RUclips 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.
@@borisogou4756 Awesome! Kindly spread the word about my RUclips Channel to your friends and academic community in France 🇫🇷 for awareness. They'll learn some useful tips and hints too. Thanks 😊
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, I have three variables 1) Price 2) harvested Area 3) Yield rate of rice. I am trying to test if there are relationship of each variables between each other so that i can draw some conclusion regarding whether stabilise the rice price can help to stabilise the yield/ harvest area. I tried to take a log of these three variables in order to better understand them in % change. not that they are skewed. I then test the stationarity by ADF, showing that only ln (Yield) is stationary. I then perform the commond differencing and test again and so on. I finally got ln (Price) after one-time differencing (ln(1)) and ln (Harvest Area) after two time differencing (ln(2)). I then also conducted the Johansen integration test on the variables ln (Yield), dln (Price) and dd ln (Harvest Area), the result shows that they have cointegration relationship for at least 2 variables. Cointegration test result here enter image description here On the final part, I want to fit those three variables into Vector error correction model. The problem is here, in order to fit the VECM, all variables should ideally be integrated of the same order, typically I(1) which is not my case here. if that is a absolute requirement? can my case use VECM? if no, then i should go for VAR instead? or you have other suggestions?
Please try to be very brief. Depending on the outcome of your unit root test AND your research objectives, you can decide between ARDL-ECM or VAR-VECM techniques.
Hahahaha Omar, it's the least I can do. Glad to be of help. I will appreciate if you can share the link to my YT Channel with your colleagues and friends on social media. They can gain a lot too. Thanks!
@@CrunchEconometrix This reply is a month late, so accept my apologies. I've already shared a link of your channel with about 250 students of my colleague. Your content is really valuable and extremely helpful. Your way of explanation makes econometrics an easy task. Sincere Thanks.
Hi Moses, thanks for the encouraging remarks. I have just uploaded the part 2. I'll appreciate if you share my videos and links. People need to know that econometrics is not as difficult as it seems.
Thanks. I have seen it and its very insightful. Quick Questions: 1. If i have a mixture of I(0) and I(1), can i estimate VECM or i need all of them to be I(1) only. 2. Is there a way to estimate VECM under ARDL model (just like ecm in ARDL) thanks Moses
Moses Mathu On your queries: (1) answer is no. All series must be I(1); (2) yes it's possible. That's the approach I explained in all my ARDL videos...please watch them starting with the video "this is how to specify ARDL models". Thanks again for the encouragement, deeply appreciated. Don't 4get to share😉
Thank you. the videos are really helping me (an econometric beginner). clarifications. I have 5 variables : inflation (dependent variable I(1), Money supply I(1), exchange rate I(0), deficit I(0), rgdp I(1). am interested on effect of deficit on inflation). due to endogeneity, i have also added two more equations with deficit and ms as the dependent variables. q1. do i conduct contegration test on I(1) only or on all variables q2. in the cointegration test, do the I(1) enter in levels or differenced q3. Assuming i get cointegration in 3 eqns, my understanding is that i run ECM for each eqn (separately as you explained).is this right? q4. again, do the I(1) variables enter the ecm at level or differenced. Apologies for the long post but i need to be clear. thanks
Thanks for letting me that my videos are helping you. it's my joy doing what I do. But, since the variables are integrated of different orders, I hope that you are estimating ARDL-VECM and not VAR-VECM? On the queries on cointegration, the answers are in the 2nd part of this video (How to estimate and interpret VECM (2)), so kindly watch it. Please assist me to share my videos to your colleagues and social media community they need to know that econometrics is not as difficult as it seems. Thanks!
Great video. The other EViews videos are also worth watching. There is an All Unit Unit Root test add-in in the toolbar. It saves time; you can use it too. Thanks a lot for your incredible videos
Hi Abbas, you will need to grab any textbooks for detailed differences between ECM and VECM. You can then support your reading with my video tutorials that show how to estimate both. Thanks.
Thank you very much for your helpful video. Now I have a defense for what I did using VECM, but I do not know why we interpreter the long-run coefficient reversely. please, help what is the reason behind? Thank you.
Hi Zebider, thanks for the encouraging feedback. Deeply appreciated! The LR equation is in IMPLICIT form. Move the explvars to the RHS...justifies the REVERSE interpretation.
Dr, what's "Positive or non-negative argument to function expected" means in heteroskedasticity test? is it bad result? should I reduce the lag? Thank you Dr.
Hello, please, if we have trend in some series, we must correct it and then check if the series are stationary in the first difference? second question please, we must check the stationarity in the 3 models(intercept/ intercept & trend/None) or only for one model? thanks a lot for your times.
Thanks a lot ma'am. You are helping the researchers like me. we are highly benefited by your lectures. Ma'am, I want to ask one thing can apply Cointegratioan and VECM if the variables are stationary at level (Means both are I(0).?
Hi Dr. Adeleye, this is Hussain from Saudi Arabia. thanks alot for your helpful videos. I really appreciate your efforts. I have a question, I am also using quarterly data (but just 40 observations). What’s the maximum number of independent variables I can use? How to determine it? Thank you 🙏🏼,
Hi Huss, choosing indpvars suitable for answering your research questions is at your discretion. It is better to read the literature to see what obtains.
When testing for cointegration between 2 variables, do i use the johansen test or the engle granger test, and which model do i then use to model my 2 variables. if there are videos on this please send the link, thank you.
after performing the test, trace test shows one cointegrating equation while the max-eigenvalue test indicates no cointegration. what to do when Result of trace test and Maximum Eigenvalue test differs?
Dear Madam, this is very much helpful video. Could you please let me know what can I do when my optimal lag length become as lag 1? Not as lag 2, its a trouble to analyze VECM in lag 1.
Hello Ma'am, thank you so much for the video. I have a little challenge. What do I do when the optimal lag length is identified by the criterion, but when i try using it in the model, I get an error message stating ''insufficient number of observation''?
Several reasons. Too many variables, too many lags and/or a short time span. For instance, too many lags will reduce the number of observations. Reduce the lags to 1 and re-estimate the model, but make sure you have at least 3 observations BEFORE doing so.
I have one query, if I have daily closing price of stock, why do take the airthmatic or log return for the series, why cannot we take the value of series directly as closing price and apply the test
Of course you can...only that the returns (log difference) is stationary. You can use the difference of the closing price too depending on your research focus. Using "returns" is not mandatory.
Really thankfull for the video and your explanation Mam. I have a question, in your video there is no list for VAR Stability test. Is it no need to check the stability because there is no Impulse Response Function and Variance Decomposition test? thank you so much mam
@@CrunchEconometrix okay sorrry mam, so I've read some of journal and they are using VAR/VECM stability test and also IRF and VD test, so I want to ask..is it necessary to do those test? And I have another question mam...My VECM model didn't pass the diagnostic test, is there any way to solve this problem? thank you mam
Thank you for all these videos ,I really need your help . It's better that I give brief about my case . I have one dependent variable Y and 7 independent variables . My aim is to forecast Y but i am not sure which model is more suitable. All series are non stationary and not normal at level .I took the logs and first difference for all series and I ran VAR eviews model . After that I tested the optimal lag and based on AIC it was 2 . Then i tested for co-integration using 2 lags . There was two cointegrations defined . As far as my understanding if the variables are stationary at first difference and there is co-integeration , I shall go to VECM model with (p-1) . Once i run the VECM model , the residuals are not normal, serially correlated and has heteroskedasticity. What shall I do in this case ?
Hi Ahmad, yes such can happen. You may need to reduce your regressors and re-estimate. If the problem persist, estimate the model at higher lag levels.
