(Stata13):Gregory-Hansen Cointegration Test Structural Break

[@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.

[@dr.clivemairura8266]

Good work and excellent explanations

[@CrunchEconometrix]

Glad you liked it, Dr. Mairura! Kindly inform your students and colleagues about my Channel, thanks.

[@reemelabd3994]

Thanks for the videos :) They're really helpful!

[@CrunchEconometrix]

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

[@reemelabd3994]

@@CrunchEconometrix Montreal, Canada

[@CrunchEconometrix]

@@reemelabd3994 Awesome!!! I will appreciate if you can share the link to my KZbin Channel with your students and colleagues in Canada for awareness. They'll learn some useful hints and tips too, thanks!

[@nashiputhalut978]

Great work, Keep up!!!

[@CrunchEconometrix]

Thanks for the encouragement, Nashipu.

[@spinebuster9490]

Thank you very very much for these videos. 💪💪💪

[@CrunchEconometrix]

U're welcome!

[@luisfelipe8109]

Thanks for the video and the explanation. I only have two doubts. 1- I'm trying to test if a group of financial time series are cointegrated or not. The series are not stationary, so I calculated the series of returns. My doubt is if I need to apply the test in the original series, or in the stationary return series? 2- After applying the test, and discovering cointegration reltationships, to create the ECM to anlyze the Impulse Response Functions and Variance Decomposition do I need to add the dummies?

[@CrunchEconometrix]

Hi Luis, (1) use the return series; (2) dummy variables can be included in the ECM.

[@luisfelipe8109]

@@CrunchEconometrixThanks for the answer. Just one last question: when i write the the code, the order that I put the variables modifies the results. For example, if I put ghansen X Y, the results of the test are different from ghansen Y X. So, I want to understand how to interpratate this difference. If I put X first the results show if the series X is cointegrated with the series Y? What is the interpretation?

[@CrunchEconometrix]

Luis, you have to list the dependent variable first. Follow the guidelines shown in the video and adapt the interpretations to yours. Thanks.

[@asangajayawardhana274]

Great work. Very clear from the beginning. Really appreciate this. Just one question regarding my regressors. I have six independent variables in my model and when I was running ghansen test, its mentioned that "more than 4 right hand since variables are not supported". I'm using ARDL-ecm framework,. Can you please help me with this matter.

[@CrunchEconometrix]

Thanks Asanga for the positive feedback. Deeply appreciated! Reduce regressors to 3 and re-estimate.

[@asangajayawardhana274]

@@CrunchEconometrix Hello Thank you for the feedback. I will give it a go.

[@LuzPerez-gz6lh]

Hi. A question regarding the Ecm of the video "step by step ardl" Why should the bounds test be performed uf G. H. already been performed? G. And H. It shows me cointegration but not the bounds test. Thank you for your job

[@CrunchEconometrix]

Hi Luz, your query is unclear and very confusing. Please write it again in an understandable format. Thanks

[@danteremagit9996]

Thank you ma! I want to ask again please; 1. Why did you interact the dummy variable you created after the Gregory-hansen test? 2. Can the dummy variable be added as exogenous variable to the ARDL without the interactions using exog() in Stata? Thank you

[@CrunchEconometrix]

1) interacting dummy variables is to address structural breaks in the model. 2) yes...you can interact depending on what your research objectives are.

[@fuzhufeifei]

Dear Ngozi, Thanks very much for the video again! It is always helpful. I'd like to consult Ngozi for the following questions: 1. How to understand the third point of the structural break: when there is a visible difference between the past and future movements in a series? What does it mean? Does it mean: the slope of the series is going up or going down in a little fast speed? 2. My Gregory -Hansen tests show that 3 models have 3 different breaking points. However, Model 3 with constant and slope reject H0. Model 1 and Model 2 accept H0. How to interpret such results? How to make a judgment in this case? Is there any criteria that I can look for to understand those results? Which model I should look at? Is it model 3 which rejects H0?

[@CrunchEconometrix]

These queries are too many, so I'll let it pass. I explained to some extent in my video but you can do more search online for other issues mentioned. My apologies, thanks.

[@fuzhufeifei]

@@CrunchEconometrix No problem at all. I totally understand. Thanks so much for the reply as usual.

