Thanks for your comment! I am glad to hear that. Happy to know it has helped you. Feel free to share my channel with anyone who you think may be interested. Best Regards and good luck with your thesis! JD
Thanks Youcef for your comment! I am Glad to hear it was useful! Feel free to subscribe to my channel for more videos and also, feel free to check my website where all the courses/videos are free and organized www.jdeconomics.com/courses/ Regards! JD
No problem! I am glad to hear it was helpful! Feel free to subscribe for the coming video - Error correction model. I hope to submit it soon, Regards, JD!
Great and well educating video 👍 Please, what should I do when my Engle and Granger and P. Oularis cointegration P-values are greater than 0.05 level of significance?!
Hello Elena, thank you for your message. I will be posting error correction method and then I can plan ahead and do vecm. Unfortunately I won’t be able to do the VECM video within 2 days as it takes me a lot of time doing the slides, finding the appropriate dataset, testing, filming, editing the video, posting etc. For VECM, you need to first do the Johansen cointegration test, and if there is cointegration you can proceed with the Vecm estimation. I’m sorry I won’t be able to submit that video on time. I wish you good luck! Best Regards, JD
Hello Sabina, thanks for your message and suggestion. I will consider it for future videos. Feel free to subscribe (if you haven’t) so you get notified of the new content. Kind Regards, JD
For unit root test, how should i know whether to use intercept or intercept & trend option? One of my variables shows non-stationary at first difference in intercept but stationary when I choose intercept & trend.
Hi. Including an intercept means that the series is stationary around a non zero mean. Trend and intercept means its stationary around a trend. In that case you still got to detrend the series using first differences or modelling the trend. Regards
Thanks for the positive feedback! And yes. I will do one about it some time. I’m hoping to start teaching python! Free software, and great results. Regards, Jd
Hello Everyone! Many people emailed me to publish a video about cointegration, I hope you find it useful! Thanks a lot for watching! ✅ You can buy for a small amount the EViews Wokfile complete explained step by step + video slides + dataset at : payhip.com/b/x9N7v ✅Watch Part 2: Estimating the Error Correction Model Link: th-cam.com/video/lvyTjcI9po4/w-d-xo.html ✅ Visit my website for more information about cointegration covered in the video: www.jdeconomics.com/cointegration-and-error-correction-model/ ✅ Subscribe to my channel for more videos!: th-cam.com/channels/5P21WGFO4WRUlAiGLcwymg.html Have a great day and wish you good luck on your research and assignments. JD.
Thank you Mr. Juan, Thank you for the video and the great effort you have been doing through your channel. I have an inquiry. I run the cointegration test on various economic pairs of 7 economic variables. What's weird is that I always get a high Prob. value on the Engle and Gragner test (around 0.55). Given that I only have 20 annual observations, could it be the problem? Or what am I doing wrong? how to diagnose the problem? Much appreciated. Thank you
Hi! Yes, it could be the data frequency. Engle and granger only works with two variables at the same time. You need the Johansen test for multiple variables.
We are testing the unbiased forward rate hypothesis, meaning we are testing forward rate is equal to the future spot rate (T + 30) by using ADF and co-integration. Do we need to lag the spot rate by 30 days to test the null hypothesis? Or do we only need to use the regular logs without adding lags?
Hi, first of all thank you very much for the video, it helps a lot! I have a question concerning the "regression residual test table": you have mentioned that you used 2 variables so the critical values are in the first 3 rows. But don't we have 3 variables as it says "Number of variables - N+1", so the critical values in this case would be 3.47 and 3.78? Maybe i am wrong but thought it would be nice to hear ur opinion on that. Kind regards :)
Hi , first all thanks you for the video. I have a question,how can residual itself be stationary, and how does it relate with other variables so that to make be co-integrated. Please 🙏 help.
Hello, may I aske you? When you test the order of integrability of the variable lm you does not select the specification trend and constant in ADF test, although the graph shows that there is a trend. Why is this so?
Hey, just for illustration purposes. Feel free to include any specifications you may consider. (The videos would be 3 hours long if I went through every little detail). I hope that was helpful! Best wishes. JD
Hello, I have conducted the same analysis on daily data for 4 years. Data included daily % price changes of two series. When i conduct the analysis, everything is fine. So I generate residual series, I conduct ADF test and prob. of adf test is 0.000 but t-statistic value is -34.65958. What should I understand from such high value ? Thank you...