@@CrunchEconometrix Thanks , I hadn't checked for Multicollinearity before . Now i did and i found two regressors with high VIF .will eliminate them and try again. Will let u know .
May you please advise how may I reverse the log function . I took the logs of all series but now after running the model and forecast , I would like to interpret the results by seeing the real numbers .
Thank you very much for the good content! I was wondering, why do you opt for AIC when you conduct the unit root test in the beginning? Is there a specific motivation for you to do so? Additionally, if you were dealing with daily data (5 days a week), what lag length would you recommend? Many thanks
Thanks, Owen for the encouraging feedback. Deeply appreciated. The researcher decides on which information criterion to use. I will advise you watch my clip on "Optimal Lag Selection" for background guides to selecting lags. Kind regards.
Thank you so much for this extremely helpful video! I just wanted to ask if I have 2 cointegrating relationships how should I interpret the long-run equation? and which cointegrating equation to use?
Respected Ma'am , The researchers are getting too much benefits from your lectures. You explain the difficult things in a very simple manner. thanks a lot for guiding us in econometric techniques. . Ma'am, I want to ask one thing that can we apply cointegration and VECM (If Variables are Cointegrated) on the variables which are stationary at level, means I(0) ?
It's my pleasure, Suresh. Thanks for the encouraging feedback. If all the variables are I(0) proceed to perform just OLS. I have a video on that. Search through my Channel and watch. Kind regards.
A Bundle of Thanks for replying Ma'am. Can we apply VAR model also on two I(0) series? In my paper, I have applied VAR model, Wald test for joint significance, Granger causality test, Impulse Response Function and Variance Decomposition model on the panel data of two variables both of which are I(0). M I going right? I need your expert suggestion. Please help.
Thank you very much for this video. It helped me a lot with my study. Can you please help me to know how we can know that it is significant at 1%? I understood how to get the t-stat but I don't understand what value to compare with.
Thanks, Jean for the encouraging feedback. Deeply appreciated! I will refer you to read about hypothesis testing and confidence intervals from any econometrics textbook for deeper understanding of statistical significance.
Dear Mam, thank you so much for this informative video. It helped me a lot. However, I have a few questions and I will be really grateful if you help me with the answers. In Johansen Normalitaization Interpretation, what percentage should be compared for testing significance of coefficients at 5% level, and will it be statistically significant if the t-statistics is greater than z value both for plus and minus? Thank you again.
Thank you very mich for the video I have two questions: 1. Does the Normalized cointegrating coefficient interpretation replace the Granger Causality test in the case VAR and VECM (for short and long run). 2. If so, should we consider the same method to interpret the Normalized coefficients which is reversing the sing in the case of short run ? If what I have mentioned is possible... I highly appreciate if you share with me some references to support my analysis Best regards
Hi Hechem, NC and GC are not the same. Please watch my videos again on these procedures. There are tons of papers on the internet that you can download, read and reference.
Dear Ma'am, I applied the given steps while doing my estimations on my time series dat of 40 annual observations. After estimating the VAR in computed the optimal lag length which is 5. Now doing the Johnson coIntegration test , when I am putting the lag interval 1 to 5, it give an error "Near singular matrix". What should I do to eliminate this error in order to proceed further.........................................Moreover although the optimal lag length is 5 (which in this case is not helping me to further proceed my estimations) so i put arbitrary lag length of 3 and with this arbitrary lag length the Johnson coIntegration test is not giving any error. Concluding it, my question is " If the estimation with the optimal lag length is giving an error, can we apply arbitrary lag length for conduction of Johnson coIntegration test ?............. Provided that, my observation is betwen 37 to 40 (the data is annual)
@@CrunchEconometrix Dear Professor, How can I interpret my results? 1) ARDL-Bond test:Cointegration Confirmed 2) Model: ARDL-ECM through OLS 3) Probability of coefficients for ECM and all independent variables is higher than 0.05. Thank you!
@@CrunchEconometrix I didn't have seen your response, i'm sorry. Thanks for you, i get it now. I'm from Côte d'Ivoire, the greatest french speaker country of western Africa 😂lol
What if in the normalized cointegration equation these coeff divided by the standard errors give 1.18 for the first variable & -3.53 for the second variable? Also: there is no cointegration, so should this normalized cointegration equation be interpreted?
By now you should understand your t-stat and give appropriate interpretation same for if there's no cointegration. You get better when you read textbooks and articles to support video tutorials.
Thank you very much for the videos, they make econometrics doable. I am trying to estimate fiscal multipliers and i have variables gdp, government expenditure and tax. Is it ok to put gdp as my target variable? another question; is it possible for the results of trace and eigenvalues to differ when doing co integration test? if so which ones do i take?
U're welcome, Lydia. CrunchEconometrix was set up to make econometrics crunchy, delightful and fun. To your query, please watch my video on Johansen Cointegration Test and you have your answers....thanks once again for watching my videos. Please tell others too :)
Hello ma'am, is it permissible to conduct the Johansen Cointegration test on RAW DATA, and then if I find cointegrating relationship in the results, can I conduct VECM on LOG TRANSFORMED DATA for better interpretation of results? Or is it necessary to perform the Johansen cointegration test on log-transformed data if I intend to perform the VECM test on log-transformed data? Thank you so much for your videos and assistance. They have literally helped me complete my research work!
@@CrunchEconometrix Thank you so much for your reply ma'am. I watched your tutorial on Johansen Cointegration test. But I think I couldn't explain my query clearly. Ma'am my confusion is that suppose I apply the Johansen Coint. test on RAW data that is not log-transformed and I find a cointegrating relationship exists, then can I use log-transformed data to conduct the VECM test or will I have to use raw data for the VECM test because I used raw data for the Johansen Coint. test? So, In case if in my study I use Raw data ( not transformed into log) for Johansen test and then Log transformed data for VECM then will this affect the reliability/accuracy of my results?
CrunchEconometrix hi Prof, i think Darcy said in this video at ,7:23 You mentioned that the t statistic is above 2, and it is significant. Thus, what is the rule of thumb in identifying the p-value and its significant? How do we know that it is significant.
@@meditatewithme4933 I remember, it is basic t stat knowledge. You can easily check on Pvalue from t stat, with degree of freedom: observations - number of variables
Hello Crunch, Hope everything is going well! I have a question: If my variables (1st diff of log_youth unemployment and 1st diff of log_GDP) are not cointegrated... I can't do the VECM right? Should I do the unrestricted VAR? What do you mean with unrestricted VAR? Typical VAR Model? Thank you!
@@CrunchEconometrix Thank you very much Crunch! I am working on the problem of Youth Unemployment and I am modeling a VAR Model with 4 Lags (the best based by Lag Length Criteria) but I don’t know if an ARDL Model could be better... I understand that VAR Model has dynamic (and Impulse Responses also)... Am I right? Should I stay with my VAR or maybe the ARDL is also a good option? Thank you Ma’am! 🙌🏻❤️
Thank you for your excellent presenation. if i the log variables on thier level not all of them non staionary while in thier nature (without log) level all of them nonatiaonary, can i apply their logarethim? if i find the varaible are not normal distributed while other statistics are fine (no serial correlation no heterscadisicty) can i accept the model? i got the following results LKSEI(-1) 1.000000 0.000000 LOG(TT(-1)) 0.000000 1.000000 LHINDEX(-1) -0.154923 -1.038932 (0.25311) (0.08584) [-0.61208] [-12.1034]
Hi Jasem, there are no rules to taking logarithms. It is at the discretion of the researcher. But if the variables are nonstationary, take their 1st difference. Also, I will give a model a good pass if it only suffers non-normality but has no evidence of serial correlation or heteroscedasticity.
what we do if the long run coefficient of the variable is not statistically significant but they have con integration and the speed of adjustment is statistically significant? specially in the case of exchange rate devaluation and trade balance if the coefficient of real exchange rate is not statistically significant can we say there is long run causality from real exchange rate to trade balance in the long run? in VECTOR ERROR CORRECTION MODEL.
