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, help me stay online.
@@tajudeenodetayo2333 Simply click on the SUBSCRIBE button showing in any of my videos or better still click on the SUBSCRIBE button on my TH-cam Channel and the BELL so that you can get my videos whenever I upload them.
Thank you so much for your explanation in the video!! I legit found the answer I was hoping for - to understand how lags work within the selection criteria of VAR!! Very good work and analysis, keep it up!!! THANK YOU again, God bless you!!!🙏❤❤
Hahahaha Adedeji, not the greatest just teaching the little I know😊. Thanks for the positive feedback, deeply appreciated! Please may I know from where (location) you are reaching me?
@@samueladedeji6883 Wow! The Great Ife!🙌 I'll appreciate it if you can share the link to my TH-cam Channel with your friends and academic community on social media. They will find the content helpful too. Thanks!
Selecting Optimal lags for Annual, Quarterly and Monthly data is a rare insight that comes only through experience. Thanks for providing this lag selection insights.
Madam, I had one question - I am running an ARIMA on Bitcoin Close price data with daily data of (12,12) when I select Max AR as 12. Is this still fine to have such high lag values considering I have daily data.
Wow! It's been three days and I already love econometrics, thanks to your extremely helpful materials. My question is on the way forward once have established that there is no co-integration.
Hi Margaret, thanks for the positive feedback and remarks on my TH-cam videos. Deeply appreciated! Depending on the outcome of your unit root test, you perform either a VAR or ARDL technique. May I know from where (location) you are reaching me?
hi thanks; but criteria for var give me 0 lags for all criteria how can i choose between 1 or 2 lags ? the 1 and 2 are both stable model with all tests( normality, AR, homoscedasity, LM) for VAR thanks
Hello, Professor Adeleye. Thank you for this informative video. I hope you can shed light to my problem. I am working on a panel data analysis, spanning 32 years (annual). I followed your instruction on finding the optimal lag, and the results show 8. However, upon integrating it in the regression, the model suffered a normality problem. I tried experimenting with lags for the dependent (GDP), and when I set the lag to 30, the normality and autocorrelation problem was fixed. May I use a lag of 30 or shall I continue with the optimal lag given by the lag length criteria results (8)?
Thank you so much for this video. If one of my series is stationary at first difference then we should use that series at level or at first difference while calculating lag for the model.
What a fantastic video to explain ma'am! May I kindly ask however, how would you choose lags for mixed-frequency data such as Quarterly And Monthly data?
I’m doing johansen cointegration and try to find optimal lag. My data are stationary at I(1), so in finding optimal lag, which type of data to use whether to use level data or 1st difference data? Thank you
Amina, this clip explains the lags to deploy with different time structures. You may want to watch it again and apply the information to your study. Thanks.
Dr Ngozi, can you make a video for censored and truncated regression models? I have a difficulty to understand these models, so I need your explanations.
Hi Patrick, thanks for the suggestions. Are you referring to TOBIT, LOGIT and PROBIT models? If yes, they are already on my to-do-list. Once I clear up my backlogs, I'll read up on them and create the videos. Thanks for believing in me. Keep watching, keep sharing too! :)
Thank you so much for your videos they have really helped us in our research and everything is clear. I just have a question regarding the optimal lag length, we are running an ARDL model on annual 30 observations and the optimal lag selection according to VAR was 5 is this ok or should it be reduced to 1 or 2?
Hello dr, im using daily data from 2018 to 2022 and i have hard time to finding optimal lag since alot of video explain quarterly/ monthly/ yearly data only . do you have suggestion on what the optimal lag for daily data ? my observation is 1443 data
@@CrunchEconometrix yes dr, I'm currently learning basic econometrics, so my prof gave me some data which is the number of observations 1443. This is for my assignment. I try to use 365 lag but the data is too small so I try to use 50 lag, and it shows my optimal lag for the data which is the smallest aic is 50. Do you have any suggestions for the lag range? I'm trying to figure out the lag because I want to do LM test in eviews. Thank you in advance !
