TH-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.
I am really thankful to you. Tomorrow, I have to given a presentation in zoom about inflation and economic growth in Bangladesh.I use all the econometric method in eviews and your videos are really helpful for me.your interpretation's are very helpful. I have several papers I want to work with you .
Great 👍...let me know what you have in mind but I have strict rules about collaboration (based on past experiences). If you have a Scopus publication send to cruncheconometrix@gmail.com for evaluation...that's one of my conditions. Thanks and take care.
@@CrunchEconometrix I am from Milan, Italy. And I did my studies in Frankfurt, Germany. It's a just while I got the passion for econometrics, so cool and useful.
@@SamWall93 Awesome!!! Kindly share my YT Channel link with your colleagues and academic community in Italy and Germany. They will find the content on my Channel very helpful...thanks!
Thanks for the positive feedback on my video, deeply appreciated! 💕 Kindly share my Channel link with your students and academic networks. May I know where you are reaching me from?
Respected Madam, I have Learnt a lot from your valuable videos specially in VAR environment. I am preparing electricity demand forecast using conventional multiple regression analysis since long. Now I am shifting to VAR model, I have gone through all steps 1. lag selection criteria for each variable 2. unit root test for each variable 3. found I (1) Cointegration 4. Combined lag selection criteria found to be I (1). 5. Johansen cointegration showed one integrating equation with using 1 Lag 6. Then I ran VECM with 1 lag 7. Speed of adjustment come negative and significant 8. No serial correlation, no heteroskedasticity, normality test is ok 9. It also passed cusm stability test My VECM model is statistically ok. In Johansen Equation, we interpret it by changing the sign, while after changing the sign results are according to econometric relationships e.g. if price is increased, the consumption will decrease. My question is that in the VECM Long Term equation. the results are not coming according to econometric theory i.e. Price elasticity is not negative. so, is it ok? But in short Term Model in VECM the electricity elasticity is coming to negative. My next question is, now i have computed both Long run and short run equation from the VECM model and, I can compute the residuals ECT t-1 from the long-term equation and also, I know the coefficient of ECT i.e. speed of adjustment My data was from 1985 to 2019 and my independent variables are total GDP, real price of domestic category electricity and population of Pakistan and my target variable is domestic electricity consumption. I have taken all these variables in logarithmic form. now my request is how I can get the forecast of my target variables for my next 10 years while I can project the independent variables like GDP, price and population for next 10 years. but how can I make projection of ECT t-1 for next 10 years by transforming this short run equation in a form that I can easily calculate forecast of my target variables. currently i am facing the issue that in equation my target variable is in differenced form which is the form of VECM equation. In OLS regression analysis after finding out the elasticity of the best equation we can easily transform this equation into Electricity Sale = (1 + Growth Rate GDP) GDP elasticity x (1 + Growth Rate Price) Price elasticity x (1 + Growth Rate Lag sale) Lag elasticity Is there some method to transform VECM equation like the above-mentioned equation to compute the forecast? Can you tell me if there is any option in EVIEWS that it can give us forecast for the period beyond the data if we provide to EVIEW the projection of independent variables? Normally this projection work doing in excel. Is it possible can you make a video on VECM forecast for next 10 years to understand the concept? Best regards
Thank you so much professor, amazing explaination. You have made things so easy for me. Please continue posting videos on empirical econometrics, you are too good. Cheers and godspeed
SuperReddevil23 I will interpret it that to mean the evidence of a long-run equilibrium is not significant. You can also do a little Google search on it.
thank you so much! your way of explaining has been such a great help. for my cointegration test- trace stat indicates 1 and max indicates 0. following the trace stat result i went ahead and estimated a VECM on lag 1 (i.e.2-1) . while cointegrating eq coefficients are statistically significant, i find ECT and all 6 variables' coefficients are statistically insignificant in the error correction estimation. 1.how to interoperate this? 2. do i go ahead and alternatively estimate unrestricted VAR instead (according to max stat result of no cointegration).
Thank you very much for the class, please how do we get the durbin watson and p-value for each variables on the e-view? Because authors provide these info in the data presentation.
Thanks for your encouraging feedback...deeply appreciated. EViews displays the Durbin Watson statistic in most its results...as for p-values, you may have to use the t-stat to determine the statistical significance of the coefficient.
Hello m'am. One question: In the short-run part, the CointEq1 of the Error Correction, If I have a value of 0.0783 where you have -0.067 in min 02:00, is this correct? I read that the value should be between -1 and 0 but I don't know if this is correct. The long-run coefficients are significative, but this 0.0783 it's not. This 0.0783 is placed where you have the -0.067 in the short-run equation. Thank you very much m'am. Best regards!!!
Great tutorial on VECM. Only one query, the model is showing that in the long run, a 1% change in GDP is associated with 0.099% decrease in PDI. While statistically, the model is appropriate. However, in economics, a rise in GDP in the long run or the short run should result in an increase in personal disposable income (PDI), rather than a decrease in PDI. Can you please comment on this. It will help a lot.
Hi Sher, results are a function of several factors: variables, scope, and technique used. Besides, there are tons of empirical findings that contradict a priori expectations. Yours could be one of such.
Great videos on VECM. However, there is a clarification I seek. In economics, whether in the short run or the long run an increase in GDP should lead to a rise in PDI (personal disposable income). Whereas, this VECM model is showing a negative relatiopship i.e. a rise in GDP results in a decrease in PDI. Could you please explain this? Or was your attempt more to show the econometrric/statistics side of VECM rather than economic.
Thank You for your reply Now my specific questions are as follows Q # 1 If in Johanson test lag length criteria tells us lag = p= 1 . So for VECM as you said in your vedio take lag as ( p-1). What should I do Q. # 2 I have computed the VECM conventional equation with 4 variables like electricity sales, GDP, electricity real price, population, which is also pasted below. Kindly guide me with a video or with some other means, how I can transform this equation to compute the forecast of the electricity sales. The target variable is LDOM8519 electricity sales which is in differenced form. ∆ LDOM8519t = 0.611109 + 0.185012 ∆ LDOM8519t-1 + 0.5014 ∆ LGDPTOT8519(t-1) - 0.014722 ∆LDOMRP8519t-1 -24.51419 ∆ LPOP8519t-1 - 0.48071 ect(t-1 )
Hi Bilal, my responses: Q1: In this case, use 1 lag since VECM cannot be estimated as a static model. Q2: My videos on "forecast error variance decomposition" will be helpful. Thanks.
Thank for this inestimable lecture it is very precise and timely. If the long run relationship is to be discussed upon, is it the normalized Johansen cointegration that will be adopted or the lag ECT cointegration equation?
How do you proceed when you obtain non-uniform results during johansen cointegration test i.e you find that there is no cointegration under trace stats but there is cointegration under max eigen value? will you estimate a VAR or VECM?
Hi Dennis, I mentioned in my video on Johansen Cointegration Test that researchers take decision when the results conflict which guides subsequent approach.
Great video. Loved it. Got one question just to be sure. You described the lambda (speed of adjustment) and its meaning, but no mention as to its sign. But then noticed that in your example it was negative. Must it be negative to have some meaning? What happens when it's positive?
Hi Douglas, I appreciate the positive feedback. Deeply appreciated. You skipped the prerequisite videos that contained the information you seek. Kindly watch them. But note that theoretically the sign must be negative. if positive it implies there is no convergence to long-run equilibrium and the model becomes explosive. Thanks.
Thank for your video 😍 I have a question. If there is serial correlation or the residuals are not normally distributed, what would we do? Can we do anything to fix the model?
I watched most of the videos from CrnchEconometrix and found very much useful to the researchers in secondary data. Now i am interested in structural VAR estimation. I request you to make one video about i, if possible.
I have a question, Suppose if we get the no. of lags for cointegration as 1, then what are we going to use as the no. of lags for VECM, we can't use 0 (since we get an error) isn't it madam? P.s. I love your videos, they're very beginner-friendly and clear also they're really helpful for my analyses
Hello! Thanks for videos... it's very helpful for me. I am doing panel vecm right now and my errors are non-normality distributed. Please help me to correct that non-normality. Thanks!
