Vector Error Correction Model (VECM) - Step 4 of 4
Вставка
- Опубліковано 7 сер 2017
- This video demonstrates the estimation of the VECM on EViews. Additionally, I provide interpretations of the output. Sorry, I inadvertently omitted the current values of X (i.e. Xt=0) in the estimation. But no worries, the process of estimation and interpretation remain intact. Enjoy :-)
This 4-part series was really useful for me when I was studying for my MSc. So much so, that it inspired me to upload 5 of my own videos on estimating a VECM and how to interpret the results.
Thanks so much, Professor. I had searched for so long time the kind of your video. All the best. From Mizoram, India
thank you so much!! you have brought me to another level of understanding of VECM. :)
Magnificent presentation! Thank you very much for the lesson!
thank you pat for helping people in understanding VECM model.
Your explanations are very clear.Thanks for this
You got a great teaching skill. Keep it up!
Crystal clear as always, thank you very much prof!
You are one of my best mentors. Thank you
Sir thank you very much for such a simple and nice interpretation. its a great help to the researchers.
This is very helpful Sir. Thank you so much for sharing this materias. 🙏🏻
Perfect explanations there, thanks
Thanks Pat, excellent insight.
thx for making this video!!! it helps a lot
Very helpful! Congratulations!
this was really helpful, thanks a lot.
thank you for wonderful teaching sir... it helps a lot..
Thank you for your good video.
Thanks so much Pat, your videos are very informative indeed! Can you please do a video to show how you can impose a cointegrating restriction using Eviews? I couldn't find a video on this anywhere....
Thats amazing, thank you kind sir.
You got a mazing teaching skill..thnx
well explained, thanks man
Dear Dr. Pat Obi. Thank you so much for such videos .. you have helped me a lot to finish my recent research paper. However, may I ask which software did you use to capture your lecture? I am really interested in recording my lectures as well, but I don't know the way!
Great but what happened to the C3 and C4 that came out insignificant and you started carrying out tests on them to point to the problem? Secondly do we report the long run coefficients from the long run equation the way we report in case of OLS ?
For the VECM, would you not have to drop from 2 lags to 1?
sir can you help me? i have panel data and i make the steps you say but in the end on eviews it doesent appears the lm test or cusum test?why ??please help me
Thank you for this video! However, I am still not clear if I can use VECM for panel data containing 4 variables, 2 of which are I(1) and 2 are I(0). Could you please tell me if VECM can be used? Thanks!
Hello sir
i have some questions on the error correction term. as u have mentioned in your lecturer that if the coefficient of the error term does not fall within -1 and 0 then there is problem with the model . in such case what is your suggestion to correct the model. I have my error term -2.00 and am confused about how to deal with it.
i tried to reduce the lag length of my model from 4 to 3 and the error term falls within the range. if i do this step will it be a problem or is it ok to reduce the lag length. I have a sample of 30 observations and when i generate the optimal lag, the model gives me 4 lag.
if you could clarify on this.
thank you
Thanks so much for your explanatory video. I found 2 vectors with my variables with the helpof Johansen test so I want to ask what if my 2nd cointegration equaiton CE(2) is positive and not significant, I can still say that there is long run relationship? because my 1st cointegrated equation CE(1) is negative and significant. Please help me about this issue. Thanks..
Dear sir, can you please tell me how can I write asymmetric Vecm model after VECM analysis?
Thank you so much !!!! i'm gonna graduate soon!!! :)
thank you so much!!
Good day Dr,
if the result for cointegration is not cointegrated for linear trend but cointegrated for Restricted linear trend, should we change the default setting of VECM from 3)linear trend to 4) Restricted linear trend when we want to run VECM?
Thank you for your time if you read and comment on this DR.
I have non cointegration data but when i change trend specification to be anything except beside constant i find cointegration, how to determine trend specification?
Question: The VEC model have two cointegrating equations. The coefficient, C(1) is negative and significant, but C(2) is positive and significant, what will be the overall conclusion for longrun equilibrium?
a question if there are two variables, the result is also two equations of vecm because only the first equation is found and why not the second equation, is it not possible to make an inference?
Sir I have 5 variables in my vecm and 4 cointegration equations . Can I proceed with the model and how to interpret each cointegrating equation
Dear sir,
In the video where you estimated an unrestricted VAR, you suggested that 2 lags were optimal to use based on SIC criteria. If I understand correctly, one must use (p-1) lags for Johansen's test and for VECM (where p is the optimal number of lags suggested by unrestricted VAR). But then you use 2 lags for both of these instead of 1. I am a little confused about that. Thank you in advance!
Can Impulse response function can be applied to VECM or it is applicable to VAR only??????
Dear sir, can u make other example about VECM analysis? i am very insterest with ur explaination.
