(Stata13): VECM Estimation, Discussion and Diagnostics
Вставка
- Опубліковано 5 сер 2024
- So, what do you understand by vector error correction model (VECM)? You may say any of the following: that it is a system having a vector of two or more variables that all the variables in a VECM are endogenous there are no exogenous variables VECM is constructed only if the variables are cointegrated cointegration implies evidence of a long-run relationship among the variables it is a restricted VAR model with cointegrating restrictions built into the specification constructed to examine long- and short-run dynamics of the cointegrated series restricts the long-run behaviour of endogenous variables to converge to their cointegrating relationships that the cointegrating term is known as the error correction term it is a representation of cointegrated VAR (courtesy of granger’s representation theorem) and that the resulting VAR from VECM representation has more efficient coefficient estimates. Also, note that VAR specified in differences is a mis-specification while VECM is obtained by differencing a VAR, hence losing a lag. So, you construct a VECM with a (p-1) lag lengths for all the variables in the system. These are the basic steps required to estimating a VECM. (1) series must be stationary (integrated of same order) (2) determine optimal lag length for the model (3) perform Johansen cointegration test (4) if there is no cointegration, estimate the unrestricted VAR model (5) but if there is cointegration, then specify the restricted VAR model (i.e. VECM). In this video using Stata13, I show you the rudiments of the VECM specification. Kindly check my channel and playlist for all simple and exciting hands-on tutorials using EViews, Stata and Excel applications.
Here is the link to the ex21-1.wf1 dataset (EViews file) used for this tutorial (endeavour to have a Google account for easy accessibility): drive.google.com/drive/u/1/fo...
Follow up with soft-notes and updates from CrunchEconometrix:
Website: cruncheconometrix.com.ng
Blog: cruncheconometrix.blogspot.co...
Forum: cruncheconometrix.com.ng/blog/...
Facebook: / cruncheconometrix
UA-cam Custom URL: / cruncheconometrix
Stata Videos Playlist: • (Stata13):Estimate and...
EViews Videos Playlist: • (EViews10):Interpret V...
UA-cam 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, support my Channel with your subscription and sharing my videos with your cohorts.
could understand comprehensively. Thank you. Subscribed also.
@@khankhalid7 Thanks for your subscription Khalid......may I know from where (location) are you reaching me?
I did subscribe, these lessons are refreshing indeed. Am estimating the VECM in stata. Why is it advisable to estimate only 1 equation when I have 2. ?
@@DicksonKhaingaMr Hi Dickson, thanks for your subscription. Deeply appreciated. The reason is just to simplify analysis and explanation. May I know from where (location) you are reaching me?
The video is very helpful for my research. My question is in Johansen normalization restrict, I got one variable which has positive relationship (negative sign) but p value is not significant (0.386). In that case how I interpret it? thanks.
I'm so grateful. Keep it up with the excelent work, teacher!
Kisses from Brazil
Thanks for the encouragement, Rev!!! Appreciated!
Thank you for the amazing videos and clarity of them too, this video has allowed me to fully understand my data output for the VECM for my undergraduate thesis which I otherwise would have been struggling with!
Thanks, Kyle for the encouraging feedback. Deeply appreciated! 🙏☺️
These are great videos! Thank you!
Thanks for the encouraging feedback, Tim...deeply appreciated!
So clear and detailed, thank you!
Glad it was helpful, Celestino!
Thank you, Professor. Your videos are really God sent, extremely understandable. I could not have completed my Masters without them. God bless.
I'm encouraged by your feedback, Thato. Thanks for lifting my spirit and wish you more progress in life! 🙏❤️😊
Dear Professor, thank you for this amazing video.
You are welcome, Khant😊
This video deserves so much more love, it doesn't have the number of likes it deserves
Thanks for the encouraging words and feedback, Nelson. Deeply appreciated! Please may I know from where (location) you are reaching me?
@@CrunchEconometrix of course. Im watching your videos from Central America
Was looking for something concrete since morning and then I found this. Got immense relief. Thank you so much ❤️
Thanks for the positive feedback, Mri. Deeply appreciated! Please may I know from where (location) you are reaching me?
@@CrunchEconometrix You are welcome. I'm from Mumbai, India.
Thank you so much for this video! So clear and helpful!
U're welcome, Monique...may I know from where (location) are you reaching me?
@@CrunchEconometrix I am doing my master's coursework in London. So glad to have found your video!
Monique Adi Suwiryo Ok girl, don't keep me to yourself (lol)...kindly share my YT Channel link with others too!😀
Thank you so much professor. Like other comments stated, your video has been extremely helpful. I would not have managed to complete my master thesis without it. May god bless your kind soul
Wow!!! I'm so encouraged by your positive feedback. Deeply appreciated and may God bless you, amen 🥰🙏
Great content, thanks from UK
You're welcome, Rowan...thanks!