Hi William, as explained in my VECM and Johansen cointegration test videos, keep it simple by using 1 CE except you are on top of your interpretation then you can use more than 1 CE....I still use only 1 CE regardless of if the JCT yields more than 1.
I have created d(ln_variable) for transformation purposes and further on used d(ln_variables). But should cointegration test be done with d(ln_variables) or level variables?
Hello CrunchEconometrix. I appreciate a lot your contribution with these videos, they amazing. Thank you! Please can you help me with a question: Why ECT=-0.0670 is interpreted as "an adjustment speed of 6.7%", but for the short-run effect of PCE=0.44 you say "a percent change in PCE is associated with a 0.44 percent increase in PDI"?. Why using a different rule for interpreting long-run and short-run coefficients, or was it a small lapsus? Thank you again for your great videos!
Hi Sabrososh, thanks for the encouraging feedback. Deeply appreciated! Those are the generic interpretations. Look up the the references listed at the end of the video or my papers listed on the Community Page. Thanks
thank you for your great explanations about VAR and VECM. I have a problem with heteroskedasticity in residuals of my VAR and VEC model residuals, could you please guide me on how to deal with this problem?
and according to your responses to other people, you suggested choosing higher lags to remove heteroskedasticity, therefore this is a violation of choosing suitable lag regarding information criteria
Hi Kaveh, thanks for the encouraging feedback on my videos. Deeply appreciated! It appears that the choice of your variables does not respond to heteroscedasticity. You may request to change them to closer proxies and re-estimate.
Thank you Dr. for this video. Pls how do I get to know the P-values of the coefficients since what is provided is the standard errors and t-statistics? thanks
Alright, Kenechukwu. You made me go into EViews10 at 2.30am. To obtain pvalues from VAR output: Go to ->>Proc, Make System, Order by variable, Estimate, 🆗.
Hello My Sister! Thumbs up!! you are making us proud here. I used this video as a specimen to estimate my vecm model and arrived at the same conclusion with you but I still need clarification concerning the issue of normality. You rightly said there that while some of the variables are normally distributed others are not and in general, the model is not normally distributed but seems you were a bit relief with the outcome of the heteroscedacity . Please, what should I do as regards the normality issue? should I just leave it or correct?
@@CrunchEconometrix Hi, Can you please give me some references for what you just mentioned so that I can mention in my research paper, I have a VECM which is free of autocorrelation and heteroskedasticity but two equations are not normally distributed. Thanks.
Hello, thank you for the videos. Are we not supposed to change the signs for the long-run model? So should it not be (+) 3.63? Or do we only do that in the Johansen part? I was told to flip the signs in the long run part.
hi professor, thank you for the great video, it really helped, ill be sure to share your channel. I have a question regarding the optimal lag for the VECM model. my optimal lag is 7. does this mean for the VECM my optimal lag would be 6 (p - 1). If so which short-run dependent variable lag should I use. aka d(GDP(1)). D(GDP(2), D(GDP(3) etc..
Hi Emira, from the output the t-stats are in [ ]. On how to get that from the t-table, I will advise you read up HYPOTHESIS TESTING from any econometrics textbook for detailed guidance. Thanks.
Thanks Zama, estimate VAR when JCT shows no cointegration. Click on the Playlists to watch my videos on the procedure. May I know from where (location) you are reaching me?
4:27 I don't understand this part. the coefficients have low t-stats, so how am I allowed to put those coefficients into the equation when they're not significant?
@@CrunchEconometrix I do but I have not found one that says its OK that the error correction in 4.27 is insignificant. I need to back this up somehow so I would love a suggestion on one that says that we must include the error term irrespective of its significance!
Thank you very much for the videos!! Is it a problem because in my model, the probabilities are less than 5% level so there is serial correlation and on normality all components showed a high prob value hence are distributed normally however the total shows that residuals are not normally distributed. And on heteroskedasticity the prob value is 0.44 so the model is not heteroskedastic. So what should i do or its fine like that?
@@CrunchEconometrix Thank you, much appreciated!! I used 6 lags for serial correlation only and starting from the 2nd lag up to the 6th, the model is now showing that there is No serial correlation. However, i wanted to know is it not a problem because from Var lag order selection criteria i was using 2 lags but i now change them to 6 lags on residual tests
@@CrunchEconometrixThank you very much for your answer. In the mean time, I've already watched almost all your videos and understood that. I'm from Portugal. Thank you
coefficient values shows e (t-1) values in the running model as you hove shown after performing VECM ?. If it is significant and negative it means there is long run relationship exist.
Hello maam I am here again. Thanks for the video again and I really thankfull, but I have another question. How to interpret the ECT Coefficient that has (+) positive sign? So, I have a significant ECT coefficient at 5% and the coefficient is 0.534676. Is the way to interpret it different compare to ECT coefficient with negative sign?
@@CrunchEconometrix thanks for the answer mam. You are so humble and answering each of the question here in comment section, I already transform my variabel into log natural now it shows there is one cointegrating equation, but from the VECM it shows that the ECT not significant from t statistics also p-value and the ECT has negative sign. So should I interpret like " there is long equilibrium between variabel but not statistically significant?"..Thank you again mam 🙏🏼
@@CrunchEconometrix thank you so much mam, really appriciate it, I am working on my undergraduate thesis..thanks a lot for all of your video about this method 🙏🏼💙
Thank you very much for your lessons! It is incredible and very helpful for me! Also, I would very much like to clarify a few points. Please tell me what we should do if after Lag Length Criteria and Lag Exclusion Test my lag intervals are 1 1 4 4 7 7? How exactly can I reduce these intervals by 1 lag for the VECM? In this case, I would also like to ask what exactly the number of lags I should use for the LM-test (if lag intervals are 1 1 4 4 7 7)? Thank you very much in advance for your answer!
Thank u so much as u have explained the concept very well. I have a query that my vecm is not fitted in residual diagnosis. There is no autocorrelation but rest two conditions are not met. What can I do ? Kindly help me out.
Thank you for your explanation, I really appreciate it! Further, I want to ask, what if the Cointegration Equation is not significant? Does it mean that both in the long run and short run/only short-run have no relationship between the dependent and independent variables?
Hi Prof...thanks for this video...so many doubts of mine have been cleared and the way you demonstrated the interpretations was outstanding. I have one question - how do you decide on a target variable, when you have many variables in the Johanssen method? Should we go for a causality to determine the target variable? Look forward to hear from you. I am based in India...
Thanks for the positive feedback, Jyoti. Deeply appreciated! The target variable is the 1st variable listed in the VAR system. It is the variable upon which the JCT is normalized. Hope this helps. Kindly share my TH-cam Channel link with your friends and academic community in India 🇮🇳. They will find the content helpful too 😊.
Thank you so much ma for a good presentation. In a case where there are two cointegrating equations (that is none and at most 1 are asterisked) , do we interpret the D(...)(-1) or the D(...)(-2) of each variables for the VEC estimates? Also following the video preceding this one, in a case where we have the variables stationery at first difference but different max-lag lengths for the unit root test, is it still acceptable to use or not acceptable to use if the are different unit root lag lengths for the variables?
Hi Praise, thanks for the encouraging feedback. Deeply appreciated! Follow the guide and explanations. Use 1 cointegrating equation. Same lag length is used for all variables.
1. In the long run part of ECT model, the coefficients of regressors must be interpreted in reverse signs. 2. When we run VECM practically, reduce one lag for D(Endogenous).
Hello, Thank you for all your uself videos. Can you please clarify if you use a Normal distribution to assess the variables significance and if you use a two tail test?
Hello! I have been following your videos and I want to thank you for the help they have given me in my dissertation. One question: How do you resolve heteroscedasticity if it arises on White's test?
@@CrunchEconometrix Thanks for your feedback! By the way, since the dummy variables are exogenous, how can we measure their impulse/response function? Thanks!
Dear ADELEYE, would you please answer these two queries? (1) In my model all variables (both IDV & DV) are cointegrated at I(1). Thus I used VECM. While running varsoc , lag one [1] is suggested by all criteria. Now, while checking cointegration, what lag should I use? one or two? And what lag should I use in estimating vecm? (2) In my other model, cointegration exists but while running ecm error correction term is found insignificant, in this situation what should I do?