Hi Juan! I'd like to ask for your help. See, I'm trying to see if theres a relation between industrialization and inflation by using Industrial Production Index and Consumer Price Index. Both series look similar, at least for my country (Mexico). Of course I used the cointegration method, first by relying only on ADF test which accepts the null hypothesis (using the cointegration tables), and then by doing the cointegration test you showed in this video. The Engle - Granger test also points that there's no cointegration by giving me statistics of -1.17 and -5,7, and P values of 0.86 and 0.67. But just to be sure I tried with the Phillips - Ouliaris test and it gives me statistics of -3.88 and -30.01, and P values of 0.01 and 0.005, therefore, rejecting the null hyphotesis. So, what should I pick as 'real'? Idk a lot about this tests but I bet the Phillips - Ouliaris test is more precisely and... well, idk. What should we do in this situation? Have a nice week Juan!
@@alaeberradi9384 Hey pal, how is it going? Well, even after a year I still have no way to answer my question with a 100% of confidence, maybe because after all, this is statistics and it's all related to trends and probabilities, not 'yes/no' answers. I don't really remember what he said in the video about the ADF (I'm too lazy to watch it again lmao), but maybe I can help you. I remember having that question for the industrialization-inflation relation for my bachelor's degree thesis, which was divided into four chapters, being the third about econometrics. While it's true that the Engle-Granger and Phillips-Ouliaris tests both help us to see if there's a long run relationship, once again I tell you that I don't think there's such thing as an absolute answer for these things. Cointegration is not meant to show us if our assumptions are real or not, it is a tool that we can use to help us collect more evidence for a relationship we are assuming. I don't really remember the words, but Damodar N. Gujarati, in 'Basic econometric's, said something precisely in the cointegration tests like 'wheter we like it or not, causality can only be answered with philosophy'. In my case, in my thesis I had already collected information in the first and second chapter about the causality of industrialization-inflation from a lot of sources, mainly purely theoretical sources. So when we are trying cointegration tests, if we get results such as ours where there one way shows what we want to see and the other shows the opposite, I'm saying that you should just pick what is in line with your theory, and if you are in the need to show both results, go ahead and edit the second one hahaha. It's actually more usual than you think, and usually economists using econometrics, when dealing with these kind of problems, simply say something like 'while we can see there's opposite econometric evidence, theory tells us that blah blah blah...'. So there you are my friend. Although this might not be the answer you were looking for, I'm sure you can actually use it. And I encourage you to search that chapter about causality in Basic Econometrics by Gujarati to see that I'm not lying and this is all about, again, trends. Good luck!
@@TheViportsPYN This is actually helpful in a sense, in my case im trying to find a good basis for a pairs trading strategy im trying to implement. o I'm just tryingto make sure my theory works before doing further efforts
@@alaeberradi9384 Ohhh! I see. So you need something more about a personal use than a professional one. Well, what kind of strategy are you using and in what field? Like crypto, stocks, forwards…? I know some ideas of portfolio theory, and I don’t think you really need to use cointegration to use them if that’s what you’re dealing with. Besides, portfolio theory tells us that whatever you got into your portfolio, you’re looking for those things to have an almost 0 level of correlation which we can say is a ‘step before’ econometrics. You are looking for almost 0 correlation because you want to diversify. It’s not much related to ‘causation’. Once you have a bunch of, let’s say, stocks that are non correlated, you can go ahead and calculete the sharp ratio an minimum variance ratio to see where you need to put more weigh in. Finally, you simply predict using ARIMA models and see if you got a good portfolio and investing strategy!
Dear Juan, I used two data series govt expenditure and gov revenue (both are I(1)) to see whether they are cointegrated or not, I found they are cointegrated, but when I test cointegration after taking the log of both series, it seems that the series are not cointegrated. What could be the reason that after taking the logarithm, the series are not cointegrated?
That's a good question and never came across that case myself. I believe what happened is that when you applied a log transformation to your series, it got smooth in some spots where the series had some "connection" with the other series you were testing. I am not sure of the procedure for your case. Regards, JD
Hi, Thanks for your message! So as I explained in the video, the variables hold a long run relationship. If the variables didn't, then the regression would be spurious. The model in levels reflects the long run regression. In my next video I will be estimating the error correction model, which will show the short run dynamics and speed of adjustment. I trust the next video will help clarify some doubts if you still have some. Kind Regards, JD.
could I ask why from the model 1, the variables are found to be stationary at I(1). But why we don't need to do first difference on the variable when running OLS to extract the residuals? can anyone answer my question?