This video has been very instrumental in my final year research project. I highly appreciate the insights I have learnt. God bless you professor.
Hi Leo, u just made my day with this comment. Glad to be of assistance...and you owe me one! 💕 Please share my videos with your academic community and colleagues...gracias!😎
@@CrunchEconometrix I certainly will do share. I am sure it will be of help to many young researchers. Have a wonderful afternoon Prof.
you're the best, just saved my Seminar paper on time ;-;
Thanks a lot, love from India!
Happy to help Somrat!
Excellent video!
Thanks, Emre for the encouraging feedback. Deeply appreciated!
Thank you so very much! Ma'am. For delivering in such a elucid manner.
It's my pleasure, Divya!
@@CrunchEconometrix I wish you were here in India.
Hahahaha, Divya. Hopefully, I will lecture there someday. Fingers crossed!
@@CrunchEconometrix I'll pray to god. Soon the day will come.
Good video on longrun model.
Glad it was helpful, Haggai!
Cheers from Greece! OUR QUEEN OF ECONEMTRICS!
Hahahaha...thanks, John for the title. Humbly taken. Much love from Nigeria 🇳🇬! 🙏 ❤️
Amazing sir thankyou so much...
You are so welcome 💗
thnks a lot for your video it has been really helfull
Good to hear, Nelson👍🏽
Thank you very much madam, you made it much more easier to understand !
Hi Ismael, thanks for the positive feedback. Deeply appreciated! Please may I know from where (location) you are reaching me?
London, UK!
Sorry for the late reply.
As always Very helpful ☺️.You're amazing. A quick question, do you always have to interpret the long run model?
Thanks so much for your encouraging feedback, appreciated 🙏🥰. Yes, good to interpret the long run results.
Thanks for your amazing work! I wanted to confirm that this is valid only if there is cointegration found in all the three equations, right? If only one was found to have cointegration, for instance,say the first one, with pdi as the dependent variable, we will use the the ecm framework instead of the vecm framework for the first equation,which will give us both the long run and the short run relationship. For the remaining two equations, we will use the ardl framework ,which will give us only the short run relationships. Am I correct in my interpretation?
Please do not confuse VAR/VECM with ARDL/VECM. The mechanisms are not the same.
Hi Dr Ngozi, Thanks again for your video. In your example on specifying VECM model, there are 3 variables, and you use some Greek letters (beta, gamma and phi) for coefficients, that is the short run dynamic coeficients of the model's adjustment long run equilibrium . I would like to ask, if i have 6 variables, what Greek letters do I use to denote coefficients?
Hi Alfred, you can use any Greek alphabet. Kindly do a Google search on that. Thanks.
Thank you! By the way, as VECM's are for I(1)'s, I saw you chose to execute vec on variables that are not differenced (9:20). I assume Stata differences I(1)'s for us automatically just as it does p-1 automatically? Should I execute vec on a set of level variables or first-differenced variables in Stata?
Hi Mint, my explanations are quite clear. I will advise you watch the VECM series all over again...VECM deals with I(1) and not I(0) variables.
Thanks for the amazing work! I was facing a question regarding Step#3 where you say that the series must not be I(2). Why is that the case? Since we are performing the Johansen Cointegration test, would it not be okay if all the series were I(2) in order to check for cointegration and perform the Johansen test. I thought that I(2) variables should not be present in only if we use the ardl model .Thanks !
VECM from VAR analysis requires all series to be I(1).
Hello, ma'am
If the AIC, HQIC and SBIC agree that my optimal lag length is 6, am I to use it for my co-integration test or go with the default 2 lags in the menu?
Similarly, am I to use the 6 lags to estimate my VECM or just the 2 lags that appear in the VECM menu by default?
Many thanks
First, watch my video on OPTIMAL LAG SELECTION. It will help you understand the use of lags. Second, re-watch my videos on VAR-VECM and adapt to suit your data.
One of my model's dependent variable has I(2) stationarity and its cointegrated with RANK2 with other variables . The independent variables are all stationary at I(0) and I(1). Can I still proceed with ARDL or is there any way out for this .
Hi Jyothsna, kindly watch my foundation video on ARDL. Details all the nitty gritty about the technique. Thanks.
Thank you very much, subscribed as well, I am writing my final year project in my university, and your videos really really helping me for doing that. However, I got a real quick question if you don't mind, what was the point making variables stationary if we don't use those d.pce, d.pdi and d.gdp anywhere? I really apprecite if you reply to my question)
Hi Mubinjon, kndly watch my VAR videos for detailed response. Thanks.
Thanks for making these tutorials. Can I run VECM If my variables are stationary at level and at first difference? Or I have to use AEDL-ECM approach as the three variables have long run equilibrium?