Hi Chinta, for (1) I gave clear explanations. Kindly watch my VECM videos again. For (2), simply interpret your results. There is a long-run relationship (cointegration) but reversion to long-run equilibrium (EECM) is not statistically significant. Thanks.
Hi Arup, NORMALITY TEST under VAR-VECM can be overlooked because it's SYSTEM of equations. Focus more on the model passing both HETEROSCEDASTICITY and AUTOCORRELATION tests.
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.
Hi Jacky, the outcomes we get have a lot to do with the variables used for analysis not necessarily because the econometric procedure is wrong....assuming things are done properly.
Thank you very much for your videos. Why did you interpret the ECT coefficient term from the VECM differently (as a log-log rather than as a log, I guess) from the other coefficients? It was a lapsus-linguae?
Thank you for your good work. Please, I have a time series, all are I(1) and cointegrated but my optimal lag is 1, can I still go on and run VECM given p-1 condition
Hi Dayo, this is not a problem since VECM cannot be a static model but must contain lags of the regressors. All you have to do is to maintain the 1 lag and put a footnote to explain why....and thanks for the kind words on my TH-cam Channel. Humbly appreciated!
Reza, you have muddled up this query. You can't do VECM with level-stationary variables but Granger causality can be done regardless of the the level of integration of the series.
Mam if we found 3 cointegrating equations, then while running vecm, should we enter 3 or 1 in rank option? If i put one, will it make any difference. Or if i put 3 then how to interpret three cointegrating equations?
Yes When I changed the lag order to 2 autocorrelation was not present.However both normality and heteroskedasticity did not agree.Does this mean the model is insignificant?
@@annalexander6384 Try a log-log model and increase lag to 4 if you have enough observations. Don't worry too much about "normality" it is rare to obtain a not significant pvalue due to the different equations in the system.
Dear teacher, is there a heteroskedasticity test like that in Stata as well? Also, you have chosen the (No-Cross Terms) option in your test. There is the (With cross terms) option too. Would you kindly explain when should we use each of those two options?
hello profesor, i have monthly data for 14 years. in Johansen test I have at most 2 cointegration. when i run with the optimized lag, i have a lot of coeficients
Hi, really appreciate this video! How do we know which of the variables are cointegrated with the dependent variable? or am I missing/misunderstanding something?
Hi Mouanguissa, you can explore a combination of measures: use logs, estimate with the "robust" option, change regressors etc. May I know from where (location) you are reaching me?
I have two integrated variables of order 1 and two integrated variables of order 0 and I used the ARDL model. please how can i use causality between these variables
CrunchEconometrix so in this case I should interpret that like this or may I change my model . But after researching on other plateforme I found that it is possible if there’s a structural change or break is this right .
There is no right or wrong. The result is what it is. Surf the web and you will find papers with positive ECT. You may also decide to re-model by changing or removing some regressors.
Mam I have 5 variables and 4 cointegrating equations. I have taken 2 exogenous variables too. Mam can I proceed with the model and how to interpret 4 cointegrated equations
Dear Ma'am, as in your ECT equation the constant term is negative and you are saying it should be negative. But in ECT equation it is positive. My question is if the constant term in ECT equation is positive, is it a good sign in estimation or not, or there is any problem in my estimations ?
@@CrunchEconometrix Dear Professor, Thank you for the prompt response. Question: How can I Interpret my results?if 1) ARDL-ECM (OLS estimations) results are insignificant. 2) ARDL-bounds test: cointegration Exist Appreciate! Stay blessed!
Is the interpretation of the ECT coefficients in percentage because of the logs you take (lndpi, lnpce etc..)? Or would this be a percentage for normal variables as well?
@@CrunchEconometrix At 4:30 you say that 'the previous years deviations from long-run equilibrium is corrected at a speed of 6.7%.' If it wouldn't be a log variable but a regular variable, would it just be 'corrected with a value of -0.067'? Thank you in advance
Max, you need to read to be sure of the information you need. You may not get a YES or NO answer all the time. I've done my best to guide you but you aren't paying attention. So, my honest and candid advice is that you read. I've made that clear in almost all my clips: video tutorials are not substitutes for reading. Please READ to strengthen your academic confidence.
After doing the VECM test I ran the diagnostic to see if there was heteroskedasticity and the residuals were found to be heteroskedastic. How can I correct it?
hi , you said "a percentage change in pce is associated with 0.44 % increase in pdi on average ceteris paribus in the short run" but what you mean with change in pci increase or decrease ??
I have a problem with my regression. I'm using Eviews as the econometric software and I'm running Vector Error Correction Model (VECM). I have a problem of heteroskedasticity and autocorrelation. I don't know how to correct them in Eviews using VECM.
I have some issues about this model.. 1.Is it necessary to include the ECM equation (raw equation,not computed one) in my article ? 2. Though my speed of adjustment is negative but it is statistically insignificant ( p value is higher than .05) ,is it a matter of concern? 3.Relevance of Macro Economic factors for the Indian Stock Market by Aman Srivastava (2010) order the equation in another way, was it wrong?
Hi I am watching your videos from Swaziland. In your Johansen normalization test we changed the signs and the impact of LPCE was positive on LPDI and GDP had a negative impact in the long run. However, in the VECM results we kept the signs the same way when interpreting for the long run model. LPCE had a negative impact. Is there such a thing as normalizing VECM results as well?
CrunchEconometrix Im sorry for asking again , because long run cointegration equation is the same as normalized JJ right ? In ur other video, normalized JJ is interpreted by reversing their sign. So in VECM we don’t have to do that?
That is because the Cointegrating Equation (CE) is in IMPLICIT form. That is, CE = (Y - X1 - X2 - X3). Therefore, to know the impact of the regressors on the depvar the equation must be stated EXPLICITLY which is the reason for INVERTING the signs of the coefficients.
Hello Prof, many thanks for the youtube lectures. I have a question please: my step 1 and 3 are all met but i have two integrating equations after doing step 3. I am a bit confused, shouldn't i have one cointegrating equation? many times i see one cointegrating equation. what do i do next? below text is copied from the table in step 3. Trace test indicates 2 cointegrating eqn(s) at the 0.05 level * denotes rejection of the hypothesis at the 0.05 level **MacKinnon-Haug-Michelis (1999) p-values
Ma, in the preceding video, you did not difference the variables before running the Johansen test, after they proved to be I(1). May I please know why? Thank you
Hi Reine, JCT is performed on the variables in their natural and not transformed form. Refer to any econometric textbook on the fundamentals of cointegration. May I know from where (location) you are reaching me?
@@CrunchEconometrix please at what point can we say fstatistic is significant or insignificant in VECM? AND can you please help me make video on only the interpretation of Structural VAR? THANK YOU.
The pvalue of the F-statistic will point you to the right direction. You can easily pick up publications that use structural VAR and read up on the interpretation. Thanks
In STATA 13 episode of estimating and interpreting the VECM you said we should interpret the direct of influence opposite to what we obtain in the software (that is if lnpdi comes out as negative we should interpret as positive). Please why is it that over here, in interpreting the results, you did so just as the signs are and never interpreted them at the opposite direction?
Enoch, the reason is because the long-run equation is expressed in linear form. Please watch my video on Johansen Cointegration technique for deeper understanding. Thanks.
@@CrunchEconometrix i have watched the video over again and i cant seem to find the part where you interpret the long run coefficients. you just state that this is the equation that signifies long run cointergration
Hello professor, may I just ask one more thing regarding your model? is the ECT also the same for all target variables? Like, for example, is if i choose D(LNPCE) instead as a target variable?
Hi Zama, I will say it does because you have a priori expectations for the signs and significance of the coefficients. May I know from where (location) you are reaching me?
Hello mam, after running VECM, the diagnostic tests show auto correlation, heteroskedasticity and no normality. what to interpret now? and if the interpretation says given VECM model is invalid, then what to do next?
Hi Boni, re-estimate the model by playing around with lags...or change regressors if you have it. You have to try several solutions to fix the problems.