Syazz, you said your data is 2018-2022. So, how come you have 1443 observations. I am not understanding. I suggest you ask your Prof for clarifications.
Hi madam, If the residual autocorrelation test results indicate that the p-value for the chosen lag is 0.000, which is less than 0.05, suggesting the presence of autocorrelation, should I change the number of lags and re-estimate the VAR model with this new lag?
@@CrunchEconometrix Thank you for your reponse. I have another question when I estimated the model ,i found the R-squared coefficient is low for both variables, even though I tested the model's stability ,and the significance test of the coefficients indicates a probability less than 5%, so the model is significant .Does this low coefficient pose a problem?
Hello, I'm using E-views 11 for student and there is no option for "Lag Lenght Criteria" in Lag Structure option. My data is an annual panel data of 27 countries with a total of 810 observations. Since I can not determine the length of the Lag, is it appropriate to go with the automatic option which is for annual is 1-2 lag? all data series are cointegrated at 1st difference, I applied Johansen and VECM with lag 1-2.. please give me suggestion. Thank you. Your channel is very educative and informative :)
Hi Andrian, thanks for the encouraging feedback. Deeply appreciated! You can use up to 365 days. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix thank you very much! I'm currently studying in Edinburgh and it's the first time I'm learning econometrics and we are asked to run an analysis that requires quantitative research method. Your vids are really helpful.
Thank you for your video, if I understand you correctlym does it mean that if AIC shows ,for example, to use 2 lags for regressor, then to make the model better we need to add lag1 and lag2 of this regressor to the model?
I have to thank you first for your helpful videos. I just have a technical question. You said we use unrestricted VAR to estimate the optimal lag length because we assume that the variables are not co-integrated. Why we assume "not co-integrated"? Thank you again.
Hi Mona, because the cointegration test has not been performed to establish that. Hence, determine the optimal lag and other steps follow. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hello to you. I am in Italy. I am a student and I am preparing a project with Eviews. If I may I would like to ask is it better to run the co-integration test or unrestricted VAR to find the optimal lag. Because I have run the co-integration test and it exist among my variables. But interesting thing is that when I run the co-integration test the lag that it suggest me is the same when I run the unrestricted VAR.
I am running ARDL ECM model here and which method should I apply to choose the optimal lag length? Including all variables in the VAR or run VAR on each variables ? what are the difference between including all variables in VAR and running VAR on each variables?
Hi there, for optimal lags selection for ARDL models, VAR is performed on each variable. But if a VAR model, all the variables are listed. Since, you are performing an ARDL model, I'll advice that you watch all my videos (Stata or EViews, check out the playlists) on how to perform the procedures from start-to-finish. My videos are quite easy to understand. Thanks and good luck with your research.
Suppose there are 4 variables and in cace of ARDL bound test approach we get 3 Cointegrating equation by F test. Now my question is what is the best equation to show short and long run impact and why? and how the lag of the equation like (1,0,2,1) or (2,1,0,1) is selected. Thnx
Hi Bikash, I covered quite a lot on ARDL models and the Bounds test. Find the time to watch them. Practice along, listen to the explanations and jot some notes. You'll get better if you do these.
Dear Professor, What is the difference between determining the optimal Lag Length for a model and a variable? Do they have different usage or different purpose? In the video the optimal Lag Length for the model is 2, but when you checked for the individual variable it showed different optimal Lag Length; one of the variables showed two optimal lag, the other four and the other one. Please explain the purpose and usage of determining optimal lag for a model and each variable.
Hi Shaf, as rightly said optimal lags for a model and for variables differ as shown in the video. This applies to ARDL modeling not VAR. But the process is simplified with EViews and Stata as these software can automatically select the appropriate lags for each variable during estimation. Thanks.
Hello Dr. I watched your video on ARDL estimation and specifically your demonstration on how to determine optimal lag. Its quite different from what you outlined here. My question is, what is the best way to determine appropriate lag structure.
Kenechukwu, while this clip gives a general overview on optimal lag selection, that of ARDL is specific to the estimation technique....and there is no "best way" as I explained it to be "an empirical issue" subject to several factors I have pointed out. I will also suggest that you read up other sources on how to determine optimal lags for deeper understanding.