Hello sir please what can cause the THRESHOLD technic to disappear from the available technique in e-vews if one want to run a data using the threshold analysis for a number of selected countries in the ecowas region.
Hello Eduardo. As we have seen in the tutorial, before applying any transformations, you need to check whether the variables are cointegrated. If they are not cointegrated, you can use first differences and estimate your model as you normally would. Good luck!
Hi, as I explained in the video, the unit root test is performed on a regression output and not a raw series. That affects the critical values. Therefore, you have to rely on a stats table. Regards, J
The dataset is here: www.jdeconomics.com/eviews-tutorials/cointegration-in-eviews Also, you can buy the slides and file at: jdeconomicstore.com/b/cointegration-eviews Regards
You are the lighthouse to my obscure thesis =)
Thanks for your comment! I am glad to hear that. Happy to know it has helped you. Feel free to share my channel with anyone who you think may be interested. Best Regards and good luck with your thesis!
JD
Excellent.....I have also purchased your online material
Awesome, thank you! I hope you enjoy it! Feel free to check out all my videos! Best, JD
You are insane. that's video amazing. it's so useful for my thesis last part. i m looking forward to tar ms var. THANK YOU VERY MUCH !!!! FROM TURKEY
Great to hear! Feel free to share my channel to others! Good luck on your thesis! JD
Thank you ..it was very well explained
Thanks for your feedback! I am glad to hear it was helpful. Feel free to subscribe for more videos coming!
Kind Regards,
JDEcon.
thank you so much sir for this course.
Thanks Youcef for your comment! I am
Glad to hear it was useful! Feel free to subscribe to my channel for more videos and also, feel free to check my website where all the courses/videos are free and organized www.jdeconomics.com/courses/
Regards!
JD
Thank you so much!
No problem! I am glad to hear it was helpful!
Feel free to subscribe for the coming video - Error correction model. I hope to submit it soon,
Regards,
JD!
@@JDEconomics I look forward to it! please please - completing a project on it.
@@yesandno22 The video has been posted! Cheers
Hello sir...Really, it was well explained...And I also request you to explain about Johansen Cointegration model and ARDL model.
Thanks! Those will come out soon. Just subscribe and stay tuned! Cheers
Great and well educating video 👍
Please, what should I do when my Engle and Granger and P. Oularis cointegration P-values are greater than 0.05 level of significance?!
ALL p-values under the Engel-Granger must be under 0.05 for all variables to be cointegrated? Good vid.
Can you do a video explaining the VECM method, please? I have an exam in 2 days and it will be very helpful.
Hello Elena, thank you for your message. I will be posting error correction method and then I can plan ahead and do vecm. Unfortunately I won’t be able to do the VECM video within 2 days as it takes me a lot of time doing the slides, finding the appropriate dataset, testing, filming, editing the video, posting etc. For VECM, you need to first do the Johansen cointegration test, and if there is cointegration you can proceed with the Vecm estimation. I’m sorry I won’t be able to submit that video on time. I wish you good luck! Best Regards, JD
Thank you very much.
Happy to hear you liked it! Regards, JD
Would you be kind enough to do a video explaining SARIMA methods?
Hello Sabina, thanks for your message and suggestion. I will consider it for future videos. Feel free to subscribe (if you haven’t) so you get notified of the new content. Kind Regards, JD
For unit root test, how should i know whether to use intercept or intercept & trend option? One of my variables shows non-stationary at first difference in intercept but stationary when I choose intercept & trend.
Hi. Including an intercept means that the series is stationary around a non zero mean. Trend and intercept means its stationary around a trend. In that case you still got to detrend the series using first differences or modelling the trend. Regards
Hello Juan, incredible video as always, have you thought about doing a video on ARDL model?
Thanks for the positive feedback! And yes. I will do one about it some time. I’m hoping to start teaching python! Free software, and great results. Regards, Jd
Hello Everyone! Many people emailed me to publish a video about cointegration, I hope you find it useful! Thanks a lot for watching!
✅ You can buy for a small amount the EViews Wokfile complete explained step by step + video slides + dataset at : payhip.com/b/x9N7v
✅Watch Part 2: Estimating the Error Correction Model
Link: th-cam.com/video/lvyTjcI9po4/w-d-xo.html
✅ Visit my website for more information about cointegration covered in the video:
www.jdeconomics.com/cointegration-and-error-correction-model/
✅ Subscribe to my channel for more videos!: th-cam.com/channels/5P21WGFO4WRUlAiGLcwymg.html
Have a great day and wish you good luck on your research and assignments.