[@harpreetkaur-bo4ko]

Thank you for imparting your knowledge with us on this topic. I have two queries 1. why we need to add interacting dummies (e.g dx1, dx2) along simple dummy 2. In my data there are two time series and there is a structural break in both series and both these variables are I(1) as per ADF Breakpoint unit root test. My query is Can i go with this test instead of Johansen test because my ultimate objective is to run VECM.

[@CrunchEconometrix]

Hi Harpreet, use ARDL and follow my procedure. I'll advise you read the Gregory-Hansen paper for better understanding. Thanks.

[@harpreetkaur-bo4ko]

@@CrunchEconometrix Thanks for prompt reply.

[@basudebkhanal7510]

Dear professor, thank you so much for the video. However, I am facing a num list error (invalid numlist has elements outside of allowed range) in my test result while running ardl (step 5 in your video). May I get any help to solve this problem? (I have 4 regressor in my model)

[@adityarazpokhrel7626]

I had the same problem sir. Now I restricted my variables and it is solved. However, I am getting error on the model with intercept shift with slope.

[@basudebkhanal7510]

@@adityarazpokhrel7626 Now it is solved. I found my error on the placing of lags to run the ardl. That error was not related to number of variables. Thank you for your reply.

[@CrunchEconometrix]

Glad to hear this is resolved, Basudeb🥰🙏

[@CrunchEconometrix]

Glad to hear that you have resolved this, Aditya🥰🙏

[@carlottapenone7820]

Thank you very much for your videos! I have a question about the ECM you mentioned in the intro slide. How do I test for a VECM stability if I allow for the break dummies? thank you very much for the support!

[@CrunchEconometrix]

Carlotta, you are mixing up the terms. VECM relates to VAR while ECM is to ARDL. Which one is your query based on?

[@carlottapenone7820]

@@CrunchEconometrix Thank you very much for the quick reply! I've been using VECM , which now I realise it's different from what you've been analysing in this video. I'm having problems with VECM and dummies.

[@CrunchEconometrix]

I'm not familiar with that approach.

[@simaykzlkaya9078]

Hello, thank you for sharing invaluable information again! I would like to ask a question. I have a model with one of the independent variables contains structural break(according to its tsline). I tried all 3 models that you suggested but none of them gave a statistically significant result. Before to encounter such an analysis that you suggested in this video, I checked for the ADF test and one of the variables is problematic. What would you suggest me to go further? Thank you in advance.

[@CrunchEconometrix]

Hi Simay,you may need to check other resources on how to handle structural breaks in explanatory variables. I've never done such estimations before. Thanks.

[@daphneashba]

Simay Kızılkaya one independent variable may be far away to estimate your model correctly that is why the residuals might have been increased due to misspecification of the model.

[@mayarbakeer2256]

Hello Professor, The video is amazing and is always of great use, I am working on my PhD and I watch them almost every day I have a primal question because I think I am confused, to reject the null hypothesis of no cointegration, should the test statistic be greater than the critical value, or only the absolute of the test statistic should be greater than the critical value Thank you so much in advance

[@CrunchEconometrix]

Hi Mayer, I suggest you watch my COINTEGRATION videos (Johansen and Bounds). Detailed and well-explained. Thanks.

[@angelastanton3588]

Thank you so much!!

[@CrunchEconometrix]

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

[@angelastanton3588]

@@CrunchEconometrix Cambridge in the UK

[@CrunchEconometrix]

@@angelastanton3588 Perfect! I will appreciate if you can spread the word about my videos to and KZbin Channel to your students and academic community in the UK 🇬🇧 for awareness. They'll learn some useful tips and skills too...thanks 😊

[@doudadada9552]

Hi, Thank you very much That was very helpful 🙂

[@CrunchEconometrix]

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

[@doudadada9552]

@@CrunchEconometrix from Paris

[@CrunchEconometrix]

@@doudadada9552 Awesome, Douda!💕 Kindly spread the word about my videos to your students and academic community in Paris! 😊

[@miwi9883]

What do you conclude when the ADF value does not indicate cointegration but the Zt does for a different break date? While also only the Zt break date is statistically significant ?