Debela, the H0 is no causal relationship. With insignificant coeffs, H0 cannot be rejected. However, a significant ECT evidences joint causality from the regressors to the outcome variable. May I know from where (location) you are reaching me?
Dear Professor, I need your help. I have monthly data for unemployment rate and for the number of persons who checked-in in turism units. The lag length that i have chosen is 24, and i obtain that the optimum lag for LR is 20 and for AIC is 5. With the optimum lag 20 i have cointegration, but with the 5 i don t. Do you think i can choose the optimum lag 20? I just need to have cointegration in my data for a project... Thank you so very much!
Hi Georgina, read up on the underlying assumptions for LR and AIC to know if your data conforms to using either after which you decide. I know AIC/SBIC are efficient in small small samples but don't know about LR.
I only have annual data but the lag length criteria selected 5 lags as the optimal - then when I proceed I get 5 cointegrating equations. Is this strange for annual data? and does this imply all 5 of my variables are cointegrated? Thank you.
Thanks for sharing. Really helps! I am a beginner in VECM. May I ask one question? There is 2 cointegrating equations but there is no granger causality at all. In this case, is it meaningless to build the VECM model? Thanks in advance :)
Hi Keron, from a 2-sided t-test, a computed figure of 1.96 is significant at 5%. Any >2 is significant at 1%. Please may I know from where you are reaching me?
I have non cointegration data but when i change trend specification to be anything except beside constant i find cointegration, how to determine trend specification?
Thank you mam... Mam i got long run casuality in Johannessen Co integration test with 3 independent variables... But while performing the Wald tests diagnostics, i got only one of the independent variables out of 3 to have casual short run relationship. Does that affect my results? And mam, how to draft final estimated long run equation after johannessen co integration test ? Waiting for your reply at your earliest... From Nepal, South Asia.
I heard you emphasised that it should be done if the serries are stationary at first difference and definitely not second difference. If the stationary is stationary at second difference what method is recomended
Hello ma'am, I have a question, while find out the optimal lag length of the underlined variables if it comes 1 then at the time time of performing VECM we will take (p-1) lag then accordingly it becomes 0 so what should I do next..??? Please do the needful.. Thanks..
Hi, thank you for nice video. I have a question about when you perfom Johansen Test. Why didn't you specify only 1 lag instead of using 2 lags since the test itself tells that lag interval is in first differences.
@@CrunchEconometrix The weakness of Johansen approach is that it is sensitive to the lag length. So, the lag length should be determined in a systematic manner. Following is the normal process used in the literature. 1. Choose maximum lag length "m" for VAR model. Usually, for annual data this is set to 1, for quarterly data this is set to 4, and for monthly data this is set to 12. 2. Run the VAR model in level. For example, if the data is monthly, run the VAR model for lag lengths 1,2, 3,....12. 3. Find the AIC (Akaike information criterion) and SIC (Schwarz information criterion) [ there are also other criteria such as HQ (Hannan-Quin information criterion), FPE (Final prediction error criterion) but AIC and SIC are mostly used) for the VAR model for each lag length. Choose the lag length that minimizes AIC and SIC for the VAR model. Note that SIC and AIC may give conflicting results. 4. Finally, you MUST confirm that for the lag length you selected in step c, the residuals of the VAR model are not correlated [use Portmanteau Tests for autocorrelations]. You may have to modify the lag length, if there is the autocorrelation. Usually, beginners in time series econometrics tend to skip step d. 5. For the cointegration, the lag length is the lag length chosen from step d minus one (since we are running the model in first difference now, unlike in level when we used VAR to decide the lag length). (Culled from Cross Validated site)
Hello mam. All your videos are very good, easy to understand and very much helpful. Can you please guide me on how did you get the t-critical value of 2 to check for significance of normalized cointegrating coefficients?
Hi, After selecting a lag length and proceeding to cointegration test, I obtained positive results under max eingen stats but not under trace stats. must I obtain positive or negative consistently? How do I proceed from here?
I have also read that in ARDL model the dependent variable should be I(1) and I am running the ARDL model and the dependent variable is of I(O) and the problem which I am facing from last one month is that ECT is -0.99 which means its is approx 100 % which should not be the case. Can we have some other test where all variables are of Io order for long term integration
Hi Ramandeep, my videos on VAR and VECM are well explained. Kindly watch them again and consult relevant references listed at the end of the video. Thanks.
@@CrunchEconometrix I have already published some research paper using your video's. Now I was working on my phD, it could only be possible because of you.
Hello Madam.. What kind of data should be taken for johansen integration test because I have tried so many different set of data and it indicates no integration test. Thanks in advance
Hello ma'am, I have a question, while find out the optimal lag length of the underlined variables if it comes 1 then at the time time of performing VECM we will take (p-1) lag then accordingly it becomes 0 so what should I do next..???
Mam, at 2.34 you say that if the test statistics is greater than critical value we have to reject the null hypothesis.. but I have found in the eviews page the other way. www.eviews.com/help/helpintro.html#page/content/advtimeser-Unit_Root_Testing.html
Hi Asha, results interpretation is the same across all techniques. Reject the null hypothesis if the test statistic is significant...if greater than the critical value.
Good afternoon Ma'am. Thanks for reply. I have a question. i am analyzing my work with panel data of 28 years with five different countries. May i use VECM model in it ? Kindly advice me. Thank you.
Hi Alfred, once you have I(2) series the Toda-Yamamoto technique is applicable. Read more about it because I've not had cause for it. Thanks for the positive feedback on my video. May I know from where (location) you are reaching me?
I have also performed this test on my data with three variables with optimal lag selection based on AIC. All variables are stationary with the same order at the 1st difference. My results are highlighting 3 cointegration equations with significant p values. But, unfortunately, "normalized cointegration coefficients" of my data are showing insignificant results by calculating the t-value with help of the standard errors. I need your suggestions to solve this problem.
Rashid Latief Re-run these JCTs: lny lnx lnz; lnz lnx lny... Compare the outcome. The VAR model is sensitive to the arrangements of variables in the system.
Thank you for the quick response, I put lnx at first, as the target variable, and rotated the lny and lnz re-run it. but the "normalized cointegration coefficients" don't significantly change. Can i ignore "normalized cointegrtion coefficents" and elaborate the main results of trace value and max eigenvalue as those are significant?
Rashid Latief The normalized cointegrating equation represents the LR equation which from your results is not significant. But b4 we outrightly conclude or suggest you change variables (using better proxies), go ahead and perform the VECM but impute 1 cointegrating equation just to simplify interpretation of the results...then see the statistical significance of the ECT.
I am very thankful for your guidance. I have moved to VECM model with 1 cointegrating equation, ECT of one of the three models is significant. It has also showed short-run relationships. It has also an evidence from the literature.
Hi man love your videos, I am a master student at University of Antwerp in Belgium. I have a question: My dependent variable is I(1) and my independent variables are I(0). What do I do?
Is it possible to analyze this if one the series is stationary at level and the other is I(1)? I mean can you analyze this if you have to differentiate one of the series?
hello, thanks for the video very much! Ir really helps me a lot for my project. There's a little question about step 5, why we specify VECM with (p-l) lags instead of just using p lags? The ect estimation is different from what we get in step 3.