ARDL/ECM approach.
Hello,
Is it possible to run VECM if some variables are not stationary after the first difference? Can you run VECM with I(2)? Thanks!
No, you can not. VECM must be run with I(1) variables.
Hello,
If I run the Johansen test and it shows no cointegration, do I just stick with my VAR model and abandon VECM?
Thanks.
Yes, Anoush.
Final question. And btw thanks so much for the help!! How do i interpert the short run coefficients. For example in the D_lnpdi section, what do the coefficients show and how to know if they return to equilibrium in the long run?
Thanks
Any variable with a "D" indicates a SR variable. The sign of the ECT tells if there is reversion to long-run equilibrium.
Hi,
Are the notes and descriptions you talk about take from a particular book?
Thanks
I use different sources. See reference list at the end of the video.
More understandable video
Thanks Sawiki!
Professor, Your videos are great. I have shared your videos with many students and friends. It would be great if you could show how to create a publication like a table using VECM results. Thank you. Best, HP
Hi Hum, thanks for the encouraging feedback and for sharing my videos with your students and academic network. May God bless you, amen 🙏. You will find my videos on exporting Stata output to Word or Excel. I showed publication formats in those clips. Please may I know from where (location) you are reaching me?
Professor I have a question. What if the veclmar of all lags gives us the Prob>chi2 values below than 0.05? What lag we choose then? Thank you in advance.
Increase your lag length if you have sufficient data points. Choosing lags is an empirical issue. Watch my video on Optimal Lag Selection.
Great video. I have a question though. If I have 2 lags in the VECM (as opposed to the 1 in the example), does the notation for the ECT change? Would it change to ECT(t-2)=[pdi(t-2)-3.6pce(t-2) + ...] ? How would you also interpret the LR effects?
Hi Max, thanks for the positive feedback but obviously you are yet to understand VECM technique. My advise is that you watch the VECM videos again and READ published papers for proper understanding. Regardless of the lag structure, ECT is ALWAYS with one lag.
@@CrunchEconometrix Great thank you!
How do you interprete the shortrun model? Do you also reverse the signs like in the long run model?
Hi Haggai, only signs of the Johansen Cointegration results are reversed not the VECM. Don't mixed it up and give the usual ceteris paribus interpretation for both long-run and short-run results.
I had one doubt in this one, in my model the AIC selection criterion gives me p=4, so while performing VECM, should I mention 'maximum lag...' option to be 4 instead of 2 as in this video?
Thanks!
You can estimate with 4 lags if your time series is long to avoid loss of observations.
Stupid question- I am confused as to how to construct the cointegrating equation. Say I have 4 variables and rank 1, meaning there is only one cointegrating equation. But before I run the VECM, I construct 4 independent equations with the each variable as the subject. After running the VECM, which equation will I select to carry on for my interpretations? And how do I discard or comment on the rest? Thank you! I hope I make sense..
Zaeem, my VECM covers about 95% of what needs to be done by any researcher. But you can check out other online resources for more information. Thanks.
I have question related to the VECM model. If I specify "trend" in the VECM model then how to do the post estimation tests. Because when I specified trend in the VECM model and after that doing the post estimation tests using the active vec results does not show any results in Stata.
Cyborg, I don't understand what you mean by specifying trend. Can you be more explicit? Thanks
Warm greetings @Cruncheconometrix! while I am doing ardl bounds test, the F-statistics falls below the lower bound critical value. So, as you explained in your video we couldn't reject the null, which implies that there is no long-run relationship among variables. If so, what next? I mean what am I gonna do?
Solomon, your query is about ARDL and this video explains VAR/VECM. Kindly post your query on the appropriate video thread for others to learn from the discussion. Thanks.
Hello, thank you for your videos, they are very clear and helpful. I have, however, a question. You often say we cannot have exogenous variables in a VECM. Is this really true or is it a simplification? I've read in other sources that we can do it by treating exogenous variables as dummy variables in STATA, but it is not very clear how to proceed. Can you help? Thank you
Hi Diogo, I teach what I know. You may want to check out other online resources for exogenous VAR modelling. Thanks
Hello! After running VECM, i found that error correction term (ce1) is negative and significant but one variable in the Johansen Normalization Restriction Imposed is not significant. Does that mean the insignificant variable is omitted in the ECT equation?
Hi Kelv, not at all. It means that variable has NO statistically significant relationship with the depvar
Thanks a lot for these videos. I have a question: if I find 4 cointegration relationships in Johansen (in step 5), how many cointegrations should I put in the main analysis (step 7)? should I keep 1 cointegration as a default?
Mohammad, I already mentioned this in my video on Johansen cointegration that you use ONE, except you are able to interpret results with more than one CE.