[@CrunchEconometrix]

You consider only dates with significant statistics.

[@davidmoreno5539]

Hi Excuse me, How should I interpret the output of sr "LD"? Thank you

[@CrunchEconometrix]

Hi David, give it the "lag difference" interpretation.

[@tebohomosikari4067]

hi Prof, I have tried to follow your steps on ghansen, however, when I run the ardl I get the message " ardl GDPC tld GCF z z_tld z_GCF, maxlags(3 3 3 0 0 0) ec "expression too long" what could be my problem here Prof?

[@CrunchEconometrix]

Teboho, not sure. The syntax looks ok. Click on the error code and follow the guide on how to fix the problem.

[@miwi9883]

Does the dummy variable also work in a VAR model with a structural break?

[@CrunchEconometrix]

Hi Wi Wi, I've never used dummy variables in my VAR analysis. You may want to check out other online resources for more details about this.

[@nurehh2271]

Thank you!

[@CrunchEconometrix]

Thanks Nur!!!

[@hakanuslu2773]

Excellent work, do you think if this test can be used in NARDL model? Thank you in advance

[@CrunchEconometrix]

Hi Dr. Uslu, honestly I cannot say precisely if it can....and thanks for the encouraging feedback. Deeply appreciated!

[@AqsaGull-l8w]

your lectures are always informative explaining things in easy bits. Thank you ma'am for all your work. I want to share my concern here expecting a response: I run the code for ghansen but every time I get the error: unrecognized command: ghansen. why is that?

[@CrunchEconometrix]

You need to install the syntax. Type "help ghansen" into the Stata Command Window and follow the prompt to get the syntax.

[@AqsaGull-l8w]

@@CrunchEconometrix thank you for your prompt response.

[@sibylla553]

My Zt of break(regime) is significant. My CUSUM is still fall outside the boundary, please advices me.

[@CrunchEconometrix]

Bylla, identify that break point and create additional dummies then re-estimate.

[@paolomaregatti7212]

hi, i have a doubt, when i use the dummy variables, the long run is significant. However when i do the breusch pagan test, indicates there is heteroskedasticity, is this normal, or the model is bad specified?

[@AlMamun-ko9dt]

Madam, when will I check the structural break in the time series data? Is it before the Unit root test or after the unit root test?

[@CrunchEconometrix]

Before unit root test is most ideal.

[@AlMamun-ko9dt]

@@CrunchEconometrix thank you, madam.

[@chintamanigautam2523]

Thank you madam for you video. In my study private credit is my dependent variable and REM, Trade, FDI and GDP is my independent variables. All are annul series. There is visible structural break in REM (in 2001) and FDI (in 2005) and GDP (in 2011). These has been tested by Chow test, ZA unit root test. While using ADF and PP unit root test all variables are I(1). How should we model if there is break point in independent variable?

[@CrunchEconometrix]

Chinta, to the best of my knowledge breakpoint analysis is concerned about only breaks in the depvar. You may check other online resources for more clarifications. Thanks.

[@thetruth4712]

The tutorial was very usefull but unfortunately we can't read the command as they were blured.

[@shemokoth4497]

Dear prof Thank you for this wonderful tutorial. I however need your help on this pont. I followed your explanation but my data has more than 5 variables and i need to check the structural break for each. It is confusing because i am getting different break years depending on the order of my variables selection. How do i know that the indicated break date is for which variable. eg for the case of your date 2007 is break date for lngini2, then how about other variables?

[@CrunchEconometrix]

Hi Shem, thanks for the encouraging feedback. Deeply appreciated. Only the break point on the DEPENDENT VARIABLE is of paramount interest. Watch the clip again. I also advise you get articles on break point analysis and study what they did.

[@enitanwale-odunaiya7350]

Good Day Ma, While writing my dissertation, I used unit root test with break point and the variables am using are stationary at I(1) with different break points. My questions are; 1. can I proceed with Johansen cointegration Test 2. if no, can you explain the Gregory-Hansen cointegration test using E-Views 10 3. How do I proceed after item 2 above. Enitan

[@CrunchEconometrix]

Hi Enitan, from the G-H paper, JCT is not applicable in the event of a break in the model. Unfortunately, I am only familiar with the procedure in Stata the reason I don't have the EViews version. You may need to check other online resources. Please may I know from where (location) you are reaching me?