Two issues I faced: first, the optimal lag to choose came to 1. So, I used 1 lag for cointegration; how many lag shall I choose for estimating VECM. Another issue, I am getting 1 cointegrating equations under trace and 2 cointegrating equations under maximum eigenvalue . Please help me out
Hi Pravakar, I'll suggest you maintain 1 lag for VECM since you can't estimate with 0 lags (that'll make it a static model and not autoregressive). Also, I have always suggested keeping things simple by using 1 cointegrating equation unless you are on top of interpreting your results which can become a little knotty when using 2 cointegrating equations.
@@pravakarneupane7299 Well, sometimes it's not absurd if it can supported by economic theory. But if not, you can look for better proxies for the variables and re-run your model.
@@CrunchEconometrix Hello Ma'am, I too face the same issue of having 1 lag as optimal through my VAR estimation. So as per theory, I need to estimate my VECM as (1-1 =0) lag, which is not possible. So as you suggest I can carry my VECM estimation with 1 lag. Secondly, my trace test indicates 2 co-integrating equation whereas maximum eigenvalue suggests none. Is this wrong or how to interpret this? Kindly suggest!
@@saakshijha9689 As explained in my Johansen cointegration test video, you are disposed to using EITHER of the results this is because oftentimes the two tests yield similar results and sometimes they don't.
Hi. Thank you so much for your assistance. I have been following your tutorial with no problem until I got to the part where I need to run the White Heteroskedasticity Test. Eviews is not responding and just gives me an error message saying: Positive or non-negative argument to function expected. Then the output I get from Eviews is gibberish in a way. Would you be able to advise what to do? I notice that when I run it with optimal lags of 1 (even though my optimal lags are 6 per the AIC) I don't have this problem.
@@Kndzu Awesome, Khanya! Please spread the word about my TH-cam Channel to your friends, students and colleagues in South Africa for awareness. Thanks 😊.
Hi Bosede. I am in need of your assistance again. I am running a VECM for my study and it appears the data has many outliers. To deal with this, my plan is to add dummy variables. Would you mind advising if there is any best practice on how to decide where to add dummy variables?
TH-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.
Looking forwards to your video on how to posit restrictions in VECM modelling.
Thank you very much. So far, this is one of the best video on ADF, cointegration and VECM I have watched.
Glad it was helpful, Kojo!
your videos are great, you explain everything very clearly and with a good mix of theory and how to do it in practice in Eviews. thank you very much!
U're welcome, Gab! 💕 😊
I am really thankful to you. Tomorrow, I have to given a presentation in zoom about inflation and economic growth in Bangladesh.I use all the econometric method in eviews and your videos are really helpful for me.your interpretation's are very helpful. I have several papers I want to work with you .
Great 👍...let me know what you have in mind but I have strict rules about collaboration (based on past experiences). If you have a Scopus publication send to cruncheconometrix@gmail.com for evaluation...that's one of my conditions. Thanks and take care.
Congratulations, great channel! Thank you for being so clear! You have my support.
Wow! Thanks for the positive feedback and support for my content. Deeply appreciated! May I know from where (location) you are reaching me?
@@CrunchEconometrix I am from Milan, Italy. And I did my studies in Frankfurt, Germany. It's a just while I got the passion for econometrics, so cool and useful.
@@SamWall93 Awesome!!! Kindly share my YT Channel link with your colleagues and academic community in Italy and Germany. They will find the content on my Channel very helpful...thanks!
Thank you for this clear presentation. Currently writing my undergrad dissertation and it is very helpful
Thanks for the positive feedback on my video, deeply appreciated! 💕 Kindly share my Channel link with your students and academic networks. May I know where you are reaching me from?
@@CrunchEconometrix I'm from England
Respected Madam, I have Learnt a lot from your valuable
videos specially in VAR environment. I am preparing electricity demand forecast
using conventional multiple regression analysis since long. Now I am shifting
to VAR model, I have gone through all steps
1.
lag selection criteria for each variable
2.
unit root test for each variable
3.
found I (1) Cointegration
4.
Combined lag selection criteria found to be I (1).
5.
Johansen cointegration showed one integrating
equation with using 1 Lag
6.
Then I ran VECM with 1 lag
7.
Speed of adjustment come negative and
significant
8.
No serial correlation, no heteroskedasticity,
normality test is ok
9.
It also passed cusm stability test
My VECM model is statistically ok.
In Johansen Equation, we interpret it by changing the
sign, while after changing the sign results are according to econometric
relationships e.g. if price is increased, the consumption will decrease. My
question is that in the VECM Long Term equation. the results are not coming
according to econometric theory i.e. Price elasticity is not negative. so, is
it ok? But in short Term Model in VECM the electricity elasticity is coming to
negative.
My next question is, now i have
computed both Long run and short run equation from the VECM model and, I can
compute the residuals ECT t-1 from the long-term equation and also,
I know the coefficient of ECT i.e. speed of adjustment
My data was from 1985 to 2019 and
my independent variables are total GDP, real price of domestic category
electricity and population of Pakistan and my target variable is domestic
electricity consumption. I have taken
all these variables in logarithmic form.
now my request is how I can get the forecast of my target variables for
my next 10 years while I can project the independent variables like GDP, price
and population for next 10 years. but how can I make projection of ECT t-1
for next 10 years by transforming this short run equation in a form that
I can easily calculate forecast of my target variables. currently i am facing the
issue that in equation my target variable is in differenced form which is the
form of VECM equation.
In OLS regression analysis after
finding out the elasticity of the best equation we can easily transform this
equation into
Electricity Sale = (1 + Growth Rate GDP) GDP
elasticity x (1 + Growth Rate Price) Price
elasticity x (1 + Growth Rate Lag
sale) Lag elasticity
Is there some method to transform
VECM equation like the above-mentioned equation to compute the forecast?
Can you tell me if
there is any option in EVIEWS that it can give us forecast for the period
beyond the data if we provide to EVIEW the projection of independent variables?
Normally this projection work doing in excel.
Is it possible can
you make a video on VECM forecast for next 10 years to understand the concept?
Best regards
Hi Bilal, your query is TOO LONG. Kindly go straight to the point. Thanks.
Thank you so much professor, amazing explaination. You have made things so easy for me. Please continue posting videos on empirical econometrics, you are too good. Cheers and godspeed
SuperReddevil23 Hahahaha, u're welcome SuperReddev (but I'm a Super bright RedGunner😊)...and please tell others about my TH-cam channel🥂
Will surely do. One other question i wanted to ask you is that what is the implication of a statistically insignificant error correction term ?
SuperReddevil23 I will interpret it that to mean the evidence of a long-run equilibrium is not significant. You can also do a little Google search on it.
ok thanks
oH my god!! I was in a real trouble in making a research assignment. You helped me so much, thank you.
You're very welcome, Tina!
thank you so much! your way of explaining has been such a great help.
for my cointegration test- trace stat indicates 1 and max indicates 0. following the trace stat result i went ahead and estimated a VECM on lag 1 (i.e.2-1) .
while cointegrating eq coefficients are statistically significant, i find ECT and all 6 variables' coefficients are statistically insignificant in the error correction estimation.
1.how to interoperate this?
2. do i go ahead and alternatively estimate unrestricted VAR instead (according to max stat result of no cointegration).
Hi there, thanks for the encouraging words and feedback. Deeply appreciated! Try option 2 and see what the outcome is.
Thank you very much for the class, please how do we get the durbin watson and p-value for each variables on the e-view?
Because authors provide these info in the data presentation.
Thanks for your encouraging feedback...deeply appreciated. EViews displays the Durbin Watson statistic in most its results...as for p-values, you may have to use the t-stat to determine the statistical significance of the coefficient.
your videos are very helpful ...thanks
Glad you like them, Preeti...thanks!
Thank you so much for the videos!!! Extremely helpful!
Thanks for the positive feedback, Huiru. Deeply appreciated! 💕 May I know from where (location) you are reaching me?
@@CrunchEconometrix Arkansas :) I'm a phd student of University of Arkansas.
@@huiruchen5664 Good to hear, Huiru! 💕 Kindly spread the word about my videos to your friends, students and academic community in Arkansas. Thanks! 😊
@@CrunchEconometrix Will do! Thanks again for your videos!
