Thank you so much!! you explain very good and slowly ... i hate those complicated materials that they try to complicate as much as possible! Econometrics is hard but explaining it good is much harder .... thanks!
+CroAuslander Thank you. I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics with me. Thank you Sayed Hossain from Hossain Academy. Please join below and post your question.facebook.com/groups/hossainacademy/
+Mohammed Hebo Thank you. I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics with me. Thank you Sayed Hossain from Hossain Academy. Please join below and post your question.facebook.com/groups/hossainacademy/
Hello, I just wanted to know when selecting the data should it be level or 1st difference incase its not stationary? The data will be used for Granger causality (where from your other videos it said the data has to be stationary for granger causality). So should we use the lag based on the stationary data set (1st difference) or non-stationary data set (level). Your help is greatly appreciated.
Hello Mr. Hossain, thank you! All the criteria tell me to choose 2 lags but when I use lag number 2 for Johansen's test and VECM, it says "near singular matrix". Does this mean that I have multicollinearity problem when I difference the data twice? I don't have any more data to add. Please advise me. Thank you!
Dear collegues I am currently stuck on testing PPP for long run effect between France and Germany during post-Bretton Woods period (April 1973-December 1990). I need to check for cointegration (Johansen test) among Fr/Ger exchange rate, Fr CPI and Ger CPI. Taking logarithms, basically, the equation looks like ln(Ger/Fr)=ln(Ger CPI)-ln(Fr CPI). Preliminary, we should test for unit root, and there is the most interesting thing: utilizing ADF(with constant and trend) and Perron's Procedure I got Ger/Fr = I(1), CPIs are both I(2). The question is how can I perform cointegration test if ranks of integration are different? Is it eligible to take the difference concerning CPIs to derive I(1) process and proceed with Johansen test? Regards
Thank you. I would like to invite you to join Hossain Academy Facebook at below link and post your question there for feedback. Thank you, Sayed Hossain from Hossain Academy facebook.com/groups/hossainacademy/
Hossain, I have a question, everywhere I read that the lag to be used in the Johansen test should be one less than the lag determined lag length criteria in VAR, but you use the same lag? Also if the maximum lag is the optimal lag, and use this in Johansen test it does not work. Another question: if the optimal lag is 1 and I use 0 0 for Johansen test (-1 lag) should I use the same lag ( 0 0) for the Vector Error Correction or (1 0)?
Sayed Hossain dear sir, i think i am pretty sure we have to use k-1 in VECM/Johansen cointegration, for k suggested by AIC/SIC/HQIC for VAR model. I repeatedly see this in different discussions. From the way i understand this, the the regression in VECM/Johansen automatically takes first difference. therefore, by including k-1 lags in those two regressions, u are involving k lag's information. but i am really puzzled what u should use for VECM/Johansen when AIC suggests k=1 for VAR.
Yes literature says so. I would like to invite you to join Hossain Academy Facebook below for greater interaction with me. facebook.com/groups/hossainacademy/
Hello Sayed Hossain When I select lag length criteria the program ask me for lags to include, I don't know any number should I insert, I have two varibles 21 observations. another question, in Johansen cointegration on the right side asking me lag intervals, how can I determine it. thanks
Abu Jumana Please join hossain academy group in the facebook and make question. Follow the following link.facebook.com/login.php?next=https%3A%2F%2Fwww.facebook.com%2Fgroups%2F841025519270747%2F843415589031740%2F%3Fcomment_id%3D843444675695498%26notif_t%3Dlike
Mr Hossain,as you showed in the video, can I use this lag selection criteria for VAR, Johansen cointegration test and also for VECM model? You showed a different criteria in another video (putting pairwise 1 1, 1 2 etc lag and selecting lowest AIC value) , which process is best?
hello, this video is great, I have a question: I am going to do a Granger Causality test on two variables for my dissertation, can I use this technique to estimate the number of lags for such a test?
Hello Sir, after watching your teaching, there are still some questions confusing me. Firstly, when we choose the lag length criteria, it requires us set the max lag length. How to set this number? In practice, I use quarterly to do this analysis. When I set the max lag length from 2 to 10, the results are almost different. From 2 to 3, the criteria shows the 2 is the best. However, from 4 to 10, the results always show that the best criteria is the largest lag length. For example, when I choose 6 as the max length, the result shows that the 6 is the best one. Secondly, does it require that the lag length for VAR, VECM and Johansen Test should be same?
