because the lag in ARDL was 0. If i increase the lag, the new ARDL model became instable. I ever read an ARDL journal, even the lag was 0 the variable also appear in short-term ECM model. Can we estimate the coefficient of that variable with eviews?

[@alishagg516]

When doing bound test, what if the f statistic is lower than upper bound at 2.5% and 1%? Also, do we need to do anything next if a variable in long run result is not significant? Thank you so much for your help!

[@obezipacademy]

It's inconclusive if the f value falls between upper and lower bounds

[@alishagg516]

​@@obezipacademy oh okay. So can i change anything to have a more better result?

[@obezipacademy]

You don't need to. You either proceed by estimating Long run and ECM or report first differenced ARDL

[@adedokunjosiah6876]

@@obezipacademy please sir . What eview package do you use..? I'm using eview9 and I can't find ECM there

[@isicheiejikeme3523]

Great

[@angelaadomokhai1231]

thank you sir. please i have a question, what can i do as my f statistics falls between 1(0) and 1(1) thus making it inconclusive but the t-test is lower than the lower critical bound value. what can i do pls?

[@obezipacademy]

If you have such kind of situation, you can either proceed with your results or better still simply run first difference ardl.. That's generating only short run parameter estimates

[@angelaadomokhai1231]

Thank you very much

[@angelaadomokhai1231]

@@obezipacademy also is it important for my jacque berra test to be normally distributed and if yes what else can i do to make it normaly distributed other than taking the log of it.

[@fitfirst4468]

What happens if there is no cointegration , where F statistic is lower than upper and lower bounds at 5%

[@obezipacademy]

You run first differenced ARDL

[@fitfirst4468]

@@obezipacademy how do you run first differenced ARDL?!

[@obezipacademy]

@@fitfirst4468 watch this video of mine: kzbin.info/www/bejne/eInReaqcr7WJjqM

[@fitfirst4468]

@@obezipacademy after you get first differenced, can we use first differenced coefficients in the analysis ?

[@mustaphadjaballah6706]

great

[@graceheavens584]

I keep getting an error message "Singular Matrix" when running bound cointegration test... please i need your help sir

[@obezipacademy]

Reduce the number of lags

[@graceheavens584]

@@obezipacademy Okay. Thank you Sir

[@gamalielkpyeinom4427]

Thank u sir

[@kanikachawla3153]

sir how can we generate residual series and run in ols long run specification

[@obezipacademy]

Watch this video on how to generate residual. kzbin.info/www/bejne/jGrZkIuIm7l0gaM

[@harrytetteh9920]

sir thank you

[@harryshenry6258]

Thank you so much for the explainations and interpretations i would give a five star this video But my question is on the ECM regression table on the coefficient column What if you get the error correction term of -7.782080 and multiplied by 100 can it adjust the r'ship of shortrun and longrun as said on the video?

[@obezipacademy]

That means the model is explosive and thus showing that there is overconvergence from short run to the long-run

[@harryshenry6258]

@@obezipacademy so how can i correct such thing?

[@obezipacademy]

Increase your lag length and rerun to see

[@obezipacademy]

Try increasing the lags... If it doesn't adjust, reduce it beyond the previous lags

[@harryshenry6258]

@@obezipacademy when i increase the lags it comes with the message SINGULAR MATRIX

[@user-qi7dp2ee3p]

you are talking very fast, and not understand u what r u saying, please make it understandable and slow down ur voice too.

[@monarch8378]

He is african

[@usmansaleem1253]

PLEASE SPEAK SLOWLY U SPEAK TOO FAST

[@okpihwoblessingeseoghene2299]

Thank you Sir

[@happydauda5699]

Thank you sir

[@preciousfatuase369]

Thank you sir

[@Safehaven_1]

Thank you sir

[@luciaezekiel3281]

Thank you sir

[@jemimabulus2140]

Thank you sir

[@faithoche9664]

thank you sir