JD.
Hi thanks for video. Can you please send me the table
Hi. What table? Cointegration ones? Here: www.economics.utoronto.ca/jfloyd/book/statabs.pdf
Thank you Mr. Juan, Thank you for the video and the great effort you have been doing through your channel. I have an inquiry.
I run the cointegration test on various economic pairs of 7 economic variables. What's weird is that I always get a high Prob. value on the Engle and Gragner test (around 0.55). Given that I only have 20 annual observations, could it be the problem? Or what am I doing wrong? how to diagnose the problem?
Much appreciated. Thank you
Hi! Yes, it could be the data frequency. Engle and granger only works with two variables at the same time. You need the Johansen test for multiple variables.
We are testing the unbiased forward rate hypothesis, meaning we are testing forward rate is equal to the future spot rate (T + 30) by using ADF and co-integration. Do we need to lag the spot rate by 30 days to test the null hypothesis? Or do we only need to use the regular logs without adding lags?
Hi, first of all thank you very much for the video, it helps a lot! I have a question concerning the "regression residual test table": you have mentioned that you used 2 variables so the critical values are in the first 3 rows. But don't we have 3 variables as it says "Number of variables - N+1", so the critical values in this case would be 3.47 and 3.78? Maybe i am wrong but thought it would be nice to hear ur opinion on that. Kind regards :)
I think ure right
Hi , first all thanks you for the video.
I have a question,how can residual itself be stationary, and how does it relate with other variables so that to make be co-integrated.
Please 🙏 help.
Hello, may I aske you? When you test the order of integrability of the variable lm you does not select the specification trend and constant in ADF test, although the graph shows that there is a trend. Why is this so?
Hey, just for illustration purposes. Feel free to include any specifications you may consider. (The videos would be 3 hours long if I went through every little detail). I hope that was helpful! Best wishes. JD
Hello, I have conducted the same analysis on daily data for 4 years. Data included daily % price changes of two series. When i conduct the analysis, everything is fine. So I generate residual series, I conduct ADF test and prob. of adf test is 0.000 but t-statistic value is -34.65958. What should I understand from such high value ? Thank you...
I’d have to see the model, but it’s just confirming it’s strongly rejecting the null hypothesis. That’s all.
can we use three series to conduct this test
VEC Models are for those cases.
Hi Juan! I'd like to ask for your help. See, I'm trying to see if theres a relation between industrialization and inflation by using Industrial Production Index and Consumer Price Index. Both series look similar, at least for my country (Mexico). Of course I used the cointegration method, first by relying only on ADF test which accepts the null hypothesis (using the cointegration tables), and then by doing the cointegration test you showed in this video.
The Engle - Granger test also points that there's no cointegration by giving me statistics of -1.17 and -5,7, and P values of 0.86 and 0.67.
But just to be sure I tried with the Phillips - Ouliaris test and it gives me statistics of -3.88 and -30.01, and P values of 0.01 and 0.005, therefore, rejecting the null hyphotesis.
So, what should I pick as 'real'? Idk a lot about this tests but I bet the Phillips - Ouliaris test is more precisely and... well, idk. What should we do in this situation?
Have a nice week Juan!
Hello mate, too bad no one answered i had the same question. but didnt he say in the video that the ADF test is not valid?
@@alaeberradi9384 Hey pal, how is it going?
Well, even after a year I still have no way to answer my question with a 100% of confidence, maybe because after all, this is statistics and it's all related to trends and probabilities, not 'yes/no' answers.
I don't really remember what he said in the video about the ADF (I'm too lazy to watch it again lmao), but maybe I can help you. I remember having that question for the industrialization-inflation relation for my bachelor's degree thesis, which was divided into four chapters, being the third about econometrics.
While it's true that the Engle-Granger and Phillips-Ouliaris tests both help us to see if there's a long run relationship, once again I tell you that I don't think there's such thing as an absolute answer for these things. Cointegration is not meant to show us if our assumptions are real or not, it is a tool that we can use to help us collect more evidence for a relationship we are assuming.