If i am not mistaken, VECM must have -1 VAR lag numbers. If you choose VAR lag 2, so VECM wll be lag 1. Right?
thanks a lot for this! question: what does it mean when C(4) and C(5) have high p-values (insignificant) in the ECM but indicate granger causality in the wald test?
Sorry, not sure. I'm sure there's a good explanation for that anomaly.
@@PatObi In this example, would we say that the VECM is poor given the very low adjusted r squared?
Prof ,thanks for your explanations , from the VECM how do i determine my speed of adjustment at long-run. Thanks you.
The speed of adjustment for each variable can be calculated by multiplying the coefficient of the variable in the cointegrating equation by the coefficient of the ECT where the first-difference of that variable is the dependent variable
. If you don't understand what I mean: ua-cam.com/video/FvCuHqqdasc/v-deo.html
Hello Dr. Firstly thank you for your videos.
I have a questions regarding your residual testing.
It appears that you are using the chi-squared P-value rather than the F-stat p-value within your residual testing (LM, hetro).
Why do you choose chi squared over f-stat?
Thank you Dr :)
I believe any can be used.
@@PatObi thank you
Thanks Pat for this video. My question is what If i get cointegration in trace statistics but not in Max Eigenvalue statistics? can I proceed to the VECM model with such results?
I would. Perhaps you might find evidence of l.r. causality. But be sure to ask others.
@@PatObi Much obliged.
Thank you for your good explanations and helpful videos.
I have a question. The objective of my research is to study the direction and magnitude of response of one variable due to a change in another variable. For that, I am planed to use SVAR and generating IRF. Pls tell me if variables are I(1) and there is cointegration then should I generate IRFs from VECM or from the unrestricted VAR? which approach is better?
Here's my take but I'm not completely sure: If variables are I(1) but not cointegrated, the OLS regression we run is "unrestricted VAR
." In this case, we difference each series and then run OLS on the differenced series to examine their short-run dynamics. In your case though, the I(1) variables are cointegrated. So you can estimate (1) OLS regression on the levels data (unrestricted VAR) to examine the l.r. equilibrium relationship and (2) ECM, to examine the speed of adjustment to l.r. equilibrium
The following ResearchGate Q&A might help: www.researchgate.net/post/Can_we_use_unrestricted_Var_model_if_the_variables_with_I1_series_are_not_cointegrated
Thank you
Thanks
Sir I have run VECM residuals diagnostic but my model found non normal and hetroskedastic residuals but it solution for it
I already taking my variable as natural log form.
What can I do for this problems
Pls rpy
Some of the remedies may include (not guaranteed) increasing sample size, changing the lag structure, changing data frequency, e.g. from yearly to quarterly or monthly, etc.
The lecture has been most helpful. However, I wish to know how one would conduct a wald test when they used one(1) lag and there are two explanatory variables?
Go to Coefficient Diagnostics, click on Wald Test and specify all the short-run coefficient numbers, e.g. C(3)=C(4)=0.
If the coefficient of the ECT is positive and significant, then can we say that there is long run causality? Because the coefficient is significant?
No. Positive ECT coefficient means there's no convergence.
Professor Pat Obi, I Iove your classes, and I have three questions for you. 1) In case one of the series has no unit root at level but the second one has it, and both of the series have no unit root at first difference. Are they candidates for a VECM model? 2) When you ran the VECM in the video, the R squared was 0.19. Is R squared or F statistic important when calculating the VECM? 3) I am currently running a VECM for two time series, but there is autocorreclation in the residuals. What would be the best way to tackle this issue? (10% percent of the data are outliers). Thank you very much, and if you teach virtual classes please let me know, so I can have one of your classes.
If one variable is I(0) while the other is I(1), you should not run a VECM. Instead, run ARDL (I have a 6-part video series on ARDL here on UA-cam). Don't worry too much about R-sq and F at this stage. Instead, focus on whether the series are cointegrated. Also, examine the short run dynamics regardless of whether the two series are cointegrated. Residual correlation is definitely not good. Try using different lags. Also, you could try different data structure (daily, weekly, monthly, quarterly, annually). One might also consider adding another regressor. Hope this helps! Thanks for subscribing to my channel.
Thank you very much, for all your answers, Professor@@PatObi! I am going to check the ARDL videos.
getting an error of insufficient number of observations while running VECm, kkndly guide sir
Sir please add video on multiple structural breaks and how to deal with these.