Hello m'am. One question: In the short-run part, the CointEq1 of the Error Correction, If I have a value of 0.0783 where you have -0.067 in min 02:00, is this correct? I read that the value should be between -1 and 0 but I don't know if this is correct. The long-run coefficients are significative, but this 0.0783 it's not.
This 0.0783 is placed where you have the -0.067 in the short-run equation.
Thank you very much m'am.
Best regards!!!
Thank you very much... from Namibia
My pleasure, Glenn! Thanks for the encouraging feedback, deeply appreciated!
Great tutorial on VECM. Only one query, the model is showing that in the long run, a 1% change in GDP is associated with 0.099% decrease in PDI. While statistically, the model is appropriate. However, in economics, a rise in GDP in the long run or the short run should result in an increase in personal disposable income (PDI), rather than a decrease in PDI. Can you please comment on this. It will help a lot.
Hi Sher, results are a function of several factors: variables, scope, and technique used. Besides, there are tons of empirical findings that contradict a priori expectations. Yours could be one of such.
Great videos on VECM. However, there is a clarification I seek. In economics, whether in the short run or the long run an increase in GDP should lead to a rise in PDI (personal disposable income). Whereas, this VECM model is showing a negative relatiopship i.e. a rise in GDP results in a decrease in PDI. Could you please explain this? Or was your attempt more to show the econometrric/statistics side of VECM rather than economic.
Hi Sher, I have responded to you on this on a another thread.
U r perfect at explaining everything... U made my day
Glad to hear that, Deepty!
Dear Madam, thanks for your effort and explanation, my question is can i use pedroni and kao for panel data instead of using johansen ???
Yes, you can. Johansen is NOT applicable to panel data. Kindly watch my videos on panel ARDL. Thanks.
Really helpful for my thesis....thank you sooo much
You are welcome, Sadaf! 🙏 ❤️
Thank You for your reply
Now my specific questions are as follows
Q # 1
If in Johanson test lag length criteria tells us lag = p= 1 . So for VECM as you said in your vedio take lag as ( p-1). What should I do
Q. # 2
I have computed the VECM conventional equation with 4 variables like electricity sales, GDP, electricity real price, population, which is also pasted below. Kindly guide me with a video or with some other means, how I can transform this equation to compute the forecast of the electricity sales. The target variable is LDOM8519 electricity sales which is in differenced form.
∆ LDOM8519t = 0.611109 + 0.185012 ∆ LDOM8519t-1 + 0.5014 ∆ LGDPTOT8519(t-1) - 0.014722 ∆LDOMRP8519t-1 -24.51419 ∆ LPOP8519t-1 - 0.48071 ect(t-1
)
Hi Bilal, my responses:
Q1: In this case, use 1 lag since VECM cannot be estimated as a static model.
Q2: My videos on "forecast error variance decomposition" will be helpful.
Thanks.
Thank for this inestimable lecture it is very precise and timely.
If the long run relationship is to be discussed upon, is it the normalized Johansen cointegration that will be adopted or the lag ECT cointegration equation?
Hi Simeon, I'm not quite clear about your query but the Normalization captures the long-run content of the model.
How do you proceed when you obtain non-uniform results during johansen cointegration test i.e you find that there is no cointegration under trace stats but there is cointegration under max eigen value? will you estimate a VAR or VECM?
Hi Dennis, I mentioned in my video on Johansen Cointegration Test that researchers take decision when the results conflict which guides subsequent approach.
Thanks.
Great video. Loved it. Got one question just to be sure. You described the lambda (speed of adjustment) and its meaning, but no mention as to its sign. But then noticed that in your example it was negative. Must it be negative to have some meaning? What happens when it's positive?
Hi Douglas, I appreciate the positive feedback. Deeply appreciated. You skipped the prerequisite videos that contained the information you seek. Kindly watch them. But note that theoretically the sign must be negative. if positive it implies there is no convergence to long-run equilibrium and the model becomes explosive. Thanks.
Thank for your video 😍 I have a question. If there is serial correlation or the residuals are not normally distributed, what would we do? Can we do anything to fix the model?
Hi MooMoo, there are several measures you can deplo: readjust the lags, change regressors etc and re-estimate the model.
I watched most of the videos from CrnchEconometrix and found very much useful to the researchers in secondary data. Now i am interested in structural VAR estimation. I request you to make one video about i, if possible.
Thanks for the suggestion, Jhabindra. Appreciated!
I have a question, Suppose if we get the no. of lags for cointegration as 1, then what are we going to use as the no. of lags for VECM, we can't use 0 (since we get an error) isn't it madam?
P.s. I love your videos, they're very beginner-friendly and clear also they're really helpful for my analyses
Hi Manoj, VECM cannot be estimated in static mode. Retain the 1 lag and put a note to that in your work. Thanks.
Hello! Thanks for videos... it's very helpful for me.
I am doing panel vecm right now and my errors are non-normality distributed. Please help me to correct that non-normality.
Thanks!
Delphine, I don't have videos on panel VECM. You may want to check out other online resources. Thanks.
@@CrunchEconometrix Alright! Thanks!
please how do we interpret VECM with 2 cointagrations equations? which equation to choose and how to interpret variable with1.000
Hi William, as explained in my VECM and Johansen cointegration test videos, keep it simple by using 1 CE except you are on top of your interpretation then you can use more than 1 CE....I still use only 1 CE regardless of if the JCT yields more than 1.
Thank you!! But what can we do if there is heteroskedasticity.....?
You can initiate a battery of measures: re-estimate at higher-order lags, and/or change regressors.
CrunchEconometrix Thank you! Can I re-estimate the equation by using the LS method but changing the covariance method to Huber-White?
Not advisable. Use the VAR routine to estimate.
Thank you for the video. So all the steps are done on stationary variables, even cointegration test?
I have created d(ln_variable) for transformation purposes and further on used d(ln_variables). But should cointegration test be done with d(ln_variables) or level variables?
Hi Ksenija, kindly watch my video on Johansen Cointegration Test. Well explained, thanks.
Hello CrunchEconometrix. I appreciate a lot your contribution with these videos, they amazing. Thank you!
Please can you help me with a question: Why ECT=-0.0670 is interpreted as "an adjustment speed of 6.7%", but for the short-run effect of PCE=0.44 you say "a percent change in PCE is associated with a 0.44 percent increase in PDI"?. Why using a different rule for interpreting long-run and short-run coefficients, or was it a small lapsus? Thank you again for your great videos!
Hi Sabrososh, thanks for the encouraging feedback. Deeply appreciated! Those are the generic interpretations. Look up the the references listed at the end of the video or my papers listed on the Community Page. Thanks
thank you for your great explanations about VAR and VECM. I have a problem with heteroskedasticity in residuals of my VAR and VEC model residuals, could you please guide me on how to deal with this problem?
and according to your responses to other people, you suggested choosing higher lags to remove heteroskedasticity, therefore this is a violation of choosing suitable lag regarding information criteria
in my data, the suitable lag for var model is 4, and the lag in which heteroskedasticity disappears is 25!
Hi Kaveh, thanks for the encouraging feedback on my videos. Deeply appreciated! It appears that the choice of your variables does not respond to heteroscedasticity. You may request to change them to closer proxies and re-estimate.
Thank you Dr. for this video. Pls how do I get to know the P-values of the coefficients since what is provided is the standard errors and t-statistics? thanks
I can't recall. Use the t-stats.
Thanks for the prompt response. But sorry to draw you back. What process am I to take with the T-stat? I really don't know
Alright, Kenechukwu. You made me go into EViews10 at 2.30am. To obtain pvalues from VAR output: Go to ->>Proc, Make System, Order by variable, Estimate, 🆗.
Lolzz. So sorry about the inconvenience Dr. I appreciate. Thanks for the info. I will try it out.
Hello My Sister! Thumbs up!! you are making us proud here. I used this video as a specimen to estimate my vecm model and arrived at the same conclusion with you but I still need clarification concerning the issue of normality. You rightly said there that while some of the variables are normally distributed others are not and in general, the model is not normally distributed but seems you were a bit relief with the outcome of the heteroscedacity . Please, what should I do as regards the normality issue? should I just leave it or correct?