Yes results will vary but if you take more lags, you will be loosing more degree of freedom and that is why I am not asking more than 5 lags as maximum. So that lag selection will be within 5.
Dear Sir Sayed Hossain, thank you for the videos. They really help me. However, I am confused how to tell which VAR type we should choose (Unrestricted VAR or Vector Error Correction) while we don't know if the variables are cointegrated? Because I thought Johansen Cointegration test is the only way to find out cointegration between variables.
If you enter 3 on the box of lags to include instead of 8 then result will be different...why this happens? and more importantly these variables does not have unit root.
Hossain sir! I've got some confusion seeing this video that you have shown lag selection for Johansen cointegration and VECM using level data. But you have not tell the lag selection for unrestricted VAR. According to your version, VAR should run on stationary data. So, should we run lag selection criteria on stationary data for unrestricted VAR separately??? Or, Does lag selection criteria give the same lag for VAR, VECM and Johansen cointegration??? If you miss lag selection for VAR in this video, please re-demonstrate the video again so that visitors will more clear. Thank you. Hoping your earlier reply.
I watched your video entitled: Lag selection, Cointegration and VECM......I found there that Firstly, you selected lag on level data that has been used in Johansen cointegration test and VECM. But if there is no cointegration, we should run only VAR. At that time, data should be converted in first differencing as your version. So, now my question is that do we have to select lag again on statationary data (i.e., after first differencing) or on level data (i.e., already selected). Thank you sir. hoping your earlier reply.
Lag selection you do only one time for the all system equation such as VAR, VECM etc. VECM automatically converts data into first differenced when u run it but VAR can not do it automatically. You have to do it.
sir i have done that, its still 12 lags, Hello, it does not show co-integration relationship, while up to six lags it shows at least one CE, what should i do. please help. Reards
thx dear..my question is why LR criteria differ from the other we cannot follow in LR the lowest value the better the model why??????? what is the criteria in LR? thx bro. about all your helpful videos
I watched the video thanks, very useful! Specifically, how can I determine "lag intervals" in VECM? You did differently at different places: 1) using "lag selection criteria" after running "unrestricted VAR" ? or 2) automatically selecting "1 1" (because VECM refers to a model with "lagged first difference terms")? thanks for your help!!
Hi, thanks for your video. I have problem that my VAR model can't pass Portmanteau test, all variable in natural log and i take 1st difference. how can i correct it ? Plz help! Thanks so much!
Sayed Hossain Thanks! I saw that some research used Portmanteau Test to selected lag lenght. It's right ?? How can I use this test cho choose the best lag lenght for VAR model
in model selection crietria we say that the model will be best model if AIC or SBC is minimum, but if in one model AIC is minimum and in other model SBC is minimum, so then how we will select the best model
Sayed Hossain I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics. Thank you Sayed Hossain from Hossain Academyfacebook.com/groups/hossainacademy/
Sir, if LR, FPE, and SC citerian suggest 6 is the optimal lag length, while SIC suggest 7 is the best one. Should 7 is the lag length I should choose? or which one has the most suggestions I should choose. like 6?
Hello Sayed, Thank you for your vide. In your video on lag selection for Johansen test you found that the maximum number of lags for the VAR model is 7. Then you use the same lag selection, namely, "1 7" in the Johansen cointegration test. Two questions: 1) The Eviews 9 guide on Johansen says that "the lags are specified as lags of the first differenced terms used in the auxiliary regression, not in terms of the levels". The Johansen test uses the ECM form of VAR. The ECM form of VAR has lags of the first differences one less that the original VAR orde. For example, if we had VAR(2) then in the ECM form there will be just one lag of the first differences. Does that imply that for example in your case you should specify for the Johansen test the lag structure "1 6" but not "1 7" The Eviews guide page is below (www.eviews.com/help/helpintro.html#page/content/coint-Johansen_Cointegration_Test.html) 2) What does a pair of the lag structure selection mean? For example "1 7": 7 means use up to 7th lag. And what does 1 mean? Use lags from 1 to 7? So "2 7" would imply use lags from2 to 7 Waiting for your comments. Regards, Argyn
Dear Argyn, I would like to invite you to join Hossain Academy Facebook at below link and post this question there, I shall respond. Thank you once again, Sayed Hossain from Hossain Academy. facebook.com/groups/hossainacademy/
+Hà Nguyễn DDo not have. Sorry. Thank you. I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics with me. Thank you Sayed Hossain from Hossain Academy. Please join below and post your question.facebook.com/groups/hossainacademy/
Thank you so much!! you explain very good and slowly ... i hate those complicated materials that they try to complicate as much as possible! Econometrics is hard but explaining it good is much harder .... thanks!