I have a question ... if some of my series are stationary I(0), can I do my "VECM"? Because I have three variables I(1) and one variables I(0) Greetings from Chile
Thank you ma'am for your very good and qualitative videos. Great job!!! I'm a bit confused with your response here. In line with one of your previous videos, I performed a Bounds cointegration test for a mix of I(0) and I(1) variables and established cointegration. Wont the next step be determining short run relationship using ECM, I'm assuming it's the same as VEC. Also I'm studying the impact of pension funds on capital market development using 5 key development indicators. Will it be right for me to construct 5 models with pension funds as the independent variable and each development indicator as dependent variables
@@temiolanipekun6579 Thanks for the positive feedback on my videos, deeply appreciated! Remember that unlike VAR, ARDL is a SINGLE-EQUATION model. I only created a vector to show that such an approach can be explored by a researcher.
Hello Ma'am, Can we conclude that the normalized cointegration coefficient measures the exact long-run relationship between the variables? Secondly, normalized cointegration coefficients are insignificant after calculating the t-value. I made sure my specification of the variables for the test is correct. Is there a solution here?
Hi Perry, yes we can say that because cointegration implies that the variables move together over the long-run horizon. It will be good if the coeffs are significant but if otherwise it simply shows that though the series move together in the long-run such movement is not significant, whatever that means. May I know from where (location) you are reaching me from?
RUclips 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.
You have really helped me a lot. I gained many things i initially didn't know
Thanks for the encouraging feedback, Gasana. Deeply appreciated! Please may I know from where (location) you are reaching me?
And now u have 24k...well done man 😎
can you please tell me shortly what is the difference between ECM and VECM?
Not all hero’s wear capes, watched your videos undergrad and currently for my masters. God bless you
Hi Ikhide, I'm encouraged by your positive feedback. Deeply appreciated 💖
You are one of the best channels I have ever seen about econometrics... regards from Madrid, Spain!
Wow, thanks Haddo!
The best econometrics channel I have seen. Simple and straightforward explanation. With millions of applications from Zambia 🇿🇲
Thanks James, for the encouraging feedback. Deeply appreciated!
Very good work, well explained! Thank you very much for this. I have been looking for such an explanation since a long time!
U're welcome, Boris😊. May I know from where (location) you are reaching me?
@@CrunchEconometrix From France Ms Bosede.
@@borisogou4756 Awesome! Kindly spread the word about my RUclips Channel to your friends and academic community in France 🇫🇷 for awareness. They'll learn some useful tips and hints too. Thanks 😊
@@CrunchEconometrix Well understood Ms! Thank you!
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, I appreciate your watching my videos but your query is TOO LONG. Kindly go straight to the point. Thanks.
I have learnt a lot from this channel and am grateful to the author
Thanks, Joseph for your encouraging words. Deeply appreciated 🥰🙏
Hi, I have three variables 1) Price 2) harvested Area 3) Yield rate of rice. I am trying to test if there are relationship of each variables between each other so that i can draw some conclusion regarding whether stabilise the rice price can help to stabilise the yield/ harvest area.
I tried to take a log of these three variables in order to better understand them in % change. not that they are skewed.
I then test the stationarity by ADF, showing that only ln (Yield) is stationary. I then perform the commond differencing and test again and so on. I finally got ln (Price) after one-time differencing (ln(1)) and ln (Harvest Area) after two time differencing (ln(2)).
I then also conducted the Johansen integration test on the variables ln (Yield), dln (Price) and dd ln (Harvest Area), the result shows that they have cointegration relationship for at least 2 variables. Cointegration test result here enter image description here On the final part, I want to fit those three variables into Vector error correction model.
The problem is here, in order to fit the VECM, all variables should ideally be integrated of the same order, typically I(1) which is not my case here.
if that is a absolute requirement? can my case use VECM? if no, then i should go for VAR instead? or you have other suggestions?
Please try to be very brief. Depending on the outcome of your unit root test AND your research objectives, you can decide between ARDL-ECM or VAR-VECM techniques.
I cannot thank you enough. You saved my future.
Hahahaha Omar, it's the least I can do. Glad to be of help. I will appreciate if you can share the link to my YT Channel with your colleagues and friends on social media. They can gain a lot too. Thanks!
@@CrunchEconometrix
This reply is a month late, so accept my apologies. I've already shared a link of your channel with about 250 students of my colleague. Your content is really valuable and extremely helpful. Your way of explanation makes econometrics an easy task. Sincere Thanks.
@@OmarAhmed-cm3mv Better late than never...thanks Omar! 💕 😊 💃
@@CrunchEconometrix Mine too Professor 💖
Thank you for a wonderful tutorial. Waiting for the (EViews10): Estimate and Interpret VECM (2)
Hi Moses, thanks for the encouraging remarks. I have just uploaded the part 2. I'll appreciate if you share my videos and links. People need to know that econometrics is not as difficult as it seems.
Thanks. I have seen it and its very insightful.
Quick Questions:
1. If i have a mixture of I(0) and I(1), can i estimate VECM or i need all of them to be I(1) only.
2. Is there a way to estimate VECM under ARDL model (just like ecm in ARDL)
thanks
Moses
Moses Mathu On your queries: (1) answer is no. All series must be I(1); (2) yes it's possible. That's the approach I explained in all my ARDL videos...please watch them starting with the video "this is how to specify ARDL models". Thanks again for the encouragement, deeply appreciated. Don't 4get to share😉
Thank you. the videos are really helping me (an econometric beginner). clarifications. I have 5 variables : inflation (dependent variable I(1), Money supply I(1), exchange rate I(0), deficit I(0), rgdp I(1). am interested on effect of deficit on inflation).
due to endogeneity, i have also added two more equations with deficit and ms as the dependent variables.
q1. do i conduct contegration test on I(1) only or on all variables
q2. in the cointegration test, do the I(1) enter in levels or differenced
q3. Assuming i get cointegration in 3 eqns, my understanding is that i run ECM for each eqn (separately as you explained).is this right? q4. again, do the I(1) variables enter the ecm at level or differenced.
Apologies for the long post but i need to be clear. thanks
Thanks for letting me that my videos are helping you. it's my joy doing what I do. But, since the variables are integrated of different orders, I hope that you are estimating ARDL-VECM and not VAR-VECM? On the queries on cointegration, the answers are in the 2nd part of this video (How to estimate and interpret VECM (2)), so kindly watch it. Please assist me to share my videos to your colleagues and social media community they need to know that econometrics is not as difficult as it seems. Thanks!
Great video. The other EViews videos are also worth watching.
There is an All Unit Unit Root test add-in in the toolbar. It saves time; you can use it too.
Thanks a lot for your incredible videos
Thanks for the encouraging feedback, Abba. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix From Turkey! Your videos are being really helpful. I came across your blog too, Great work!
Great explanation, big thanks
Thanks for the encouraging feedback, Annur... deeply appreciated! 🙏
Perfect tutorial dear Dr
Thanks Dr. Alnour for the encouraging feedback. Appreciated!
can you please tell me shortly what is the difference between ECM and VECM?
Hi Abbas, you will need to grab any textbooks for detailed differences between ECM and VECM. You can then support your reading with my video tutorials that show how to estimate both. Thanks.
Wonderful lecture keep up .
Thanks, Kitesa!
Thank you very much for your helpful video. Now I have a defense for what I did using VECM, but I do not know why we interpreter the long-run coefficient reversely. please, help what is the reason behind? Thank you.
Hi Zebider, thanks for the encouraging feedback. Deeply appreciated! The LR equation is in IMPLICIT form. Move the explvars to the RHS...justifies the REVERSE interpretation.
Dr, what's "Positive or non-negative argument to function expected" means in heteroskedasticity test? is it bad result? should I reduce the lag? Thank you Dr.
Kevin, I have not encountered such a message.