@@CrunchEconometrix Thank you so much
If the adjustment term is not significant how can we interpret? Please, I need your answer.
Hi Shakira, it means there's no convergence to long-run equilibrium.
Thank you for the video. May I know the intuition behind the interpretation of the Johansen normalization restriction table regarding the signs of the coefficient. Thank you.
Kindly check my Playlist to watch my video on the JOHANSEN COINTEGRATION method in Stata and EViews.
hello teacher i have a question , when i use 'vecrank EU_Price EU_Yield' , stata respond :the sample has gaps
but i have checked data
drop if EU_Price==.
drop if EU_Yield==.
stata said (0 observations deleted)
but i implement 'vecrank' again
the answer of stata is the same "the sample has gaps"
how come ? what can i do?
You told Stata to drop both variables on the condition that they have NO observations. That is, if you have data from 1980 to 2017, you want both variables dropped because they have 0 observations across those years. What you want Stata to do is to remove those years having no observation within the dataset. In my opinion, this will distort your dataset and reduce your time span apart from the fact both variables do not have equivalent missing obs in those years. The best solution: change those variables with closer proxies having sufficient data points. OR post this query on Stata.org FORUM for further assistance.
Thank you very much for the insightful video. I would like to ask if VAR and VECM can be used where all the variables are stationary at level. Your assistance on the matter would be highly appreciated.
Not at all, Luther. OLS will do in that case.
hi , why always in the long run stata omitted ther first or second...... variables? for exemple rank:3 , stata omitted the 3 first variable . haw can i interpret them when i don't know their pvalue ( i have just the coef) ?
Omission could be due to several factors which I may not have the answer. You can post your query on Statalist.org for more constructive feedback.
Do we need to reverse the signs of long run coefficients in the report as well?
Yes.
Thanks for this video. However, I am at loss at what to look out for in the vecstable output. How do I determine, through the vecstable output that the vecm is stable?
Hi Kenechukwu, for more clarification you can go through published papers on VECM. I have always maintained that video tutorials are insufficient. They must be complemented with readings. This is because no one gets all from one video/source.
Hi, I have one doubt why did u say we keep it simple by using 1 con integrating equations during the estimation of the VECM model. In my model, I got 3 cointegrating equations .what should I do?
Kiara, I explained that in the video. If you can interpret 3 CE, please go ahead.
Thank you so much for this video CrunchEconometrix. I am writing my thesis and this video has been my biggest help so far! Is there any way to support you besides subscribing?
Quick question: in the video you specify rank(1) to "keep it simple". I am estimating a vecm and there are 6 cointegrated vectors. I am only interested in explaining one of them (exchange rate). Can I follow the procedure in your video, even when vecrank told me there are six cointegrated relations?
Hi Hero, thanks for the offer to support my Channel. I'll appreciate if you help share my videos with your friends, colleagues and academic community. I need the global academic community to be aware of my Channel. My niche is to assist students and upcoming researchers.
...yes, keep-it-simple, always use 1 cointegrating equation (CE). As a confirmation, check papers on VECM to find that rarely will you see any using more than 1. Interpretations will be a bit clumsy with more than 1 CE.
I have found each of your videos so so helpful. Thank you! May I please know...my series are only stationary after second differencing. Can I still go-ahead to perform VECM?
Hi Adjeley, thanks for the positive feedback. Deeply appreciated! No. Use the Toda-Yamamoto procedure. Check out other online resources for this. Thanks.
@@CrunchEconometrix thank you so much. I’ve sent an email and Facebook message too, please. Urgently awaiting your response
I responded to you on Facebook.
Thank you for the video!
I have a question: what should I do if the results for both Lagrange-multiplier test and Jarque-Bera test show a Prob>chi2 below 0.05? Does it mean that the model is biased and useless?
I started with level data and tried to improve my results with log but I got the same results.
Hi Alessis, these indicate that the model suffers from serial correlation and its not normally distributed. To control these, several measures may be required depending on your data and variables such: including a lag of the depvar as a regressor, replacing some variables with better proxies etc. Try these and observe the outcomes, thanks.
@@CrunchEconometrix But since I'm applying a VECM model I already have a lag of the depvar as a regressor. What do you mean? Should I increase the number of lags and disregard what the optimal lag determination criteria said?
If that's the case, you may try increasing the lags to 2.
is there video where did you calculate lnpdi as the target variable??
Hi Abdukakhkhor, not sure. You may need to do a search WITHIN my Channel to find out. Thanks.
Hello professor. Thanks for the amazing content. What if the diagnostic test showed that we have serial correlation in residuals? Thank you!
Hi Fred, re-estimate the model at higher-order lags.
Great insight,Thanks!what if there is autocorrelation after estimating long run relationship?how do I correct it?