[@enitanwale-odunaiya7350]

From Abuja. After the gregory-hansen, how do I proceed

[@CrunchEconometrix]

@@enitanwale-odunaiya7350 If you have a combination of I(0) and I (1} series, you estimate ARDL model.

[@iftekharimran9667]

Respected professor, I found ADF has the lowest value in all three models, but the breakpoints are different as well. In model 1 (1978), in model 2 (1979), in model 3 (1993). So, which year I need to consider as a breakpoint for the dummy variable. Please kindly let me know.

[@CrunchEconometrix]

Use the breakpoint that is plausible with the year of policy change in that country.

[@iftekharimran9667]

@@CrunchEconometrix Thaks a lot, Professor

[@dilpreetdhillon4725]

Hello ma'am, your video was very informative. kindly clear my doubt that in case of presence of structural break in our time series data, if our all variables (one dependent and two independent variables) are I(1) , do we have to use Gregory-Hansen cointegration only or we can use Johansen Cointegration also by introducing dummy variable in the model?

[@CrunchEconometrix]

Dilpreet, your queries are always confusing. G-H cointegration test is applicable when there's a break in the data. I'm sure I mentioned that.

[@adityarazpokhrel7626]

Thank you Madam. I learnt a lot. But what to do when there exists no cointegration but there is a breakpoint in depvar ? Greetings from Nepal.

[@CrunchEconometrix]

Hi Aditya, no cointegration is exactly what is it. Simply proceed to interpret the breakpoints results. Much love from Nigeria 🙏❤️

[@adityarazpokhrel7626]

@@CrunchEconometrix Much Thanks Madam. Madam what if I get a break year 2011 in the level and 2014 in the trend. Which dummy should I be using. And also in the ECM if "z" is not significant and rest (Such as ADJ and LR coefficients are significant) are signifcant then what to do Madam ?

[@CrunchEconometrix]

Please watch my videos on structural break analysis for guidance.

[@suleabubakar6049]

Thanks Ma for this great lecture. Kindly help me with the do file on gregory-hasen cointegration test

[@CrunchEconometrix]

Hi Sule, thanks for your enquiry. Please know that due to abuse and unethical conduct, Stata dofiles used in my videos are no longer free but available on my website upon payment. Here's the link cruncheconometrix.com/view/datashop.php

[@dr.rayhanulislam6609]

Dear Professor, Thanks a lot for your support with many videos. my data gives an evidence of structural break when I take level and regime but not trend under Gregory-Hansen co-integration test. I got a very good result in LR. But model is not stable as per Cusum test, now what is solution. please advise me details with all options.

[@CrunchEconometrix]

Hi Rayhanul, thanks for the positive feedback on my videos. Deeply appreciated. It implies that there's another break in the model. All you have to do is identify the break date, construct dummy variables and re-estimate. May I know from where (location) you are reaching me?

[@dr.rayhanulislam6609]

@@CrunchEconometrix Dear Professor, I am in Beijing at University of International Business and Economics, I am really grateful to you for your excellent feedback. But how I can find another break? I am confused about ADF test, My questions--1) is it essential to check stationarity for all variable including dummy variables? 2) After checking stationarity you are using non-stationay form of variable at bounds test, at regression/cointegration test which form of variables (level/log/first difference) I have to use? 3) I got a structural break in my data series (found at chow test), now I shall go for gregory-hansen test, Is there any necessity to bounds test? In my data series, one variable is stationary in ADF test ( at level), in pp test stationary ( at first difference) and non-stationary in kpss, what will be decision? In your another video named ardl with dummy variables, you have select lags of all variables, why and when this lag selection is necessary? pls pls reply as soon as possible. if I get your phone no I can call you if you don't mind.

[@CrunchEconometrix]

@@dr.rayhanulislam6609 Hi Rayhanul, I have covered most of these queries in my videos. Optimal lags are selected at the onset. Stationarity is not required for dummy variables. Once you have a break in the model, Bounds test is no longer applicable (explained in my video). If the ADF, PP and KPSS give different results, you have the prerogative to use any so long as the outcome is stationary. Read up on Zivot-Andrews unit root test.