Hello, please, if we have trend in some series, we must correct it and then check if the series are stationary in the first difference?
second question please, we must check the stationarity in the 3 models(intercept/ intercept & trend/None) or only for one model? thanks a lot for your times.
Hi Zakaria, kindly watch my videos on stationary test using the Augmented Dickey-Fuller. Thanks.
Thanks a lot, very well explained... life saver
Hi Kyriacos, I'm encouraged by your positive feedback. Grateful! Please may I know from where (location) you are reaching me?
I love your explanations
Thanks for the positive feedback, Muyiwa. Grateful! 🙏
Thanks a lot ma'am. You are helping the researchers like me. we are highly benefited by your lectures. Ma'am, I want to ask one thing can apply Cointegratioan and VECM if the variables are stationary at level (Means both are I(0).?
I have responded to you on this on a different thread, thanks.
Hi Dr. Adeleye, this is Hussain from Saudi Arabia. thanks alot for your helpful videos. I really appreciate your efforts.
I have a question, I am also using quarterly data (but just 40 observations). What’s the maximum number of independent variables I can use? How to determine it?
Thank you 🙏🏼,
Hi Huss, choosing indpvars suitable for answering your research questions is at your discretion. It is better to read the literature to see what obtains.
Question: If all of my variables are not significant after doing the Johansen test as in 10:05 of video, should I still go ahead with the VECM model?
If there's a cointegration relationship, you proceed to VECM. You execute VAR if otherwise.
@@CrunchEconometrix Ok great.
When testing for cointegration between 2 variables, do i use the johansen test or the engle granger test, and which model do i then use to model my 2 variables. if there are videos on this please send the link, thank you.
Kindly watch my video on JOHANSEN COINTEGRATION TEST...check through my Time Series Playlist.
Do we need to estimate all the two equations in case of two variables in order to determine bidirectional and unidirectional causality
Kumari, I gave clear steps on what to do. Watch the videos.
You are great , thanks a lot.
You are welcome, Oyun 😊
after performing the test, trace test shows one cointegrating equation while the max-eigenvalue test indicates no cointegration. what to do when Result of trace test and Maximum Eigenvalue test differs?
Hi, Waleed the decision is yours to make. I mentioned that in my JCT video.
I love your videos, mam!
Thank you so much, Khyati!
Dear Madam, this is very much helpful video. Could you please let me know what can I do when my optimal lag length become as lag 1? Not as lag 2, its a trouble to analyze VECM in lag 1.
Hi Ishamali, retain the one lag.
Loved the explanation. I understood everything.I have one doubt though. How do we check if alpha is 0 in a restricted vecm?
Thanks Ann, for the positive feedback. Deeply appreciated! But I don't quite understand your question. Kindly rephrase.
@@CrunchEconometrix how can we apply restrictions to the coefficients?
Hi Ann. I have no idea. You may need to check out other online resources. Thanks.
Hello Ma'am, thank you so much for the video. I have a little challenge. What do I do when the optimal lag length is identified by the criterion, but when i try using it in the model, I get an error message stating ''insufficient number of observation''?
Several reasons. Too many variables, too many lags and/or a short time span. For instance, too many lags will reduce the number of observations. Reduce the lags to 1 and re-estimate the model, but make sure you have at least 3 observations BEFORE doing so.
Thank you DR; but Why signs are reversed in long-run? what is the implication?
Hi Nahimana, the results of the VECM are not reversed. Only that of Johansen cointegration test but it is in implicit form.
I have one query, if I have daily closing price of stock, why do take the airthmatic or log return for the series, why cannot we take the value of series directly as closing price and apply the test
Of course you can...only that the returns (log difference) is stationary. You can use the difference of the closing price too depending on your research focus. Using "returns" is not mandatory.
Really thankfull for the video and your explanation Mam. I have a question, in your video there is no list for VAR Stability test. Is it no need to check the stability because there is no Impulse Response Function and Variance Decomposition test? thank you so much mam
Rizka, your query is muddled up. Very confusing to understand. Kindly recast. Thanks.
@@CrunchEconometrix okay sorrry mam, so I've read some of journal and they are using VAR/VECM stability test and also IRF and VD test, so I want to ask..is it necessary to do those test?
And I have another question mam...My VECM model didn't pass the diagnostic test, is there any way to solve this problem? thank you mam
Rizka, I have videos on those areas you refer to. Kindly watch them to see what I did and adapt. Thanks.
good job
Thank you for all these videos ,I really need your help . It's better that I give brief about my case . I have one dependent variable Y and 7 independent variables . My aim is to forecast Y but i am not sure which model is more suitable. All series are non stationary and not normal at level .I took the logs and first difference for all series and I ran VAR eviews model . After that I tested the optimal lag and based on AIC it was 2 . Then i tested for co-integration using 2 lags . There was two cointegrations defined . As far as my understanding if the variables are stationary at first difference and there is co-integeration , I shall go to VECM model with (p-1) . Once i run the VECM model , the residuals are not normal, serially correlated and has heteroskedasticity. What shall I do in this case ?
Hi Ahmad, yes such can happen. You may need to reduce your regressors and re-estimate. If the problem persist, estimate the model at higher lag levels.
@@CrunchEconometrix Thanks , I hadn't checked for Multicollinearity before . Now i did and i found two regressors with high VIF .will eliminate them and try again. Will let u know .
May you please advise how may I reverse the log function . I took the logs of all series but now after running the model and forecast , I would like to interpret the results by seeing the real numbers .
Ahmad, having taken the log, you have to use the "elasticity" interpretation.
@@CrunchEconometrix can you please explain . Guide me If it's there in one of your videos .Thanks
ma pls if we have different order what is the next step to take to test for co integration
Hi Moradeyo. ARDL-Bounds test.
very good
Thanks, Toufi for your encouraging feedback. Deeply appreciated 🙏🥰
Thank you very much for the good content! I was wondering, why do you opt for AIC when you conduct the unit root test in the beginning? Is there a specific motivation for you to do so? Additionally, if you were dealing with daily data (5 days a week), what lag length would you recommend? Many thanks
Thanks, Owen for the encouraging feedback. Deeply appreciated. The researcher decides on which information criterion to use. I will advise you watch my clip on "Optimal Lag Selection" for background guides to selecting lags. Kind regards.
This is very informative
Thanks for the encouraging feedback, Onesmus. Deeply appreciated! Please may I know from where (location) you are reaching me?
Thank you so much for this extremely helpful video! I just wanted to ask if I have 2 cointegrating relationships how should I interpret the long-run equation? and which cointegrating equation to use?
Hi Sara, kindly watch my video on Johansen Cointegration. I gave clear directions on what to do. Regards.
evening ma, thank you for all the videos. There have all been helpful. Please Ma, what do we do about I(2) variables ?
Hi Okputu, drop it and replace with a closer proxy.
@@CrunchEconometrix okay Dr. thank you very much
Respected Ma'am , The researchers are getting too much benefits from your lectures. You explain the difficult things in a very simple manner. thanks a lot for guiding us in econometric techniques. . Ma'am, I want to ask one thing that can we apply cointegration and VECM (If Variables are Cointegrated) on the variables which are stationary at level, means I(0)
?
It's my pleasure, Suresh. Thanks for the encouraging feedback. If all the variables are I(0) proceed to perform just OLS. I have a video on that. Search through my Channel and watch. Kind regards.
A Bundle of Thanks for replying Ma'am.
Can we apply VAR model also on two I(0) series?
In my paper, I have applied VAR model, Wald test for joint significance, Granger causality test, Impulse Response Function and Variance Decomposition model on the panel data of two variables both of which are I(0). M I going right?
I need your expert suggestion. Please help.
Kindly watch my videos on VAR. Well explained. Thanks.