You are saving our lives! Thanks for your amazing tutorials. Now I know enough via your lessons how to run my own regression model.
Thank you for your instructional video's Mr. Sayed, you helped me a lot with my econometrics studies. You are my hero!
+CroAuslander Thank you. I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics with me. Thank you Sayed Hossain from Hossain Academy. Please join below and post your question.facebook.com/groups/hossainacademy/
I liked the way you explained time series data analysis, keep it up!
+Mohammed Hebo Thank you. I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics with me. Thank you Sayed Hossain from Hossain Academy. Please join below and post your question.facebook.com/groups/hossainacademy/
Hello, I just wanted to know when selecting the data should it be level or 1st difference incase its not stationary? The data will be used for Granger causality (where from your other videos it said the data has to be stationary for granger causality). So should we use the lag based on the stationary data set (1st difference) or non-stationary data set (level). Your help is greatly appreciated.
Hello Mr. Hossain, thank you!
All the criteria tell me to choose 2 lags but when I use lag number 2 for Johansen's test and VECM, it says "near singular matrix". Does this mean that I have multicollinearity problem when I difference the data twice? I don't have any more data to add. Please advise me. Thank you!
Singular matrix can come up for a number of reasons. One is due to setting the data wrongly in EVIEWS. You need to check this data setting issue.
Dear collegues
I am currently stuck on testing PPP for long run effect between France and Germany during post-Bretton Woods period (April 1973-December 1990). I need to check for cointegration (Johansen test) among Fr/Ger exchange rate, Fr CPI and Ger CPI. Taking logarithms, basically, the equation looks like ln(Ger/Fr)=ln(Ger CPI)-ln(Fr CPI). Preliminary, we should test for unit root, and there is the most interesting thing: utilizing ADF(with constant and trend) and Perron's Procedure I got Ger/Fr = I(1), CPIs are both I(2). The question is how can I perform cointegration test if ranks of integration are different? Is it eligible to take the difference concerning CPIs to derive I(1) process and proceed with Johansen test?
Regards
salam alikm dr sayed I appreciate your effort and thank you very much , god will reward you .
Thank you. I would like to invite you to join Hossain Academy Facebook at below link and post your question there for feedback. Thank you, Sayed Hossain from Hossain Academy
facebook.com/groups/hossainacademy/
Hossain, I have a question, everywhere I read that the lag to be used in the Johansen test should be one less than the lag determined lag length criteria in VAR, but you use the same lag? Also if the maximum lag is the optimal lag, and use this in Johansen test it does not work.
Another question: if the optimal lag is 1 and I use 0 0 for Johansen test (-1 lag) should I use the same lag ( 0 0) for the Vector Error Correction or (1 0)?
Sanne Rodenburg Yes that may be less than 1 as you said..But I used all same lags for all system equation models such as Johansen, VAR and VECM...
Sanne Rodenburg on the other question, I doubt the same. did you find the answer yet?
Yes many literature suggests that in Johansen number of lags should be less than whatever has been suggested by VAR model.
Sayed Hossain dear sir, i think i am pretty sure we have to use k-1 in VECM/Johansen cointegration, for k suggested by AIC/SIC/HQIC for VAR model. I repeatedly see this in different discussions. From the way i understand this, the the regression in VECM/Johansen automatically takes first difference. therefore, by including k-1 lags in those two regressions, u are involving k lag's information. but i am really puzzled what u should use for VECM/Johansen when AIC suggests k=1 for VAR.
Yes literature says so. I would like to invite you to join Hossain Academy Facebook below for greater interaction with me.
facebook.com/groups/hossainacademy/
Hello Sayed Hossain
When I select lag length criteria the program ask me for lags to include, I don't know any number should I insert, I have two varibles 21 observations.
another question, in Johansen cointegration on the right side asking me lag intervals, how can I determine it.
thanks
Abu Jumana Please join hossain academy group in the facebook and make question. Follow the following link.facebook.com/login.php?next=https%3A%2F%2Fwww.facebook.com%2Fgroups%2F841025519270747%2F843415589031740%2F%3Fcomment_id%3D843444675695498%26notif_t%3Dlike
Hi Sir,
Thank you. Your videos are great and have been so helpful.