Thanks Ed, for the positive feedback! Normality is of no consequence compared to serial correlation and heteroscedasticity. You can ignore.
@@CrunchEconometrix Hi, Can you please give me some references for what you just mentioned so that I can mention in my research paper, I have a VECM which is free of autocorrelation and heteroskedasticity but two equations are not normally distributed. Thanks.
@@1717ratul You can always cite published papers (tons of them) that used the procedure.
@@CrunchEconometrix I couldn't find, that's why I asked you. Sorry to bother you, Ma'am.
@@1717ratul You mean you found no papers with VECM approach? That's very difficult to believe. Anyways, do a Google search.
How do you know whether it is a significant relationship or not
From the t-stat. You can compute that.
Hello, thank you for the videos. Are we not supposed to change the signs for the long-run model? So should it not be (+) 3.63? Or do we only do that in the Johansen part? I was told to flip the signs in the long run part.
Hi Owen, please watch my video on Johansen Cointegration Test. It answers your question. Thanks
hi professor, thank you for the great video, it really helped, ill be sure to share your channel. I have a question regarding the optimal lag for the VECM model. my optimal lag is 7. does this mean for the VECM my optimal lag would be 6 (p - 1). If so which short-run dependent variable lag should I use. aka d(GDP(1)). D(GDP(2), D(GDP(3) etc..
Hi there, yes use 6 lags for the VECM. Follow the guide as explained in the clip on how to proceed.
Thank you mam,
May I ask about the t-statistics? How do we know the value in the t table? And how do we know the degree of freedom in VECM?
Hi Emira, from the output the t-stats are in [ ]. On how to get that from the t-table, I will advise you read up HYPOTHESIS TESTING from any econometrics textbook for detailed guidance. Thanks.
I love all your videos. very clear and easy to follow. I need the link to a video on step 4 (When there is no cointegration)
Thanks Zama, estimate VAR when JCT shows no cointegration. Click on the Playlists to watch my videos on the procedure. May I know from where (location) you are reaching me?
@@CrunchEconometrix Thank you. i am watching your videos all the way from Eswatini (formerly known as Swaziland)
@@babaywakhe Awesome! Please spread the word about my videos to your students, friends and academic community in Eswatini! 💕 😊
Thank you really you clarify my problem on ECM interpretation
Thanks for the positive feedback, Muhd. Deeply appreciated! 💕 May I know from where (location) you are reaching me?
I have a question. if Granger test shows that there's no casuality relationship between two variables then do we have to stop the research?
Báo, not at all.
4:27 I don't understand this part. the coefficients have low t-stats, so how am I allowed to put those coefficients into the equation when they're not significant?
Hi Ecksdee, you construct your equation regardless of whether the coeff is statistically significant or not. Confirm this from any econometrics text.
@@CrunchEconometrix hi ! any text book suggestions to find this?
@@danaeellina2645 You mean you don't have any econometrics textbook that you use for studying?
@@CrunchEconometrix I do but I have not found one that says its OK that the error correction in 4.27 is insignificant. I need to back this up somehow so I would love a suggestion on one that says that we must include the error term irrespective of its significance!
Alright.
Thank you very much for the videos!! Is it a problem because in my model, the probabilities are less than 5% level so there is serial correlation and on normality all components showed a high prob value hence are distributed normally however the total shows that residuals are not normally distributed. And on heteroskedasticity the prob value is 0.44 so the model is not heteroskedastic. So what should i do or its fine like that?
Takunda, correct the model for serial correlation. Estimate the model at higher-order lags.
@@CrunchEconometrix Thank you, much appreciated!! I used 6 lags for serial correlation only and starting from the 2nd lag up to the 6th, the model is now showing that there is No serial correlation. However, i wanted to know is it not a problem because from Var lag order selection criteria i was using 2 lags but i now change them to 6 lags on residual tests
No, it's not. Put a note note in the text explaining what you did about the diagnostics.
Thank you for the video. From this, how can I proceed to assess causality?
U're welcome, Pedro. Simply watch my videos on VECM and Causality Tests. May I know from where (location) you are reaching me?
@@CrunchEconometrixThank you very much for your answer. In the mean time, I've already watched almost all your videos and understood that. I'm from Portugal. Thank you
coefficient values shows e (t-1) values in the running model as you hove shown after performing VECM ?. If it is significant and negative it means there is long run relationship exist.
Yes
Hello maam I am here again. Thanks for the video again and I really thankfull, but I have another question. How to interpret the ECT Coefficient that has (+) positive sign? So, I have a significant ECT coefficient at 5% and the coefficient is 0.534676. Is the way to interpret it different compare to ECT coefficient with negative sign?
Hi Rizka, positive ECT is interpreted in the opposite. No convergence to long-run equilibrium. Model is explosive.
@@CrunchEconometrix thanks for the answer mam. You are so humble and answering each of the question here in comment section, I already transform my variabel into log natural now it shows there is one cointegrating equation, but from the VECM it shows that the ECT not significant from t statistics also p-value and the ECT has negative sign. So should I interpret like " there is long equilibrium between variabel but not statistically significant?"..Thank you again mam 🙏🏼
Correct. Or simply "no significant reversion to long-run equilibrium".
@@CrunchEconometrix thank you so much mam, really appriciate it, I am working on my undergraduate thesis..thanks a lot for all of your video about this method 🙏🏼💙
Thank you very much for your lessons! It is incredible and very helpful for me!
Also, I would very much like to clarify a few points.
Please tell me what we should do if after Lag Length Criteria and Lag Exclusion Test my lag intervals are 1 1 4 4 7 7? How exactly can I reduce these intervals by 1 lag for the VECM?
In this case, I would also like to ask what exactly the number of lags I should use for the LM-test (if lag intervals are 1 1 4 4 7 7)?
Thank you very much in advance for your answer!
Hi Olesia, kindly watch the clip again and follow the guides shown. Do same for the video on diagnostics. Thanks.
Thank you very much!!
Thank u so much as u have explained the concept very well. I have a query that my vecm is not fitted in residual diagnosis. There is no autocorrelation but rest two conditions are not met. What can I do ? Kindly help me out.
Hi Nutan, re-estimate the model at higher-order lags.
Ma'am, why did we take t-1 subscript for the ECT model, not t minus anything else
Srishti, that's the way the ECT is specified. I suggest that you scour the literature for detailed explanation about the ECT. Thanks.
Thank you for your explanation, I really appreciate it!
Further, I want to ask, what if the Cointegration Equation is not significant? Does it mean that both in the long run and short run/only short-run have no relationship between the dependent and independent variables?
Hi Utama, if that is the case then the long run relationship is statistically not significant.
Hi Prof...thanks for this video...so many doubts of mine have been cleared and the way you demonstrated the interpretations was outstanding. I have one question - how do you decide on a target variable, when you have many variables in the Johanssen method? Should we go for a causality to determine the target variable? Look forward to hear from you. I am based in India...
Thanks for the positive feedback, Jyoti. Deeply appreciated! The target variable is the 1st variable listed in the VAR system. It is the variable upon which the JCT is normalized. Hope this helps. Kindly share my TH-cam Channel link with your friends and academic community in India 🇮🇳. They will find the content helpful too 😊.
@@CrunchEconometrix Thanks a lot Prof...I will certainly..all the best
Awesome! Thanks a million! 😀
Thank you so much ma for a good presentation. In a case where there are two cointegrating equations (that is none and at most 1 are asterisked) , do we interpret the D(...)(-1) or the D(...)(-2) of each variables for the VEC estimates?
Also following the video preceding this one, in a case where we have the variables stationery at first difference but different max-lag lengths for the unit root test, is it still acceptable to use or not acceptable to use if the are different unit root lag lengths for the variables?
Hi Praise, thanks for the encouraging feedback. Deeply appreciated! Follow the guide and explanations. Use 1 cointegrating equation. Same lag length is used for all variables.
1. In the long run part of ECT model, the coefficients of regressors must be interpreted in reverse signs. 2. When we run VECM practically, reduce one lag for D(Endogenous).
Yes.
Hello,
Thank you for all your uself videos.
Can you please clarify if you use a Normal distribution to assess the variables significance and if you use a two tail test?
Hi Martina, regression outputs come out as 2-tail tests.