Thank you very much for this video. It helped me a lot with my study. Can you please help me to know how we can know that it is significant at 1%? I understood how to get the t-stat but I don't understand what value to compare with.
Thanks, Jean for the encouraging feedback. Deeply appreciated! I will refer you to read about hypothesis testing and confidence intervals from any econometrics textbook for deeper understanding of statistical significance.
Dear Mam, thank you so much for this informative video. It helped me a lot. However, I have a few questions and I will be really grateful if you help me with the answers. In Johansen Normalitaization Interpretation, what percentage should be compared for testing significance of coefficients at 5% level, and will it be statistically significant if the t-statistics is greater than z value both for plus and minus? Thank you again.
Hi Christina, I gave clear explanations on the decision criteria. Also, kindly watch my video on Johansen Cointegration. Thanks.
Maximum lag at four for unit root test because you have a quarterly data. But what if I have monthly data?
Kindly watch my video on OPTIMAL LAGS SELECTION. Detailed to guide you, thanks.
i noticed that you have gdp and ingdp, is ingdp log(gdp) ?
lngdp not ingdp
Thank you very mich for the video
I have two questions:
1. Does the Normalized cointegrating coefficient interpretation replace the Granger Causality test in the case VAR and VECM (for short and long run).
2. If so, should we consider the same method to interpret the Normalized coefficients which is reversing the sing in the case of short run ?
If what I have mentioned is possible... I highly appreciate if you share with me some references to support my analysis
Best regards
Hi Hechem, NC and GC are not the same. Please watch my videos again on these procedures. There are tons of papers on the internet that you can download, read and reference.
Dear Ma'am, I applied the given steps while doing my estimations on my time series dat of 40 annual observations. After estimating the VAR in computed the optimal lag length which is 5. Now doing the Johnson coIntegration test , when I am putting the lag interval 1 to 5, it give an error "Near singular matrix". What should I do to eliminate this error in order to proceed further.........................................Moreover although the optimal lag length is 5 (which in this case is not helping me to further proceed my estimations) so i put arbitrary lag length of 3 and with this arbitrary lag length the Johnson coIntegration test is not giving any error.
Concluding it, my question is " If the estimation with the optimal lag length is giving an error, can we apply arbitrary lag length for conduction of Johnson coIntegration test ?............. Provided that, my observation is betwen 37 to 40 (the data is annual)
Hi Tariq, kindly shorten your query. Too long didn't read (TLDR).
@@CrunchEconometrix Dear Professor, How can I interpret my results?
1) ARDL-Bond test:Cointegration Confirmed
2) Model: ARDL-ECM through OLS
3) Probability of coefficients for ECM and all independent variables is higher than 0.05.
Thank you!
Tariq, I give interpretations in my videos. I will advise you watch them and adapt to your results. Thanks.
Thank you for a nice video, waiting for asymmetric test in EViews 10
Thanks Sharif, for the positive feedback. I'll do my best to make the video. Please may I know from where (location) you are reaching me?
Why didn't you use the first difference ?
Hi Axel, you can use the 1st difference. There are 2 schools of thought. I mentioned that in the foundation video.
@@CrunchEconometrix I didn't have seen your response, i'm sorry. Thanks for you, i get it now.
I'm from Côte d'Ivoire, the greatest french speaker country of western Africa 😂lol
What if in the normalized cointegration equation these coeff divided by the standard errors give 1.18 for the first variable & -3.53 for the second variable? Also: there is no cointegration, so should this normalized cointegration equation be interpreted?
By now you should understand your t-stat and give appropriate interpretation same for if there's no cointegration. You get better when you read textbooks and articles to support video tutorials.
i love all your videos, they are very easy to follow. i am looking for a video that estimates and interprets the VAR model in eviews
Thanks Zama. Kindly click on the Playlists for EViews Time Series videos to watch those on VAR.
Thank you very much for the videos, they make econometrics doable. I am trying to estimate fiscal multipliers and i have variables gdp, government expenditure and tax. Is it ok to put gdp as my target variable? another question; is it possible for the results of trace and eigenvalues to differ when doing co integration test? if so which ones do i take?
U're welcome, Lydia. CrunchEconometrix was set up to make econometrics crunchy, delightful and fun. To your query, please watch my video on Johansen Cointegration Test and you have your answers....thanks once again for watching my videos. Please tell others too :)
Best VECM video EVER
Thanks for the encouraging feedback, Mohd...deeply appreciated!
@@CrunchEconometrix I know right! Jesus i only understood this thing because of her
Madam, At the minute 11.49 you say that the t-statistics are clearly above 2, Why 2 ma'am ?
Hi Manoj, using the 95% CI.
Hello ma'am, is it permissible to conduct the Johansen Cointegration test on RAW DATA, and then if I find cointegrating relationship in the results, can I conduct VECM on LOG TRANSFORMED DATA for better interpretation of results? Or is it necessary to perform the Johansen cointegration test on log-transformed data if I intend to perform the VECM test on log-transformed data? Thank you so much for your videos and assistance. They have literally helped me complete my research work!
Hi Saima, Yes. I explained this in my video on the Johansen Cointegration test. I'm advising you to watch the video for more information. Thanks.
@@CrunchEconometrix Thank you so much for your reply ma'am. I watched your tutorial on Johansen Cointegration test. But I think I couldn't explain my query clearly. Ma'am my confusion is that suppose I apply the Johansen Coint. test on RAW data that is not log-transformed and I find a cointegrating relationship exists, then can I use log-transformed data to conduct the VECM test or will I have to use raw data for the VECM test because I used raw data for the Johansen Coint. test? So, In case if in my study I use Raw data ( not transformed into log) for Johansen test and then Log transformed data for VECM then will this affect the reliability/accuracy of my results?
You are on track. Go ahead.
Dont we open VAR at first differences to find the appropriate lags?
Watch my videos on VAR. Thanks.
Dear CrunchEconometrix can you make a tuto on how to perform structural VECM. Thanks
Hi Omalams, thanks for the suggestion. Noted. Will do so once I understand the procedure. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix I'm reaching you from China. Thanks
Thank you Prof Ngozi, you said 7.23 is above 2, but what is the rule of thumb when identifying the p value. thank you
Hi Darcy, please can you be more explicit? I didn't quite get your query.
CrunchEconometrix hi Prof, i think Darcy said in this video at ,7:23
You mentioned that the t statistic is above 2, and it is significant.
Thus, what is the rule of thumb in identifying the p-value and its significant? How do we know that it is significant.
@@meditatewithme4933 already an answer for it?
StudentEnjo hi yes. Rule of thumb, if your t statistic is more than 2 or less than -2 it’s significant.
@@meditatewithme4933 I remember, it is basic t stat knowledge. You can easily check on Pvalue from t stat, with degree of freedom: observations - number of variables
Hello Crunch, Hope everything is going well! I have a question: If my variables (1st diff of log_youth unemployment and 1st diff of log_GDP) are not cointegrated... I can't do the VECM right? Should I do the unrestricted VAR? What do you mean with unrestricted VAR? Typical VAR Model? Thank you!
Hi Haddo...it's YES to both questions. Thanks 😊
@@CrunchEconometrix Thank you very much Crunch! I am working on the problem of Youth Unemployment and I am modeling a VAR Model with 4 Lags (the best based by Lag Length Criteria) but I don’t know if an ARDL Model could be better... I understand that VAR Model has dynamic (and Impulse Responses also)... Am I right? Should I stay with my VAR or maybe the ARDL is also a good option? Thank you Ma’am! 🙌🏻❤️
Haddo, choosing between VAR and ARDL depends on your study objectives which I don't get involved in.
@@CrunchEconometrix Okay, thank u!
hello mam, what will be lag if data is annually.