Thanks Tebogo, for the encouraging words and feedback. Deeply appreciated! Adjust the lag structure and re-estimate.
Dear Professor, thank you for the video. This video has been very helpful and easy to understand. However, I have one question regarding performing Johansen Cointegration rank in STATA. What is the implication if I write additional lag information on the cointegration command, for instance, "vecrank lnpdi lnpce lngdp, lag(4)"? I did both, without (as mentioned on your video) and with lag(4). Estimation without lag gives me no cointegration result, but when I mentioned the lag (which is lag(4)) on "vecrank" command, then the result says at least 1 cointegration.
For your information, I use yearly time series and the reason why I put lag 4 is that because the optimum lag selection give me optimum lag at lag 4 using AIC criterion. Thank you very much.
Hi Anang, you're on the right path. Use the syntax with the additional lag information since that yields a cointegration result.
Using fewer lags causes multicollinearity and autocorrelation. While using more lags reduces the degrees of freedom
James, I explained the dilemma in my video on OPTIMUM LAG SELECTION. You may want to watch the clip. Using lag is not correct science. The final decision rests with the researcher.
Hello professor, in your estimation you used 1 lag as optimal lag. if I use 3 or 4 lags, which lag result should I select as the main result? For example if I use 3 lags in the vecm estimation, the Stata output will show the coefficient and pvalue results for each lag, such as LD, LD1, LD2, etc. which lag’s result should I choose?
There's no confusion on the issue of lag, Ellis. Adapt my procedures, tailor them to yours and interpret your results.
Some of my variables are I(0) and some are I(1), I had to difference them to make them stationary. You are saying, I should make them all I(1), when you say they should be all of the same order? Or can I go foward with I(0) and I(1).
Mark, that is not what I said. You may need to watch the clip again and watch those on ARDL models. I(0) series does not require differencing because it is STATIONARY AT LEVEL. Only nonstationary series should be differenced. Regards.
Hi Professor, I used 7 lags for my VECM because I used daily data. after checking the residuals for serial correlation I found out there are different significance at different lags thus serial correlation at lag 3,4 and 5 and no serial correlation at lag 1,2,6 and 7. How can I interpret this? Thank you
Hi Justin, use the appropriate optimal lags as obtained from the varsoc command to estimate the model....and from there perform the respective diagnostics.
Hi miss, i have a question. Why stata estimate VECM with k-1 lags? Is a default from stata or exist other explication?. Greetings from Ecuador!
Hi Alex, that's the way Stata is configured. My love to Ecuadorians as you share my videos with your students and the academic community. Thanks!
Where you say that the signs have to be reversed in interpretation. Does that also apply to the constant? so in this case it would be a constant value of +4.91949 and not - ??? Thanks for your help btw
Yes, including the constant.
one more question, how to interpret it if the adjustment parameter is 0.8 and positive?
Hi Walid, it impliea adjustment to LR equilibrium is at the speed of 80%. Also, I advise that you get an article on VECM for detailed interpretation.
Thanks so much for your explanatory video. 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..
Hi Melis, as suggested in my video on Johansen Cointegration Test, keep it simple and use 1 CE. But if you are on top of your interpretation then you can use more than 1 CE. Thanks.
Hi there, what is the justification to assume one cointegrating vector even if the Johansen cointegration test yields lets say 3 cointegrating vectors?
No assumption is made as the result tells us you the number of cointegration equations (CE). I maintain that you use ONE CE unless you are good with interpreting results with more than one CE.
@@CrunchEconometrix thank you
dear professor how can i include hetro in the vecm model diagnostic test and also how can figure out the R square while using stata.
Mohd, I don't understand your query. "Include hetro"? "R square"?
Dear Dr. Ngozi
I am running a VECM model and I obtained two long run coeficients with positive values. How can I interprete this.
Hi Jose, kindly watch my videos on results interpretation and adapt to yours. Thanks.
Hi Prof, Just a question on the number of cointegrating equations (rank). Does it mean that if I get 2 or more cointegrating equations in my Johansen test I still use one in my VECM? I will also appreciate if you could provide me with some articles with similar results. Thank you very much.
Hi James, my suggestion and reasons on "keeping it simple" and use one CE is quite clear. You may check online research for likely references...not everything require references. Plausible arguments and reasons often suffice. Thanks.
Thanks Prof.
Good morning, Please running a model that requires a threshold VAR analysis. Please do you have a video for TVAR and TVECM conducted on STATA or EVIEWS? I will appreciate if you can help out
Not at all, Chidi. You may want to check out other online resources. Thanks
Excellent explanation Miss. Just I have a question
Do you have any reference or paper about put one cointegration equation in vecm even if there are more than one cointegration equation?
Thanks in advance. Greetings from Bolivia.