[@sabreenkhan3498]

hello can you please tell me if there are multiple structural breaks in panel data then how can we apply ARDL modelling in STATA

[@CrunchEconometrix]

Hi Sabreen, I do not have any information at the moment. You may want to check out other online resources. Thanks.

[@sabreenkhan3498]

@@CrunchEconometrix actually there is a new command xtbreak to know multiple structural breaks in panel and time series that have recently introduced in stata and the command was used to know the impact of covid . I hope you can go through it and make a knowledgeable vedio. Thanks

[@sabreenkhan3498]

@@CrunchEconometrix ❤

[@CrunchEconometrix]

Sabreen, thanks for this. Appreciated 🙏

[@bunyaminozkur5361]

Thanks for sharing such great contents with us. I really appreciate it! I just wanted to ask if the tutorial's dataset shared below can be re-uploaded again. (I tried to access but had an error) Regards

[@CrunchEconometrix]

Due to abuse some datasets are available upon purchase on my website. Here is the link cruncheconometrix.com.ng/shop

[@adityarazpokhrel7626]

Dear Madam, My value of Zt in ardl is positive but significant and ADJ coefficient is also negative and significant. What should I do in this case ?

[@CrunchEconometrix]

Go ahead and interpret your result.

[@aiboudaziz8318]

Hello professor, in this exemple you have checked structural break in the dependent variable. However, if one of the explanatory variable exhibits structural break point do we have to apply the same method when checking the cointegration

[@CrunchEconometrix]

Aiboud, you may need to check other resources for breaks in the expl var. Thanks.

[@daphneashba]

I have a significant structural break in my ardl series,and some ardl coefficients become statistically insignificant, and I could not make them significant. May this occur due to structural break modelled with dummies? Also adjustment coeff. is negative and nearly 1. Is this possible?

[@CrunchEconometrix]

Hi Daphne, several factors could lead to insignificant coefficients including the presence of structural breaks in the model. I don't understand what you mean by "negative adjusted coefficient". Do you imply the ECT?

[@daphneashba]

@@CrunchEconometrix yes sorry ect is now -0,29. My depvar is consumption, indepvar is income. Shouldn't be ect positive?

[@CrunchEconometrix]

No. ECT should lie between 0 and -1 to evidence reversion to long-run equilibrium. It MUST be negative and if otherwise, the model is explosive.

[@adityarazpokhrel7626]

Good afternoon, madam. I was using the same model with your reference. Upon testing the Gregory Hansen procedure, I got error in the model INTERCEPT SHIFT WITH SLOPE. What should I do now ? Can I go for the model with intercept only ? Because the model with intercept shift with trend shows co integration, however, ECM gives insignificant results. But the model with intercept gives significant results with ECM as well. How should I be reasoning for chosing model with intercept shift only ???

[@CrunchEconometrix]

Hi Aditya, I estimate all the models. You can do same and finally decide on the one that gives you the best outcome. Thanks

[@MrLothman]

Excelent explanation! Just a question Can I use a Vector Error Correction Model after gregory Hansen procedure? Or it only can be used in ARDL models? Thanks in advance

[@CrunchEconometrix]

Hi Planck, maybe. Can't say specifically. I've only used it for ARDL models. Thanks.

[@MrLothman]

@@CrunchEconometrix ok one question more... this procedure can be used in all I(1) variables?

[@CrunchEconometrix]

Yes, it can.

[@badiahahmed2085]

Thank you for your great help, Do you have a video about using a dummy variable for panel data? thank you

[@CrunchEconometrix]

U're welcome, Badiah. Please watch my panel data videos. Thanks.

[@kenechukwujohnpaul3661]

A very good morning to you Dr. Pls how can one determine which of the test statistic that is statistically significant and at what level? Am not sure I got it from your video. Thanks.

[@CrunchEconometrix]

Kenechukwu, You'll have to read up on that from any basic econometric textbook to enable the understanding of regression results.

[@belloibrahim7322]

i'm watching ur videos and found it very interesting but the issue is that i;m not good in stata how would i do?

[@CrunchEconometrix]

Hi Bello, thanks for the positive feedback. Deeply appreciated! Stata is mastered with time when you practice CONSTANTLY.