I don't really remember the words, but Damodar N. Gujarati, in 'Basic econometric's, said something precisely in the cointegration tests like 'wheter we like it or not, causality can only be answered with philosophy'. In my case, in my thesis I had already collected information in the first and second chapter about the causality of industrialization-inflation from a lot of sources, mainly purely theoretical sources. So when we are trying cointegration tests, if we get results such as ours where there one way shows what we want to see and the other shows the opposite, I'm saying that you should just pick what is in line with your theory, and if you are in the need to show both results, go ahead and edit the second one hahaha. It's actually more usual than you think, and usually economists using econometrics, when dealing with these kind of problems, simply say something like 'while we can see there's opposite econometric evidence, theory tells us that blah blah blah...'.
So there you are my friend. Although this might not be the answer you were looking for, I'm sure you can actually use it. And I encourage you to search that chapter about causality in Basic Econometrics by Gujarati to see that I'm not lying and this is all about, again, trends.
Good luck!
@@TheViportsPYN This is actually helpful in a sense, in my case im trying to find a good basis for a pairs trading strategy im trying to implement. o I'm just tryingto make sure my theory works before doing further efforts
@@alaeberradi9384 Ohhh! I see. So you need something more about a personal use than a professional one. Well, what kind of strategy are you using and in what field? Like crypto, stocks, forwards…?
I know some ideas of portfolio theory, and I don’t think you really need to use cointegration to use them if that’s what you’re dealing with. Besides, portfolio theory tells us that whatever you got into your portfolio, you’re looking for those things to have an almost 0 level of correlation which we can say is a ‘step before’ econometrics. You are looking for almost 0 correlation because you want to diversify. It’s not much related to ‘causation’.
Once you have a bunch of, let’s say, stocks that are non correlated, you can go ahead and calculete the sharp ratio an minimum variance ratio to see where you need to put more weigh in. Finally, you simply predict using ARIMA models and see if you got a good portfolio and investing strategy!
Dear Juan, I used two data series govt expenditure and gov revenue (both are I(1)) to see whether they are cointegrated or not, I found they are cointegrated, but when I test cointegration after taking the log of both series, it seems that the series are not cointegrated. What could be the reason that after taking the logarithm, the series are not cointegrated?
That's a good question and never came across that case myself. I believe what happened is that when you applied a log transformation to your series, it got smooth in some spots where the series had some "connection" with the other series you were testing. I am not sure of the procedure for your case. Regards, JD
Good video! but so what are the consequences of integration?
Hi, Thanks for your message! So as I explained in the video, the variables hold a long run relationship. If the variables didn't, then the regression would be spurious. The model in levels reflects the long run regression. In my next video I will be estimating the error correction model, which will show the short run dynamics and speed of adjustment. I trust the next video will help clarify some doubts if you still have some.
Kind Regards,
JD.
could I ask why from the model 1, the variables are found to be stationary at I(1). But why we don't need to do first difference on the variable when running OLS to extract the residuals? can anyone answer my question?
You use the residual test to check for long run relationship between the variables. The variables have to be in levels. Regards
Hello sir please what can cause the THRESHOLD technic to disappear from the available technique in e-vews if one want to run a data using the threshold analysis for a number of selected countries in the ecowas region.
Not sure.
@@JDEconomics Thank you Sir. Please can panel data be used to run threshold analysis
Hi, if I convert my all series to firts different, because all has unit root, is it correct?
Hello Eduardo. As we have seen in the tutorial, before applying any transformations, you need to check whether the variables are cointegrated. If they are not cointegrated, you can use first differences and estimate your model as you normally would. Good luck!
Why the critical values from eviews are not valid?
Hi, as I explained in the video, the unit root test is performed on a regression output and not a raw series. That affects the critical values. Therefore, you have to rely on a stats table. Regards, J
Data set please
The dataset is here:
www.jdeconomics.com/eviews-tutorials/cointegration-in-eviews
Also, you can buy the slides and file at:
jdeconomicstore.com/b/cointegration-eviews
Regards
my variables are not cointegrated. what should be my next step
Use a model in differences.
Does anybody teach Bayesian time series implementation on TH-cam ?
I don’t know. Sorry
Boss, you are missing in action. hope all is fine?
Hey! Im still alive. Hoping to submit a video this week! Cheers
Transfer pricing
?
Hi,
Thank you for these great videos
Is there any way I could get into contact with you (email)?
Greetings
Www.jdeconomics.com you can find my contact information. Regards, JD
@@JDEconomics ok ive sent you a message
@@JDEconomics did you get my message? Im from Germany
@@User-12365 Im not sure. Please resend it
@JD Economics I've tried to resend it. If you give me your email I can also resend it once again.