Mr Hossain,as you showed in the video, can I use this lag selection criteria for VAR, Johansen cointegration test and also for VECM model? You showed a different criteria in another video (putting pairwise 1 1, 1 2 etc lag and selecting lowest AIC value) , which process is best?
hello, this video is great, I have a question: I am going to do a Granger Causality test on two variables for my dissertation, can I use this technique to estimate the number of lags for such a test?
Hello Sir, after watching your teaching, there are still some questions confusing me. Firstly, when we choose the lag length criteria, it requires us set the max lag length. How to set this number? In practice, I use quarterly to do this analysis. When I set the max lag length from 2 to 10, the results are almost different. From 2 to 3, the criteria shows the 2 is the best. However, from 4 to 10, the results always show that the best criteria is the largest lag length. For example, when I choose 6 as the max length, the result shows that the 6 is the best one. Secondly, does it require that the lag length for VAR, VECM and Johansen Test should be same?
Yes results will vary but if you take more lags, you will be loosing more degree of freedom and that is why I am not asking more than 5 lags as maximum. So that lag selection will be within 5.
I have not done yet any work with Structural VAR, You can see all my videos in
Hossain Channel.
lag selection is done for the whole system where endogeneous and exogenous both can be there.
Personally I use stationary data for any time series analysis including lag selection..
From my lag selection videos, you will try to find out a method that will ask you to use least number of lags.
hello,i just want to ask, only just for lag selection in VAR ,can we run level (NON ststiionary data) OR we have to run the stationary data
This lag selection is suitable for Granger Causality also
Failed to understand your question.
there are so many ways for lag selection. You can choose any one
Dear Sir Sayed Hossain, thank you for the videos. They really help me.
However, I am confused how to tell which VAR type we should choose (Unrestricted VAR or Vector Error Correction) while we don't know if the variables are cointegrated? Because I thought Johansen Cointegration test is the only way to find out cointegration between variables.
only just for lag selection through VAR , WE have to run the stationary data or we have to run the non stationary data?
thanks. i want to know why in johansen cointegration test we usually choose: 3) intercept (notrend) in CE and test VAR
+Kooli Wissem
I choose 3
Professor Hossain, could you please tell me when do you use exogenous variables option for the lag length?
Normally I do lag selection after making the variable into stationary
It would be better if you could run VAR with stationary data even for lag selection
If you enter 3 on the box of lags to include instead of 8 then result will be different...why this happens? and more importantly these variables does not have unit root.
Hossain sir! I've got some confusion seeing this video that you have shown lag selection for Johansen cointegration and VECM using level data. But you have not tell the lag selection for unrestricted VAR. According to your version, VAR should run on stationary data. So, should we run lag selection criteria on stationary data for unrestricted VAR separately??? Or, Does lag selection criteria give the same lag for VAR, VECM and Johansen cointegration??? If you miss lag selection for VAR in this video, please re-demonstrate the video again so that visitors will more clear. Thank you. Hoping your earlier reply.
You see one of my video entitled: Lag selection, Cointegration and VECM.....You will find the sequence there...
I watched your video entitled: Lag selection, Cointegration and VECM......I found there that Firstly, you selected lag on level data that has been used in Johansen cointegration test and VECM. But if there is no cointegration, we should run only VAR. At that time, data should be converted in first differencing as your version. So, now my question is that do we have to select lag again on statationary data (i.e., after first differencing) or on level data (i.e., already selected). Thank you sir. hoping your earlier reply.
Lag selection you do only one time for the all system equation such as VAR, VECM etc. VECM automatically converts data into first differenced when u run it but VAR can not do it automatically. You have to do it.
thanks for your videos, but you talk too slow, could make your 20 min videos into 5min
Vlad Ionescu Yes I shall keep your advice in my mind....
Thank you Vlad.
Set speed to 1.5
if the variable are non -stationary, is it okay to put them without differencing when we run VAR TO get lag length ?
sir i have done that, its still 12 lags, Hello, it does not show co-integration relationship, while up to six lags it shows at least one CE, what should i do. please help.
Reards
thx dear..my question is why LR criteria differ from the other we cannot follow in LR the lowest value the better the model why??????? what is the criteria in LR? thx bro. about all your helpful videos
I watched the video thanks, very useful! Specifically, how can I determine "lag intervals" in VECM? You did differently at different places:
1) using "lag selection criteria" after running "unrestricted VAR" ? or 2) automatically selecting "1 1" (because VECM refers to a model with "lagged first difference terms")?
thanks for your help!!