@@CrunchEconometrix
Thank you for the advice.
Are you using a Student T-distribution to check the critical values?
@@martinamarino8700 I use the corresponding pvalues.
could you tell me some bibliographic references please? love the vid
Thanks Amilcar, for the positive feedback. There are several papers on VAR/VECM procedure online that are freely accessible.
When interpreting result we have to change the sign
You change the sign only for the Johansen long-run equation.
Hello!
I have been following your videos and I want to thank you for the help they have given me in my dissertation.
One question: How do you resolve heteroscedasticity if it arises on White's test?
You can take logs or change regressors.
@@CrunchEconometrix Thanks for your feedback! By the way, since the dummy variables are exogenous, how can we measure their impulse/response function?
Thanks!
@@duartegracamoreira You cannot.
Vecm will run level data or first difference data
Hi Arup, difference variables. I explained and showed this in my VECM videos.
Dear ADELEYE, would you please answer these two queries?
(1) In my model all variables (both IDV & DV) are cointegrated at I(1). Thus I used VECM. While running varsoc , lag one [1] is suggested by all criteria. Now, while checking cointegration, what lag should I use? one or two? And what lag should I use in estimating vecm?
(2) In my other model, cointegration exists but while running ecm error correction term is found insignificant, in this situation what should I do?
Hi Chinta, for (1) I gave clear explanations. Kindly watch my VECM videos again. For (2), simply interpret your results. There is a long-run relationship (cointegration) but reversion to long-run equilibrium (EECM) is not statistically significant. Thanks.
What we should do in vecm lag selection if lag is 1
Hi Neha, retain the one lag and put a note in your work on why you did that.
As error term distribution does not follow normality so what can be done
Hi Arup, NORMALITY TEST under VAR-VECM can be overlooked because it's SYSTEM of equations. Focus more on the model passing both HETEROSCEDASTICITY and AUTOCORRELATION tests.
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.
Hi Jacky, the outcomes we get have a lot to do with the variables used for analysis not necessarily because the econometric procedure is wrong....assuming things are done properly.
Thank you very much for your videos. Why did you interpret the ECT coefficient term from the VECM differently (as a log-log rather than as a log, I guess) from the other coefficients? It was a lapsus-linguae?
That's the form for interpreting the ECT.
Morning Prof, may i know Error Correction Term (ECT) stand for any used?
With it be same term to use for mine any other VECM analysis test??
Hi Eleni, there's a lot to read up about regarding the ECT. Kindly get any econometrics textbook and other online resources. Thanks.
Thank you for your good work. Please, I have a time series, all are I(1) and cointegrated but my optimal lag is 1, can I still go on and run VECM given p-1 condition
Hi Dayo, this is not a problem since VECM cannot be a static model but must contain lags of the regressors. All you have to do is to maintain the 1 lag and put a footnote to explain why....and thanks for the kind words on my TH-cam Channel. Humbly appreciated!
Thank you, Mama
You are so welcome, Dr. Tsovini! 🙏🥰
How can I correct for autocorrelation, heteroscedasticity and normality problems?
Estimate the model at higher-order lags.
Thanks for the very useful video.
I was wondering if we can use Granger causality and VECM for the variables which are stationary at their level?
Reza, you have muddled up this query. You can't do VECM with level-stationary variables but Granger causality can be done regardless of the the level of integration of the series.
Now, how can I measure the long run causality? And, what should I do with them? Simple regression?
@@rezagod7951 Watch my videos on "VECM and Causality Checks". Well-explained and practice along with your data.
Mam if we found 3 cointegrating equations, then while running vecm, should we enter 3 or 1 in rank option? If i put one, will it make any difference. Or if i put 3 then how to interpret three cointegrating equations?
Waleed, I mentioned that you keep it simple and use 1 CE.
What should we do if the residual diagnostics do not agree with the model i.e there is autocorrelation and no normality.
Hi Ann, you can do a combination of measures: estimate the model at higher-order lags, change regressors or control variables.
Yes When I changed the lag order to 2 autocorrelation was not present.However both normality and heteroskedasticity did not agree.Does this mean the model is insignificant?
@@annalexander6384 Try a log-log model and increase lag to 4 if you have enough observations. Don't worry too much about "normality" it is rare to obtain a not significant pvalue due to the different equations in the system.
Ok I will. Thank you so much.
Ok I will. Thank you so much.
Dear teacher, is there a heteroskedasticity test like that in Stata as well? Also, you have chosen the (No-Cross Terms) option in your test. There is the (With cross terms) option too. Would you kindly explain when should we use each of those two options?
Hi Revolta, you can do further online readings on both concepts. Tx
hello profesor, i have monthly data for 14 years. in Johansen test I have at most 2 cointegration. when i run with the optimized lag, i have a lot of coeficients
I explained in my video on JCT to keep it simple by using 1 CE.
Hi, really appreciate this video! How do we know which of the variables are cointegrated with the dependent variable? or am I missing/misunderstanding something?
Hi Jamie, kindly watch the clip again. Well explained. Thanks
How to do if there are evidence of heteroscedasticity and the residuals are not normaly distributed in the model? And i have annual data with lag 1.
Hi Mouanguissa, you can explore a combination of measures: use logs, estimate with the "robust" option, change regressors etc. May I know from where (location) you are reaching me?
I have two integrated variables of order 1 and two integrated variables of order 0 and I used the ARDL model.
please how can i use causality between these variables
Watch my video on "ARDL-ECM and 3 ways causality checks".
How can I interpret the coefficient of ECT if the sign is positive ?? Thank u in advance
It implies that the model diverges. There's no reversion to long run equilibrium. Model is explosive.
CrunchEconometrix so in this case I should interpret that like this or may I change my model . But after researching on other plateforme I found that it is possible if there’s a structural change or break is this right .
There is no right or wrong. The result is what it is. Surf the web and you will find papers with positive ECT. You may also decide to re-model by changing or removing some regressors.
Mam I have 5 variables and 4 cointegrating equations. I have taken 2 exogenous variables too. Mam can I proceed with the model and how to interpret 4 cointegrated equations
Hi Manali, kindly watch my video on Johansen Cointegration and take my advise to "keep it simple". Use 1 CE. Thanks.
@@CrunchEconometrix Thank you Ma'am
Hi. I have 5 variable with cointegration.
Can i choose between VAR and VECM ?
Or only VECM ?
Thanks
Please watch my VAR and VECM videos. Well explained.
I watched the video. I see in video that if the variable I(1) are cointegrated yo estimate var AND evcm.
But its (and ) or (OR)
@@CrunchEconometrix so i choose beteween var and evcm or both
You may decide to show the VAR estimates before the VECM.
Dear Ma'am, as in your ECT equation the constant term is negative and you are saying it should be negative. But in ECT equation it is positive. My question is if the constant term in ECT equation is positive, is it a good sign in estimation or not, or there is any problem in my estimations ?
Hi Tariq, the sign of the ECT is the most important thing in the model.
@@CrunchEconometrix Dear Professor, Thank you for the prompt response.
Question: How can I Interpret my results?if
1) ARDL-ECM (OLS estimations) results are insignificant.
2) ARDL-bounds test: cointegration Exist
Appreciate! Stay blessed!
Is the interpretation of the ECT coefficients in percentage because of the logs you take (lndpi, lnpce etc..)? Or would this be a percentage for normal variables as well?
Adapt the interpretation given in the video.
@@CrunchEconometrix At 4:30 you say that 'the previous years deviations from long-run equilibrium is corrected at a speed of 6.7%.' If it wouldn't be a log variable but a regular variable, would it just be 'corrected with a value of -0.067'? Thank you in advance
Max, since you are not paying attention to my guide, you can easily clarify your doubts by reading interpretation of the ECT from any publication.
@@CrunchEconometrix okay, I was hoping for just a short answer yes or no
Max, you need to read to be sure of the information you need. You may not get a YES or NO answer all the time. I've done my best to guide you but you aren't paying attention. So, my honest and candid advice is that you read. I've made that clear in almost all my clips: video tutorials are not substitutes for reading. Please READ to strengthen your academic confidence.