Thanks for the positive feedback, deeply appreciated! No reference. It's only logical to use 1 CE. Much love to the academic community in Bolivia 🇧🇴 please do share the link to my UA-cam Channel with them.
I am following all the procedures you mentioned.But in the last stage while testing autocorrelation,normality and stability stata shows 'error computing temporary var estimates '.
What is the solution of this problem?
Hi Sharmin, I am not familiar with error message so will be unable to guide you appropriately. I advise you post it on Statalist.org for constructive feedback. Thanks.
Thank you for your amazing video, if my selection lag is lag 0 then what should i do, professor? Thank you
Thanks for the encouraging feedback, deeply appreciated! Advisable you use one-period lag since VECM cannot be estimated in STATIC mode. Thanks.
Sir, I m working on VECM. U said we need to reverse the signs while interpretation. Is there any reference of it.. as I need to quote in my phd thesis
Asra, you don't need a reference for that knowing the the Johansen Normalisation result is in IMPLICIT form. If in doubt, look for articles that performed the JN and adapt their interpretation.
How was the error term of 0.067 ECT obtained??
Automatically.
As always, excellent dear Bosede. Please, do you have a video with the ARDL estimation for this very same dataset, I'd like to compare these VECM estimation results to ARDL. Greetings!
Hi Romni, no I don't. But you can still experiment with your own data.
Thank you for all. I have a question a did ardl bounds test and I look that my variables is cointegrate but when I look to my ECM term is -1.56 how can I interpret this number
Your result will be discarded, Jete. ECT of -1.56 is TOO bogus. Re-estimate the model. Change regressors/control variables. Run different model specifications (log-log, log-level)...when the ECT is within 0 and -1, adapt the interpretation given in the video. Thanks.
@@CrunchEconometrix thank you so much ,i will do this thank for the advice
@@jeteshaina6220 U're welcome 😊. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix I'm from mozambique
Hi, your video is very helpful. In my annual time series I experienced the difference in the rank selection based on trace statistics and max statistics (for example, the trace statistics suggests cointegrating rank of 2 while max statistics suggests rank 1). Is that even possible? if yes, then in such case which statistics is better to choose?
And in your example, there seem to be 2 cointegrating equations but you only chose one cointegrating equation while fitting a VECM. What happens with the other one? Can we simply ignore any other cointegrating equations and just estimate the VECM for one cointegrating equation?
Thanks in advance!
Hi Yashoda, watch my video on Johansen cointegration test where I said that the researcher is disposed to using either of the test statistics. I also explained that you keep it simple by using 1 CE unless you are able to explain your results with 2 CEs. May I know from where (location) you are reaching me?
@@CrunchEconometrix Thank you for the quick response and the explanation. I still have couple of confusions (let say questions) which I would be grateful for getting answers on because I really need some quick suggestions on this topic.
Firstly, What does it imply when the ECT is negative and insignificant? and when the ECT is positive insignificant and positive significant? (I came to have all three kinds of coefficients for ECT)
Also, while you performed the diagnostic test for vec stability and got the result " The VECM specification imposes 2 unit moduli." you just said that this is good! how can you say that is good or how to know if the vec is stable?
And I live in Norway now.
@@yashodakarkee8832 Know that any statistically insignificant coefficient need no interpretation because it is equal to zero. A positive and significant ECT implies an explosive model with no longrun convergence. Please I have always reiterated that video tutorials are not enough, they must be supported with reading. Kindly do these and you will get better faster. Thanks for watching my videos. Deeply appreciated! Kindly share my UA-cam Channel link with your students and academic community in Norway 🇳🇴. Thanks 😊
Thanks for the video. BTW, in some articles they didn't reverse the sign when interpreting the vecm long run model coefficient. Is that still correct interpretation? Regards.
Toba, watch my videos again. Signs are not reversed for the VECM (explicit equation) but for the Johansen Normalization Equation (implicit equation). Don't mix it up.
@@CrunchEconometrix sorry disturbing again prof. Yes, I mean Johansen normalization prof. as in the video it refers to the long run model. Thanks in advance.
@@thetruth4712 My explanation clarifies that.
@@CrunchEconometrix Does the reversal of signs for the implicit equation also apply to the coefficient of the constant?
@@denistiyo7193 Yes, it does apply to ALL coefficients in ( ).
Ma'am I have watched your video where you have taught us how to perform a bounds test for cointegration. While concluding the video you have said that if a series is cointegrated then the appropriate model are ARDL and VECM. But here while explaining the VECM, you have mentioned that series should be of same integrated order and we have to first perform the Johanson test. So, how will I run a VECM if I have a integrated order of both level and first difference.
Pls help me in understanding this as it is really confusing.
Thankyou
Hi Vaibhav, I doubt if I ever said this.