THANK YOU SO MUCH SIR FOR MAKING THE CONCEPTS SO CLEAR. I HAVE DID AN EXAMPLE IN WHICH MY RESULTS ARE QUITE GOOD EXCEPT THAT MY ESTIMATED MODEL PROBABILITIES WERE GREATER THAN 0.05.SO ANY OF THE COEFFICIENTS WERE NOT SIGNIFICANT. SO, WHAT IS THE INTERPRETATION? COULD U GIVE ME A SUGGESTION, SIR????
jaya subha: If the short run coefficients of the explanatory variable are jointly not significant, then there is no short run causality.
thank u :)
Hi professor. My three variables are all I(1). According to Johensen test, there are 2 cointegrations. However,when I construct VECM, C(1) is not significant and sometimes not negative.
I have no idea where the mistake exists because I followed your method step by step.
Please watch from the 12th minute for some ideas
Dear Prof. I am Tyrone from Sri Lanka once again. How could I start learning time series model fully. I have no basic knowledge . would you advise me on this
tyrone from Sri Lanka
This 4-part series is a good start
I think that you must take the p-1 lag in the VECM estimation, which is 2(var lag) - 1 = 1. Thank you for your efforts.
What happens if the optimal lag is p=1? In estimating VECM do we still subtract the lag so that we get p-1=1-1=0?
@@alphonsedismasorango8406 yes absolutely. In this case, you will take zero as the VECM's lag. Good luck.
@@yassineounnabi4396 , thank you for the quick response. However, when I do that, it gives me an 'illegal lag specification' error. What should I do? Thanks again for your response.
@@alphonsedismasorango8406 in the lag criteria for the var, try to enter 4. Then I think that the criteria chosen will be 3. Then, try to estimate the VECM by 2 lags. If this strategy doesn't work, try to change the number of lags tested in the VAR. Try this and tell me the results.
@@yassineounnabi4396 , thanks. do we use the AIC, SIC or others in selecting the optimal lag? My AIC indicates one lag same as the SIC. Kindly advice.
if non- stationary time series are integrated of the first order I(1) found not co-integrated , then what should we do ??
Test for only short run causality.
@@PatObi do you have video for that one ?
@@chhitijpoudel156 Yes. Watch from about the 8 minute till end. It's part of the VECM. Also, check out my 6-part ARDL series. One of parts is on s.r. dynamics.
@@PatObi thank you so much sir!! 🙏
why don't you reverse the sign of the coefficients?
Sir thank you for your excellent explanations. I just confuse cos I got one variable (logTT) zero coefficient for long run how I explain it? can i remove it from the equation? can we use your model for more than 2 variables? thanks a lot
Vector Error Correction Estimates
Date: 02/02/19 Time: 20:53
Sample (adjusted): 2007M04 2017M11
Included observations: 128 after adjustments
Standard errors in ( ) & t-statistics in [ ]
Cointegrating Eq: CointEq1 CointEq2
LKSEI(-1) 1.000000 0.000000
LOG(TT(-1)) 0.000000 1.000000
LHINDEX(-1) -0.154923 -1.038932
(0.25311) (0.08584)
[-0.61208] [-12.1034]
LOG(INVP(-1)) 0.435698 0.245088
(0.18477) (0.06266)
[ 2.35810] [ 3.91135]
C -11.12377 -4.368917
Obi why in delta y i is 1 and in delta x i is 0
What should you do when your results have no cointegrating relationship
Then check only for short-run relations. No cointegration means no long run relationship.
@@PatObi Thank you. This video really helped.
Sir, In the Long Run Model, how we explain it ? Increase in oil prince positively affecting the FX ? Or the sign reversed in Ect .. Please help
The sign is reversed in ECT. To see why this is, set ECT to equal 0. Eg. 1.00FX_t-1 + 0.003Oil_t-1 -1.0308 = 0. Therefore, 1.00FX_t-1 = -0.003Oil_t-1 + 1.0308. Just in case you're wondering why we set ECT to equal 0, it's because ECT should have 0 mean.
thank very much , but i have encountered this problem , all my C(1).........c(7) , are insignificant , what should i do now?
Coefficient Std. Error t-Statistic Prob.
C(1) -0.030998 0.024244 -1.278581 0.2218
C(2) -0.082780 0.360631 -0.229543 0.8218
C(3) -0.030502 0.341345 -0.089357 0.9301
C(4) -0.107738 0.297568 -0.362061 0.7227
C(5) 0.356619 0.301433 1.183077 0.2565
C(6) -0.012327 0.236723 -0.052073 0.9592
C(7) 0.200735 0.224905 0.892533 0.3872
C(8) 3087.805 1536.751 2.009307 0.0642
I believe those are the short run coefficients. Testing them jointly using Wald test will confirm that in the s.r., your X-variable does not Granger-cause your Y variable. That's what I think :-)
@@PatObi please can you send me your mail and whatsapp number.
thank you very much.
@@PatObi my whatsapp number 00818602552355
ijazyounis@njust.edu.cn