After doing the VECM test I ran the diagnostic to see if there was heteroskedasticity and the residuals were found to be heteroskedastic. How can I correct it?
Hi Ann, re-estimate the model at higher-order lags.
@@CrunchEconometrix ok Thank you :)
hi , you said "a percentage change in pce is associated with 0.44 % increase in pdi on average ceteris paribus in the short run" but what you mean with change in pci increase or decrease ??
What are you referring to exactly?
I have a problem with my regression. I'm using Eviews as the econometric software and I'm running Vector Error Correction Model (VECM). I have a problem of heteroskedasticity and autocorrelation. I don't know how to correct them in Eviews using VECM.
Hi Bentou, correction will require several measures not one. You can try increasing or decreasing the lags, changing regressors or control variables.
I have some issues about this model..
1.Is it necessary to include the ECM equation (raw equation,not computed one) in my article
?
2. Though my speed of adjustment is negative but it is statistically insignificant ( p value is higher than .05) ,is it a matter of concern?
3.Relevance of Macro Economic factors for the Indian Stock Market
by Aman Srivastava
(2010) order the equation in another way, was it wrong?
Hi Abdullah, I've read your queries over and over again I still don't understand what you mean?
Hi I am watching your videos from Swaziland. In your Johansen normalization test we changed the signs and the impact of LPCE was positive on LPDI and GDP had a negative impact in the long run. However, in the VECM results we kept the signs the same way when interpreting for the long run model. LPCE had a negative impact. Is there such a thing as normalizing VECM results as well?
Hi Zama, thanks for watching my videos. The VECM procedure is well explained. Follow my guidelines. Thanks.
@@CrunchEconometrix hello Dr, do we inverse the sign when interpreting VECM in the long run?
@@meditatewithme4933 No. Give interpretations as seen on the output.
@@CrunchEconometrix ok thank you ! because i've seen some notes , they reversed the sign in the long run model of vecm
CrunchEconometrix Im sorry for asking again , because long run cointegration equation is the same as normalized JJ right ?
In ur other video, normalized JJ is interpreted by reversing their sign.
So in VECM we don’t have to do that?
Respected madam please tell me the reason of inverting cointregrating coefficient sign while explanation?
That is because the Cointegrating Equation (CE) is in IMPLICIT form. That is, CE = (Y - X1 - X2 - X3). Therefore, to know the impact of the regressors on the depvar the equation must be stated EXPLICITLY which is the reason for INVERTING the signs of the coefficients.
Hello Prof, many thanks for the youtube lectures. I have a question please: my step 1 and 3 are all met but i have two integrating equations after doing step 3. I am a bit confused, shouldn't i have one cointegrating equation? many times i see one cointegrating equation. what do i do next? below text is copied from the table in step 3.
Trace test indicates 2 cointegrating eqn(s) at the 0.05 level
* denotes rejection of the hypothesis at the 0.05 level
**MacKinnon-Haug-Michelis (1999) p-values
Keep it simple, use 1 CE unless you can interpret your results with 2 CEs.
Ma, in the preceding video, you did not difference the variables before running the Johansen test, after they proved to be I(1). May I please know why? Thank you
Hi Reine, JCT is performed on the variables in their natural and not transformed form. Refer to any econometric textbook on the fundamentals of cointegration. May I know from where (location) you are reaching me?
Thank you, Ma for the videos. It is possible for a VECM to be unstable when all the variables are I(1)?
Hi Sherif, yes it is possible.
The F-statistic is not explained and how to obtain the probability of it is not mentioned
The F-statistic has the standard interpretation of proving the joint significance of the independent variables.
@@CrunchEconometrix please at what point can we say fstatistic is significant or insignificant in VECM? AND can you please help me make video on only the interpretation of Structural VAR? THANK YOU.
The pvalue of the F-statistic will point you to the right direction. You can easily pick up publications that use structural VAR and read up on the interpretation. Thanks
In STATA 13 episode of estimating and interpreting the VECM you said we should interpret the direct of influence opposite to what we obtain in the software (that is if lnpdi comes out as negative we should interpret as positive). Please why is it that over here, in interpreting the results, you did so just as the signs are and never interpreted them at the opposite direction?
Enoch, the reason is because the long-run equation is expressed in linear form. Please watch my video on Johansen Cointegration technique for deeper understanding. Thanks.
regards how do interpret a positive short run equation coefficient
Same way you interpret long-run coefficients.
@@CrunchEconometrix i have watched the video over again and i cant seem to find the part where you interpret the long run coefficients. you just state that this is the equation that signifies long run cointergration
Dear Ma'am, Can we still use one cointegation equation, if the total number of cointegation equation are 6.
Hi Tariq, kindly watch my video on Johansen Cointegration Test for detailed response.
@@CrunchEconometrix Thank you, Dr. for the prompt response. I really appreciate all your sincere efforts. Stay blessed!
Hello professor, may I just ask one more thing regarding your model? is the ECT also the same for all target variables? Like, for example, is if i choose D(LNPCE) instead as a target variable?
No. If you read the textbooks and published papers you will see that every target variable has its own ECT in its VECM.
No. If you read the textbooks and published papers you will see that every target variable has its own ECT in its VECM.
Hi Bosede, I just want to find out if the sign in the VECM output matters?
Hi Zama, I will say it does because you have a priori expectations for the signs and significance of the coefficients. May I know from where (location) you are reaching me?
Hello mam, after running VECM, the diagnostic tests show auto correlation, heteroskedasticity and no normality. what to interpret now? and if the interpretation says given VECM model is invalid, then what to do next?
Hi Boni, re-estimate the model by playing around with lags...or change regressors if you have it. You have to try several solutions to fix the problems.
@@CrunchEconometrix ok mam. thank you so much for your guidance
Hello Prof, how can I solve the problem of heteroskedasticity and autocorrelation in VECM?
Re-estimate the model using higher-order lags.
Two issues I faced: first, the optimal lag to choose came to 1. So, I used 1 lag for cointegration; how many lag shall I choose for estimating VECM. Another issue, I am getting 1 cointegrating equations under trace and 2 cointegrating equations under maximum eigenvalue . Please help me out
Hi Pravakar, I'll suggest you maintain 1 lag for VECM since you can't estimate with 0 lags (that'll make it a static model and not autoregressive). Also, I have always suggested keeping things simple by using 1 cointegrating equation unless you are on top of interpreting your results which can become a little knotty when using 2 cointegrating equations.
Hello again, I got speed of adjustment a positive number... is it an absurd result
@@pravakarneupane7299 Well, sometimes it's not absurd if it can supported by economic theory. But if not, you can look for better proxies for the variables and re-run your model.
@@CrunchEconometrix Hello Ma'am, I too face the same issue of having 1 lag as optimal through my VAR estimation. So as per theory, I need to estimate my VECM as (1-1 =0) lag, which is not possible. So as you suggest I can carry my VECM estimation with 1 lag. Secondly, my trace test indicates 2 co-integrating equation whereas maximum eigenvalue suggests none. Is this wrong or how to interpret this? Kindly suggest!
@@saakshijha9689 As explained in my Johansen cointegration test video, you are disposed to using EITHER of the results this is because oftentimes the two tests yield similar results and sometimes they don't.
Hi. Thank you so much for your assistance. I have been following your tutorial with no problem until I got to the part where I need to run the White Heteroskedasticity Test. Eviews is not responding and just gives me an error message saying: Positive or non-negative argument to function expected. Then the output I get from Eviews is gibberish in a way. Would you be able to advise what to do? I notice that when I run it with optimal lags of 1 (even though my optimal lags are 6 per the AIC) I don't have this problem.
Hi Khanya, then I advise you estimate with 1 lag. May I know from where (location) you are reaching me?
@@CrunchEconometrix from south africa. Thanks so much for your response :)
@@Kndzu Awesome, Khanya! Please spread the word about my TH-cam Channel to your friends, students and colleagues in South Africa for awareness. Thanks 😊.
@@CrunchEconometrix I already am. Thanks again:)
Hi Bosede. I am in need of your assistance again. I am running a VECM for my study and it appears the data has many outliers. To deal with this, my plan is to add dummy variables.
Would you mind advising if there is any best practice on how to decide where to add dummy variables?