HI PROFESSOR, SINCE I GOT 4 CO INTEGRATING EQUATIONS AND 6 VARIABLES, IN THIS CASE, WILL it become possible to estimate eECT1, ECT2, ECT3 ETC? AND ONE MORE QUESTION HOW I DO CAUSALITY TEST IN VECM?
I've always emphasised using 1 co-integrating equation to simplify explanations....and watch my videos on VECM and Causality tests.
Thankyou!!! You’ve been the most helpful econometrics lecturer thus far.
How do you interpret a VECM with 1 lag? The data I’m working on selected a maximum lag of 1, meaning that the VECM essentially ran with 0 lags. The short run estimates showed only the dependent variable for each of the 3 equations e.g under lnpdi there’s only L1 and the constant, no lnpce/lngdp. Likewise, under lnpce there’s only L1 and the constant, no lnpdi/lngdp and so on...
Hi Bupe, thanks for the positive feedback and the kind words...I'm encouraged to do better😊. My suggestion: put 2 in the box for the underlying lags for VAR such that the VECM is estimated with 1 lag as shown in the video. Reason: VECM should not be estimated as a static model (that is, with 0 lag).
CrunchEconometrix thanks once again. I would have messed up my whole dissertation.
I have solved the problem by adding a dummy variable to control for structural breaks and using the LR criterion instead of AIC/BIC.
Cruncheconometrix, Thank you for the video. I have an important question: do we use the differenced values(d1) in our estimation(like testing for cointegration) or we use the original values. I get different results when I use the original values and the differenced values.
Hi Iddriss, my videos are explicit. Follow my procedure.
@@CrunchEconometrix one last quest: if overall the variables are not normally distributed (according to the Jack-berra test), what do you do? Does that nullify your estimation results?
I'll say do nothing (unless you want to change some variables and re-estimate the model). It does not nullify the results except such can't be used for inferences.
@@CrunchEconometrix sorry prof. for disturbing. Is there any technique in stata to boost the error like robust standard error in ols? I think this will help normalize the unnormal error distribution. Regards.
@@thetruth4712 Not sure if the "robust" option is allowed for the VAR routine.
Hello! Your videos are very usefull. And it open me a lot of opportunities for my economics researches. But I have a question with VECM. Is it necessary to perform seasonal adjustment before cointegration checking and estimation? I analize domestic and world wheat prices. And I think that price pairs for some countries are not cointegrated because it needs to first, exclude seasonality.
Thank you in advance
Yes, seasonal adjustments may be corrected for considering your type of data and analytical approach.
Hi, you have written "Note: the sign of the coefficients are reversed in the long run." Do you mean the signs of beta coefficients in the long run? If so where do you base this statement!
Nevila, the long-run equation is in IMPLICIT form. You MUST change the signs during interpretation.
Dear Professor, I really appreciate your videos. They are indeed very helpful! Though, I would like to ask you how can I interpret the short-run elasticities? What do you propose an elasticity expressed in 1% or 10%? Thank you in advance for your response.
Hi Simoni, if the X coeff is 0.435, the elasticity interpretation is given as: a 1% change in X leads to a 0.43% change in Y, on average, ceteris paribus.... May I know from where (location) you are reaching me?
@@CrunchEconometrix Thank you so much! I use logarithms , I forgot to mention that previously. I have one more question and I need your advice. If I would like to interpret the causal short-run effects and my coefficient is significant indicating the existence of short-run causality from the independent vatriable to dependent, then how can I interpret this? I mean an ECT of -0,348 indicates a speed of adjustment of 35% within the first year. In the case of a coefficient that entails short-causality I would say something like that or I have to refer solely to the significance and the existence of shor-run causal effects? Thank you again for your response! I''m a big fun of your videos and I have already watched your videos on causaliy. Greetings from Greece!
@@simonisoursou217 Good. Follow the interpretations I gave in the causality videos. Same thing I would have written here.
Dear Bosede,
I love your videos. They are helping me so much to finish my Bachelor Thesis. Thank you so much for your work
However, I got a question regarding the Interpretation of the ECT. If I have 6 depedent variables ( 6 countries), In order to interpret the ECT of each country i need to run the model every time with that variable on the front? . As the Johansen Normalization Restriction Imposed changes everytime you change the variables
Hi Juan, thanks for the positive feedback on my UA-cam videos. Deeply appreciated! For simplicity, use one cointegration equation as explained in the clip. May I know from where (location) you are reaching me?
@@CrunchEconometrix Yes there seems to be only one cointegrating vector. My question is regarding the explanation of the ECT for each variable. In the clip you establish lnpdi as dependent variable, and it takes the form of 1, and you explain it. But If i want to explain the ECT of lnpce do i need to put this one first?