[@rosebanda3134]

Hello, do I use the gregory hansen test to check for structural breaks in VAR model as well?

[@CrunchEconometrix]

Hi Rose, not sure. You may need to check other online resources on that. Thanks.

[@miwi9883]

Did you found an answer to that question?

[@raihansiddika9942]

Hi, Thanks for your video. I have to questions to ask. I would be happy if you reply. 1. How did you select the lags in step 5 (lags(1 1 1 0 0))? 2. If evidence of cointegration is not established, should we run the ARDL model without "ec"? ardl lngini2 lndc sedu z z_lndc z_sedu, lags(1 1 1 0 0)

[@CrunchEconometrix]

Hi Raihan, thanks for the encouraging feedback on my videos. Deeply appreciated! Kindly watch my video on Optimal Lag Selection....and yes, for the 2nd query.

[@raihansiddika9942]

@@CrunchEconometrix Thank you very much for your reply. I know we can use "varsoc" to find the optimal lag for each of the variables. But, In one of your ardl video, you use the following command to generate lag, which later used in the 'ec' model. such as: ardl LGDP LINDT LDIRT, maxlags(1) aic matrix list e(lags) //which then produces lag for variables ardl LGDP LINDT LDIRT, lags(1 0 0) ec btest I am a bit confused. Should I just stick to using ''varsoc" for each of the variable. Thanks.

[@CrunchEconometrix]

Oh I see...either approach is fine. But use the maxlags ( ) approach for Stata to indicate the default BIC lags for each variable.

[@thetruth4712]

Sorry prof disturning. Why does stata comand show doesnt recognised ghansen when i run? Regards.

[@CrunchEconometrix]

Because you have not installed the syntax. Do so with any of these: help ghansen OR findit ghansen OR ssc install ghansen.

[@thetruth4712]

@@CrunchEconometrix thank you very much prof. It is now working. Just want to asked one more q. What can we do if model is still not stable. Is there any way that can automatically corrected the instability on stata? Regards.

[@熱乾麺]

thanks for u video. could i ask u a question? i use quarterly date but it showed "but with gaps". when i did ghansen, it showed that sample may not contain gaps. how do i deal with it? u answer will be great help to me.

[@CrunchEconometrix]

GH won't execute except you change those variable(s) that have missing values.

[@熱乾麺]

@@CrunchEconometrix thanks, i have solved it

[@iftekharimran9667]

Respected professor, thanks for your nice tutorial. I found no cointegration in GHansen structural break test but I found cointegration in ARDL bound test, so in this situation what model I need to use long-run Error Correction Model (ECM) or the Short-run ARDL model.

[@CrunchEconometrix]

Hi Iftektar, refer to my ARDL videos on what to do if there is no cointegration. Thanks.

[@iftekharimran9667]

@@CrunchEconometrix Thanks a lot, professor

[@obaidullah63]

after generating z and z-indep while running ardl there is a problem of collinearity, why is it so?

[@CrunchEconometrix]

Obaid, 'z' is common to both.

[@obaidullah63]

@@CrunchEconometrix thanks . but then how to run ardl having z and z-ind

[@CrunchEconometrix]

Watch my videos on 'ARDL and Dummy Variables".

[@obaidullah63]

thank you so much

[@rainardmutuku8476]

Hi Obaid, did you get the solution to this since I am facing the same problem? Please advise

[@enongenebetrand1119]

Thanks prof.it is wrong to run ghansen at levels or using log transformation.secondly can the test be apply to many variables with multiple breaks.

[@CrunchEconometrix]

Hi Mr. Bertrand, you can log-transform the variables. The test is applicable ONLY to the depvar having a break or multiple breaks.

[@enongenebetrand1119]

Thanks so much Prof for quick response.one more question,I have 3 indepvar with multiple breaks.which test can I apply.

[@mustaphadjaballah6706]

Hello could we estimate the model of cointegration at breakpoint by FMOLS instead of ARDL method?

[@CrunchEconometrix]

Not sure. You may need to check other online resources.

[@mustaphadjaballah6706]

@@CrunchEconometrix because FMOLS DOLS ....these are cointegration methods

[@CrunchEconometrix]

Ok. But I've never tried them with break points.