Hi Anupam, kindly watch my video on "Optimal lag selection". Detailed to guide appropriately.
ok mam, Thanks for your guidance
Thank you for your excellent presenation. if i the log variables on thier level not all of them non staionary while in thier nature (without log) level all of them nonatiaonary, can i apply their logarethim? if i find the varaible are not normal distributed while other statistics are fine (no serial correlation no heterscadisicty) can i accept the model? i got the following results LKSEI(-1) 1.000000 0.000000
LOG(TT(-1)) 0.000000 1.000000
LHINDEX(-1) -0.154923 -1.038932
(0.25311) (0.08584)
[-0.61208] [-12.1034]
Hi Jasem, there are no rules to taking logarithms. It is at the discretion of the researcher. But if the variables are nonstationary, take their 1st difference. Also, I will give a model a good pass if it only suffers non-normality but has no evidence of serial correlation or heteroscedasticity.
i appreciate that so i want to know how did you get the number 2 for testing the significant coefficient
thank you in advance
That is the standard guide. Kindly read up hypothesis testing, probability value and T-stat from any basic econometrics textbook.
Hi Cruncheconometrix, It is worth of watching and i have a query on optimal lag length, Can we take lag length first difference 1(1) or level1(0)?
Hi Mahabub, thanks for the positive feedback on my videos. Deeply appreciated! You take lag of the MODEL not variable.
what we do if the long run coefficient of the variable is not statistically significant but they have con integration and the speed of adjustment is statistically significant? specially in the case of exchange rate devaluation and trade balance if the coefficient of real exchange rate is not statistically significant can we say there is long run causality from real exchange rate to trade balance in the long run? in VECTOR ERROR CORRECTION MODEL.
Debela, the H0 is no causal relationship. With insignificant coeffs, H0 cannot be rejected. However, a significant ECT evidences joint causality from the regressors to the outcome variable. May I know from where (location) you are reaching me?
here from Ethiopia but is there any theoretical support that devaluation may not improve trade balance in the long run ?
@@debelagelana6408 You'll have to do some readings on that. You might find papers with such findings.
thank you so much
@@debelagelana6408 U're very much welcome! 💕 😊
Dear Professor,
I need your help. I have monthly data for unemployment rate and for the number of persons who checked-in in turism units. The lag length that i have chosen is 24, and i obtain that the optimum lag for LR is 20 and for AIC is 5. With the optimum lag 20 i have cointegration, but with the 5 i don t. Do you think i can choose the optimum lag 20? I just need to have cointegration in my data for a project...
Thank you so very much!
Also, with the optimum lag 20 my errors satisfy the 3 hypothesis.
Hi Georgina, read up on the underlying assumptions for LR and AIC to know if your data conforms to using either after which you decide. I know AIC/SBIC are efficient in small small samples but don't know about LR.
@@CrunchEconometrix Thank you for your answer and all the knowledge you share with us! Wish you all the best!
Hello, if I want to know whether the macro variable (LNGDP) is cointegrated with LNPDI, how can I tell it?
Watch my video on Johansen Cointegration.
I only have annual data but the lag length criteria selected 5 lags as the optimal - then when I proceed I get 5 cointegrating equations. Is this strange for annual data? and does this imply all 5 of my variables are cointegrated? Thank you.
Hi Ellie, as I explained in one of my VECM videos use only 1 cointegrating equation to keep it simple and for easy interpretation.
Thanks for sharing. Really helps! I am a beginner in VECM. May I ask one question? There is 2 cointegrating equations but there is no granger causality at all. In this case, is it meaningless to build the VECM model? Thanks in advance :)
Thanks for the encouraging feedback, Yihe. Deeply appreciated! Go ahead and estimate the VECM. I advise you use 1 CE to keep explanations simple.
Can we include only two variable one dependent and another independent to measure co-integration via johnson test?
You include ALL variables to perform the JCT.
Hello ma'am,
Could you please provide video on fully modified ols run on time series data..
Hi Iti, suggestion noted. Thanks.
Good day, Why did you say that the values are clearly above "2" which makes it statistically significant at the 1% level. How did you select "2"?
Hi Keron, from a 2-sided t-test, a computed figure of 1.96 is significant at 5%. Any >2 is significant at 1%. Please may I know from where you are reaching me?
I have non cointegration data but when i change trend specification to be anything except beside constant i find cointegration, how to determine trend specification?
I responded to you on the other thread. Please stop repeating same questions across different videos. Not nice.
I just want to make sure you see my comment, sorry if it bother you
It's alright. Duplicating causes confusion. Once you post, I will definitely see it. Thanks for watching. Grateful!
Thank you mam...
Mam i got long run casuality in Johannessen Co integration test with 3 independent variables...
But while performing the Wald tests diagnostics, i got only one of the independent variables out of 3 to have casual short run relationship. Does that affect my results?
And mam, how to draft final estimated long run equation after johannessen co integration test ?
Waiting for your reply at your earliest...
From Nepal, South Asia.
Hi Aditya,
Thanks for watching my videos but how did you get casualty from Johansen cointegration test?
I heard you emphasised that it should be done if the serries are stationary at first difference and definitely not second difference. If the stationary is stationary at second difference what method is recomended
Hi Ephraim, the Toda-Yamamoto is then applicable.
Hello ma'am,
I have a question, while find out the optimal lag length of the underlined variables if it comes 1 then at the time time of performing VECM we will take (p-1) lag then accordingly it becomes 0 so what should I do next..???
Please do the needful..
Thanks..
...and what is the needful?
Prof do we divide the Coefficient by statistics or standard error, is the figure (0.64496) in brackets the t statistics or SE? . thank you
Hi Darcy, to get the T-stat: coeff/std.error.
@@CrunchEconometrix thank you Prof
do we have to transform the data to natural logarithm? can we use raw data?
You make the choice.
CrunchEconometrix thank you for the reply
Hi, thank you for nice video. I have a question about when you perfom Johansen Test. Why didn't you specify only 1 lag instead of using 2 lags since the test itself tells that lag interval is in first differences.
Hi Panun, the procedure is that Johansen is estimated with the optimal lag length (p) and not with (p-1) lags.
This was given to us in step 2. We thus have no latitude over it.
@@CrunchEconometrix
The weakness of Johansen approach is that it is sensitive to the lag length. So, the lag length should be determined in a systematic manner. Following is the normal process used in the literature.
1. Choose maximum lag length "m" for VAR model. Usually, for annual data this is set to 1, for quarterly data this is set to 4, and for monthly data this is set to 12.
2. Run the VAR model in level. For example, if the data is monthly, run the VAR model for lag lengths 1,2, 3,....12.
3. Find the AIC (Akaike information criterion) and SIC (Schwarz information criterion) [ there are also other criteria such as HQ (Hannan-Quin information criterion), FPE (Final prediction error criterion) but AIC and SIC are mostly used) for the VAR model for each lag length. Choose the lag length that minimizes AIC and SIC for the VAR model. Note that SIC and AIC may give conflicting results.
4. Finally, you MUST confirm that for the lag length you selected in step c, the residuals of the VAR model are not correlated [use Portmanteau Tests for autocorrelations]. You may have to modify the lag length, if there is the autocorrelation. Usually, beginners in time series econometrics tend to skip step d.
5. For the cointegration, the lag length is the lag length chosen from step d minus one (since we are running the model in first difference now, unlike in level when we used VAR to decide the lag length).
(Culled from Cross Validated site)
Thanks for sharing!
I love the way you teach! my question is what if my optimal lag is 1? should it be lag 0 when doin VECM?
Thanks for the encouraging feedback, Rifki. Deeply appreciated! In that case, maintain the one lag and put a note on it in your manuscript.