From the Netherlands. This videos have been my savior
@@juansimonescobar6711 Keep it simple. Concentrate on the outcome variable of interest and arrange it as the 1st variable in the system. Every other thing falls in place. Thanks for watching my videos. Please tell all Netherlanders about my YT Channel!
@@CrunchEconometrix For sure I will tell.
But this is a multivariate time series. I need to give results for every country not only one, that is why i ask.
@@juansimonescobar6711 This video explains time series analysis for a single country.
Dear Dr.
i have a question:
if ETC_(t-1)=1*lPIB_(t-1)+〖0.1214*lX〗_(t-1)-0.9443*lM_(t-1)-3.8345
how can I calculate lPIB in 13:30 lPIB as dependent variable. thanks
Hi Adda, I have no idea. I'm wondering why you need to calculate it since it is the LAG of the dependent variable NOT the dependent variable.
Hi prof,must you difference but a lag variable or you can as well do it at levels.If both which is adviceable to do?
Hi Bertrand, kindly recast your query. Not clear to me. Thanks.
@@CrunchEconometrix oh sorry instead of log i put lag.what i mean is, in attaining stationarity we difference the series .must one difference only the log transformation of the series or it can as well be don on the raw data.secondly please what is the criteria for interpreting diagnostic test like langrange multiplier,jacque berra and stability condition.You gave no criteria for their interpretation.Thanks in advance and greetings from cameroon.
@@enongenebetrand1119 Difference the log if you are estimating a log-log model. I always explain diagnostics using the pvalues.
Hi, In my long run equation one of my variables is not significant should i drop it or just still include it ?
Evelyn, you must understand that all coefficients MAY not be statistically significant.
Thank you so much for these amazing videos they have literally taken me through my thesis For some more clarity why do we reverse the signs on interpretation of the long run effects
Hi Evelyn, I explained this in the Johansen Cointegration test results. They are in implicit form.
hi , what's mean a positive impact ? is that mean when pce increase pdi increase and when pce decrease pdi decrease too ?
Hi Ahmed, it means the former.
Thank you for your amazing videos...I have a question...I have performed all the necessary steps, I get an AIC optimal lag of 1 and the number of Cointegrating equations to also be one. When I perform the VECM model, the L1._ce1 for my dependant variable is negative but insignificant...What does this mean and what do I need to do to correct this. Adding more lags does not solve the issue
Hi Rahul, I'm unclear about your query. If the error correction term is not significant it implies there's ZERO reversion to long-run equilibrium. Thanks
@@CrunchEconometrix So what does this mean for my model? Is it not specified correctly? Like I said I get 1 Cointegrating equation at 1 optimal Lag. But when I run the VECM, the error correction term is negative and insignificant. What is my next steps?
Rahul, I have already laid out the steps for VECM. Once you have your results, go ahead and interpret.
@@CrunchEconometrix Having read the literature it states that your error correction term must be negative and significant for the VECM model to be appropriate. Since mine is not…do I have to change my model or can I just begin interpreting?
Rahul, you can re-estimate after you have changed your variables and/or modified the lag lengths. It is not erroneous if the ECT turned out to be not statistically significant. It is subject to your study and your result is what it is. Hope these tips are helpful.
Dear sir, I have a question about the vec outcome tables. If the error correction terms in the first table (_ce L1.) are all positive and not statistically significant, I should interpret it as there are divergence to the long term equilibrium in all equations, right? This result does not affect my interpretation for the second table (explained in this video) of long run causal relationships, is that correct? Thank you so much for answering my question. I am working on my thesis and you have been so helpful. I am very grateful for your videos and your answers.
Yes, Yanuo. You may also refer to how I interpreted the VECM results.
@@CrunchEconometrix Thank you!
It was very helpful ma’am. I have a short query - how do I understand whether my trend will be constant or not? And why did you select 2 lags here?
Thanks Afia, for the positive feedback. Deeply appreciated! Lags are selected as shown in the preceeding videos. If the trend coefficient is not significant, then exclude it.
Dear Sir, can I use VECM for my series which are all stationary at I(0)?
No, Trung. Use OLS
@@CrunchEconometrixAnd there's no need to test the cointegration among them, right sir? Because I want to analyze the market efficiency through testing the cointegration. Thank you
Trung, I will advise you to read the fundamentals of COINTEGRATION from any econometrics textbook to deepen your understanding.
Good work ma'am. I still don't get where you got the values in the VECM model: 0.002, -0.144, +0.44, -0.099 and -0.067. Please shed more light on this. Thanks.
Ok, if you check the 1st component of the result output, you'll see the coefficients. I only lifted from there to construct the lnpdi equation.
hi , what about heteroscedasticity test what is the command ?
Hi Ahmed, you may need to check other online resources for that. Thanks.