[@anandarajsaha8697]

Kindly upload Video Lecture on Gregory and Hansen Cointegration Test using Eviews. It will be a great help.

[@CrunchEconometrix]

Anandaraj Saha ok, I'll work on it...

[@enongenebetrand1119]

thanks alot Dr for the video.it was helpfull.but in my study with I(0) and I(1) after running G-H test ,all levels have different break but Ho is rejected.i am confused the right date to chose.any idea on how to specify the equations mathematically when using G-H test.thanks

[@CrunchEconometrix]

Hi Enongene, choose the date from one of the 3 models and look up the G-H paper for the model specification.

[@ogundipeayobolawole2971]

thumbs up.... How can I test structural break in a panel data

[@CrunchEconometrix]

Thanks Ayo...no idea for now. Kindly share my videos with your students and colleagues...thanks! :)

[@faa-e-quahiqbal3976]

thanks for your video, i just want to know how u calculate the lag values in gregory hansen model while calculating ECM model. please do reply

[@CrunchEconometrix]

Hi Iqbal, as seen from the video, I did not manually calculate it. It is automatically derived. May I know from where (location) you are reaching me?

[@faa-e-quahiqbal3976]

@@CrunchEconometrix mam I am asking same..I didn't get from where it is calculated...plz help if it is possible for u...I am from India

[@CrunchEconometrix]

@@faa-e-quahiqbal3976 Ok. I still don't understand your query but to generate lag values in Stata is simply l.y or L.y..and as I said, I didn't calculate the ECM, it is automatically generated.

[@spinebuster9490]

So, Prof I wish to ask this. Let us say Indc was originally a dummy variable and the model shows break points from the cusum test. My question is, will you then create another variable interacting Ind(dummy) and Z dummy?

[@CrunchEconometrix]

I'm not sure if that will be methodologically plausible.

[@spinebuster9490]

I agree. it doesn't sound right. I will just say the model is unstable. Thank you.

[@kenechukwujohnpaul3661]

Good day Dr. I have followed your video as you outlined it. When I run the gregory-hansen test command, I get the following error message. -Number of gaps in sample: 143 Sample may not contain gaps r(198); Pls how do I take care of this?

[@CrunchEconometrix]

No idea what that means.

[@kenechukwujohnpaul3661]

@@CrunchEconometrix I was able to figure it out. Stata is very sensitive with the kind/format of data you feed into it for analysis. For instance, if your date data does not conform with any of its acceptable format, it will not run. Again, if any of your variable is a string data, it won't run. In my case, I reformatted my date data and converted a string variable to a numeric variable (as they were what I discovered was the issue) and I was good to go. I just thought I should share. Thanks anyway for your time.

[@yaminiyadav8064]

@kenechukwu are u using monthly data?

[@CrunchEconometrix]

Thanks for sharing Kenechukwu. My viewers will find this information helpful.

[@masih5925]

Why need to use 3 different specifications? What happen if they are producing contradictory results?

[@CrunchEconometrix]

I have explained in the thread.

[@vigsubbey]

Hi ! What should I do if introducing Dummy Variables has no effect on the stability ? Does this mean there could be more than one structural breaks?

[@CrunchEconometrix]

That depends on the outcome of the GH test.

[@omarahmed8957]

S'il vous plait, comment effectuer ce test sur eviews ? sinon merci pour vos vidéos, elles sont tres utiles pour les étudiants

[@CrunchEconometrix]

Hi Omari, thanks for watching my videos. Kindly write your comments in English because I am a non-French speaker...gracias!

[@omarahmed8957]

@@CrunchEconometrix how we can perform this test in eviews, please. and thanks for your videos, they are very useful

[@CrunchEconometrix]

No idea Omari, that's the reason I don't have the EViews version.

[@meghnakhadwal4767]

Hi I am not able to run the cusum6 command in Stata? could any one help me

[@CrunchEconometrix]

Did you install the syntax?

[@masih5925]

Why need 3 test statistics? What happen if they are contradict with each other? Which one that better use?

[@CrunchEconometrix]

You pick that in which the break date can be linked to an event or policy shift in your data for the period under study.