@@CrunchEconometrix Your welcome, Dr.
what if I have I(2) and I(1) variables? they can't be cointegrated, right? should I stop the study right there or do anything else?
Change the I(2) variable.
Hello mam. All your videos are very good, easy to understand and very much helpful.
Can you please guide me on how did you get the t-critical value of 2 to check for significance of normalized cointegrating coefficients?
Hi Bowan, I will refer you to any basic econometrics textbooks to read up "HYPOTHESIS TESTING and CRITICAL VALUES" for better understanding...regards.
Hi,
After selecting a lag length and proceeding to cointegration test, I obtained positive results under max eingen stats but not under trace stats. must I obtain positive or negative consistently? How do I proceed from here?
Hi Denis, positive and negative results? I have no idea what you are referring to.
hello Madam, can the series be of Order I(O) for running VECM
I have also read that in ARDL model the dependent variable should be I(1) and I am running the ARDL model and the dependent variable is of I(O) and the problem which I am facing from last one month is that ECT is -0.99 which means its is approx 100 % which should not be the case. Can we have some other test where all variables are of Io order for long term integration
Hi Ramandeep, my videos on VAR and VECM are well explained. Kindly watch them again and consult relevant references listed at the end of the video. Thanks.
@@CrunchEconometrix I have already published some research paper using your video's. Now I was working on my phD, it could only be possible because of you.
You are so destined for the top, Raman. I give a million likes for this encouraging feedback. Well done!
Hello Madam.. What kind of data should be taken for johansen integration test because I have tried so many different set of data and it indicates no integration test.
Thanks in advance
Time series data. Quarterly, monthly, weekly, or yearly.
Hello ma'am,
I have a question, while find out the optimal lag length of the underlined variables if it comes 1 then at the time time of performing VECM we will take (p-1) lag then accordingly it becomes 0 so what should I do next..???
Same doubt here.
Not an issue. Retain the 1 lag as VECM cannot be estimated as a static model and make a note in your work explaining what you did.
Mam, at 2.34 you say that if the test statistics is greater than critical value we have to reject the null hypothesis.. but I have found in the eviews page the other way. www.eviews.com/help/helpintro.html#page/content/advtimeser-Unit_Root_Testing.html
Hi Asha, results interpretation is the same across all techniques. Reject the null hypothesis if the test statistic is significant...if greater than the critical value.
Good afternoon Ma'am.
Thanks for reply. I have a question. i am analyzing my work with panel data of 28 years with five different countries. May i use VECM model in it ? Kindly advice me. Thank you.
Hi Hemant, if N < T as your current data structure, use panel ARDL.
Thank you Dr. for this tutorial. i have a question, if after unit root test, we obtain i(0) i(1) I(1) I(2). Which model should we use in our analysis
Hi Alfred, once you have I(2) series the Toda-Yamamoto technique is applicable. Read more about it because I've not had cause for it. Thanks for the positive feedback on my video. May I know from where (location) you are reaching me?
I have also performed this test on my data with three variables with optimal lag selection based on AIC. All variables are stationary with the same order at the 1st difference. My results are highlighting 3 cointegration equations with significant p values. But, unfortunately, "normalized cointegration coefficients" of my data are showing insignificant results by calculating the t-value with help of the standard errors. I need your suggestions to solve this problem.
Rashid Latief Re-run these JCTs: lny lnx lnz; lnz lnx lny... Compare the outcome. The VAR model is sensitive to the arrangements of variables in the system.
Thank you for the quick response, I put lnx at first, as the target variable, and rotated the lny and lnz re-run it. but the "normalized cointegration coefficients" don't significantly change. Can i ignore "normalized cointegrtion coefficents" and elaborate the main results of trace value and max eigenvalue as those are significant?
Rashid Latief The normalized cointegrating equation represents the LR equation which from your results is not significant. But b4 we outrightly conclude or suggest you change variables (using better proxies), go ahead and perform the VECM but impute 1 cointegrating equation just to simplify interpretation of the results...then see the statistical significance of the ECT.
I am very thankful for your guidance. I have moved to VECM model with 1 cointegrating equation, ECT of one of the three models is significant. It has also showed short-run relationships. It has also an evidence from the literature.
Hi man love your videos, I am a master student at University of Antwerp in Belgium. I have a question: My dependent variable is I(1) and my independent variables are I(0). What do I do?
I tested for cointegration & there are 2 cointegration equations. But I don't know if this is the procedure I should follow.
I'm sure you heard me mention that all your variables must be I(1) to estimate VECM.
See my response.
you have changed the lag to 4 because you had quarterly data .. please tell me how much lad I should use for the data that is on daily basis.
Hi Manish, watch my video on "Optimal Lag Selection"...and use up to 360 lags for daily data.
Is it possible to analyze this if one the series is stationary at level and the other is I(1)? I mean can you analyze this if you have to differentiate one of the series?
Hi David, you cannot. The most applicable procedure is the ARDL model.
hello, thanks for the video very much! Ir really helps me a lot for my project. There's a little question about step 5, why we specify VECM with (p-l) lags instead of just using p lags? The ect estimation is different from what we get in step 3.
is it because of df-related issue or something?? thank you
Check econometrics textbooks and you'll see that a VECM has p-1 lag length given that the dependent variable is specified in first difference. .
@@chengchelu9731 No
I have a question ... if some of my series are stationary I(0), can I do my "VECM"?
Because I have three variables I(1) and one variables I(0)
Greetings from Chile
Hi Alvaro, you can't. ARDL is your model. Love you from 🇳🇬 ...please share my videos with your friends, students and colleagues. Gracias! 💕 😊
Thank you ma'am for your very good and qualitative videos. Great job!!! I'm a bit confused with your response here. In line with one of your previous videos, I performed a Bounds cointegration test for a mix of I(0) and I(1) variables and established cointegration. Wont the next step be determining short run relationship using ECM, I'm assuming it's the same as VEC. Also I'm studying the impact of pension funds on capital market development using 5 key development indicators. Will it be right for me to construct 5 models with pension funds as the independent variable and each development indicator as dependent variables
@@temiolanipekun6579 Thanks for the positive feedback on my videos, deeply appreciated! Remember that unlike VAR, ARDL is a SINGLE-EQUATION model. I only created a vector to show that such an approach can be explored by a researcher.
@@CrunchEconometrix Thanks for the response
Can I reach you by email?
Hello Ma'am, Can we conclude that the normalized cointegration coefficient measures the exact long-run relationship between the variables? Secondly, normalized cointegration coefficients are insignificant after calculating the t-value. I made sure my specification of the variables for the test is correct. Is there a solution here?
Hi Perry, yes we can say that because cointegration implies that the variables move together over the long-run horizon. It will be good if the coeffs are significant but if otherwise it simply shows that though the series move together in the long-run such movement is not significant, whatever that means. May I know from where (location) you are reaching me from?
@@CrunchEconometrix I'm reaching you from India Ma'am. Thank you for the response
Perry Elorm U're welcome, Perry...please share my videos with your academic community in India and beyond!😀
@@CrunchEconometrix Sure I will your videos have been really helpful
in this case can i ignore the second varaible for the long run relationship thanks alot for your help
I don't understand what you imply. What second variable?
@@CrunchEconometrix i have varaibles second long relationship coefficient is 0 means no relationships thnx
professor, if the optimal lag length is 1, and we estimate VECM (p-1), would it be 0?
No, Djallel. In that case, maintain 1 lag because VECM cannot be estimated as a static model. Keep watching, keep sharing!👍🏽
@@CrunchEconometrix thank you so much Professor. İ really appreciate your help and your quick response.
Did you selected 4 arbitrarily for the lag length?
Yes, for the stationarity test. But optimal lag is used to estimate the model.