Thank you for your lecture. I wonder why you included the number of lags of 8, not another number?
I finalized lag value 3 and when I used cointegration test it showed near singular matrix then how to solve this problems?
Do you have any video on structural VARS? Please response
Hi, thanks for your video. I have problem that my VAR model can't pass Portmanteau test, all variable in natural log and i take 1st difference. how can i correct it ? Plz help! Thanks so much!
In that case u go for trial and error. Add some relavant variable or deduct some irrevalent variables and then try again...
Sayed Hossain Thanks! I saw that some research used Portmanteau Test to selected lag lenght. It's right ?? How can I use this test cho choose the best lag lenght for VAR model
Great to know.
sir i have seen this video and run the lag selection criteria, all five LR FPE AIC SC HQ shows 12 lags. please help me , its my last year of PhD.
in model selection crietria we say that the model will be best model if AIC or SBC is minimum, but if in one model AIC is minimum and in other model SBC is minimum, so then how we will select the best model
Najeeb Ullah Both methods are efficient. You can choose any one as benchmark and can take decision.
Sayed Hossain I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics. Thank you Sayed Hossain from Hossain Academyfacebook.com/groups/hossainacademy/
Sir, if LR, FPE, and SC citerian suggest 6 is the optimal lag length, while SIC suggest 7 is the best one. Should 7 is the lag length I should choose? or which one has the most suggestions I should choose. like 6?
Thank you so much
SORRY IS NOT SIC, IS AIC.
zonghao wang Please join Hossain Academy facebook group where we can discuss all issues. Link below. @groups/841025519270747/
hello thanks for all your works ... i have problem with my vecm model how could i remove LM and heteroscedasticity from VECM model ...thank you
Convert all variables into natural log and run it again
You are welcome
sir, how to run 2SLS model in eviews?
thank you very much!
thank you very much sir for reply, i have monthly data. could you please suggest me what should i do.
Hello Sayed,
Thank you for your vide.
In your video on lag selection for Johansen test you found that the maximum number of lags for the VAR model is 7. Then you use the same lag selection, namely, "1 7" in the Johansen cointegration test. Two questions:
1) The Eviews 9 guide on Johansen says that "the lags are specified as lags of the first differenced terms used in the auxiliary regression, not in terms of the levels". The Johansen test uses the ECM form of VAR. The ECM form of VAR has lags of the first differences one less that the original VAR orde. For example, if we had VAR(2) then in the ECM form there will be just one lag of the first differences.
Does that imply that for example in your case you should specify for the Johansen test the lag structure "1 6" but not "1 7"
The Eviews guide page is below
(www.eviews.com/help/helpintro.html#page/content/coint-Johansen_Cointegration_Test.html)
2) What does a pair of the lag structure selection mean? For example "1 7": 7 means use up to 7th lag. And what does 1 mean? Use lags from 1 to 7? So "2 7" would imply use lags from2 to 7
Waiting for your comments.
Regards,
Argyn
Dear Argyn, I would like to invite you to join Hossain Academy Facebook at below link and post this question there, I shall respond. Thank you once again, Sayed Hossain from Hossain Academy. facebook.com/groups/hossainacademy/
Hello Sayed, I posted it on the Facebook page. Best, Argyn
Change the variable set
Hello, Sir. Could you give me your tapescript of this lecture please? Thanks advance.
+Hà Nguyễn DDo not have. Sorry. Thank you. I would like to invite you to join Hossain Academy Facebook for greater interaction about economics, finance and econometrics with me. Thank you Sayed Hossain from Hossain Academy. Please join below and post your question.facebook.com/groups/hossainacademy/
Hi, how can I make my data Stationary? Thanks.
hai sir, could you tell me how to overcome near sngular matrix in vecm? thanks
risma widyawati Normally it happens when you set the eviews command in wrong way.
sir i am getting 12 lags as optimal, is that ok or too many, please tell me what should i do
sir i have tried, its still 12 lags suggested by all criteria.
it is monthly data sir, please suggest me what should i do. i am really worried.
can you help me in interpretation my result??
hasan shaikh If you have any question please do so, I guess I can assist you
agreed, it was not clear. I will rephrase it .
Check Unit Root Testing in Hossain Channel
thanx
Ok convert all variables into log then try
In this video I have chosen lag 2, not lag 8
It is to many. Please try to do something else
It would be better if you could run VAR with stationary data even for lag selection