[@kenechukwunwisienyi6262]

Good evening Dr. I met with this situation while running the ghansen test; Model 1: ADT statistics shows cointegration, zt statistics doesn't. Model 2: ADT statistics shows cointegration, zt statistics doesn't. Model 3: both statistics show cointegration. Again all the statistics in model 1 and model 2 has the same break dates, model 3 has a different break date. Now can I actually say that there is cointegration? If yes, which break date am I using? Thanks

[@CrunchEconometrix]

Read the Gregory-Hansen paper if you have not. That's the underlying study that explains the different test statistics. Use the break date that gives the closest reasoning to your study.

[@mustaphadjaballah6706]

hello if i run a similar exemple but with no cointegration at break point cun i take a dummies variables

[@CrunchEconometrix]

You are not referring to this video. Are you?

[@mustaphadjaballah6706]

@@CrunchEconometrix i am referring to this video of cointegration at breakpoint but in other one case when we accept the null hypotsis What do we do in this case

[@CrunchEconometrix]

Null hypothesis of what?

[@mustaphadjaballah6706]

@@CrunchEconometrix in your exemple you rejcet the null hypothesis from gregry -hansen test and you accept H1(there is acointegration in breakpoint ) but i take on other exemple i find accept H0 (no cointegration in breakpoint ) in my case Can I add dummy variables?

[@CrunchEconometrix]

@@mustaphadjaballah6706 If there's no cointegration and you have a break point perform ARDL with dummy variables for the break.

[@jameswarrenjohnson7251]

Good evening madam. I have a database of 6 variables. I performed the ADF test on the 6 and I find that 3 are I (1), 2 are I (2) and 1 is I (0). I perform the test of Zivot and Andrew (1992) and I note that all my variables are stationary with structural breaks. My question is the following. Do I have to do Gregory and Hansen's cointegration test or simply switch to Ordinary least Square ?

[@CrunchEconometrix]

Hi James, familiarise yourself with the rudiments of VAR and ARDL models. I have videos on these. Watch them to learn the basics and you will be guided on how to proceed. Thanks.

[@jameswarrenjohnson7251]

@@CrunchEconometrix Can the Gregory and Hansen test be applied to a model containing stationary variables with structural breaks?

[@CrunchEconometrix]

@@jameswarrenjohnson7251 But you are watching the clip?

[@jameswarrenjohnson7251]

@@CrunchEconometrix yes I'm watching the clip

[@CrunchEconometrix]

@@jameswarrenjohnson7251 Ok then...follow the procedures with your data.

[@vigsubbey]

Hi, When I try to do step 3, I get the following message: "more than 4 right hand side variables not supported". What should I do about this if I want to test for cointegration?

[@CrunchEconometrix]

Vigneshwar Subramaniam It means you have to reduce your explvars to 4 or less b4 running the regression.

[@poojagomes8854]

What to do when ADF and Zt reaches different conclusion? In my model absolute value of ADF test statistics is lower than the critical value at 5% level whereas the absolute value of Zt test statistics is is higher than the critical value at 5% level

[@CrunchEconometrix]

These are different measures. You stick with one. I will advise you read up the Gregory-Hansen paper for better understanding.

[@samuelbodunrin]

This is not possible in stata 15, any reason why?

[@CrunchEconometrix]

Hi Samuel, it should run. Type: ssc install ghansen OR help ghansen OR findit ghansen in the COMMAND WINDOW and follow-up with the response.

[@samuelbodunrin]

@@CrunchEconometrix Thanks, it now worked. It wasn't install b4. I love your clear voice. Nice work

[@user-oj9rd1hf1q]

How to do this in R?

[@CrunchEconometrix]

Sorry dear I have no idea....may I know from where (location) you are reaching me?

[@mustaphadjaballah7089]

how to run ardl by stata

[@CrunchEconometrix]

ARDL is a user-written program, so you have to use the commands to execute it.

[@azzahassan84]

how to do so with eviews?

[@CrunchEconometrix]

No idea, Sir. Not sure if EViews has the routine.

[@mustaphadjaballah7089]

how to run ardl by stata

[@CrunchEconometrix]

I have several videos on ARDL using Stata.