Most times, the information criteria (AIC, BIC, HQIC) are in the negative. So, it is best to consider the absolute figures to avoid confusion in making a decision.
Dr. Ngozi, i have been following your tutorials and they are really helpful. Now concerning Lag order selection creteria, what should you do when the creteria dont agree on the lag. For instance, when AIC shows minimum lag should be 1 while SBIC shows the minimum lag should be 2??? kindly explain
great explanation. however, i have a question. now i understood how to select the lags, but you've done it separately for each variable. my question is, how should select the appropriate lags for the whole model? say THE DEPENDENT VARIABLE IS GDP= PDI+ PCE?
what if the model suing the obtained optimal lag suffer from heteroskedasticity? is it convent to take the lag after ? i dont want to take the log of my variables. Thanx
Oh sorry . I calculated the optimal lag which is found to be 1, then re-estimate the PVAR(1), no serial correlation found, BUT Heteroskedasiticity was obtained. I tested again on lag 2 and no issue found. But As far as i understand increasing lag must be based on Theory or previous studies which I couldn’t find. I applied the GLS but couldn’t retest the Heteroskedasiticity. Any suggestions?
Mam is it possible that in our model there is no constant and trend and can we select under trend specification the option"none" as all the options are giving insignificant result..meaning when I am estimating the model by selecting both constant and trend the p value for both constant and trend is coming insignificant..so is my model ok if i estimate under 'none' option as this gives the correct results
@@CrunchEconometrix thanks a lot mam for your speedy response.Its really very helpful for my research.I appreciate you for your so much concern towards the students and learners.
Good morning sir!! I want to choose a suitable las length of the variables (i have 4 variables) using unrestricted VAR. I found that lag length 3 is appropriate. but does it seem logical to find the optimal lag equal to 3 for annual data. Thanks for your help
Hi PIN, in that case, retain the 1 lag because VECM cannot be estimated as a static model. Please may I know from where (location) you are reaching me?
@@pinlyhuy7221 Awesome! Kindly spread the word about my TH-cam Channel to your friends and academic community in Cambodia 🇰🇭. They will find the content helpful too 😊.
Again once, Samia from Algeria, I thank you for your videos i am fun of your useful and simple explanation, i've one question i found AIC -6.332499 and Schwarz criterion -5.970776 which should i choose?
Madam Professor greetings.... Professor can we use these resulted "optimal lags" from VAR, into the Unit root tests for panel data? I mean can we use that method of lag selection for unit root tests in panel data. e.g. Pessaran CADF or CIPS tests
To summarize: (i) we transform our variables to be stationary (log or d_log) and (ii) look for the optimal lag length on transformed variables. Is that correct?
@@CrunchEconometrix In comparison to IMF course related to VAR (that I followed prior to yours), you are clearer, more informative and more precise. Thank you!
It means you won't include any lags. But for dynamic models, you need to include at least one lag. May I know from where (location) you are reaching me?
Normally in time series data we check stationarity of each variable. some series are stationary at level but some at first difference. so my question is that is it necessary to change the original variable data at first difference if the variables are stationary at first difference.
Hi Muhd, I think you mixed up this question: "is it necessary to change the original variable data at first difference if the variables are stationary at first difference"...are you asking whether to use 1st difference or level data?
@@CrunchEconometrix when the variables are stationary at level. Is it necessary to use data at 1st difference for using ARDL model or we have to use the original data for ARDLs
Use differenced data if you are using the OLS algorithm and level data if using the ARDL algorithm. I have videos in both procedures. Please check thru the Playlist. Thanks.
Hi Jay, as explained in the video, lag selection is not cast on stone. So, for daily data, you can use up to 30, 60, 120, ..., 360 lags. Also, consult other online resources. Thanks.
hello mam, your videos are really very helpful and easy to understand. I have a query- if the variables under study have different lag length than how to go for cointegration test?
Thanks Boni, for the encouraging words and feedback. Deeply appreciated! Watch my video on "This is how to specify ARDL model" after which you watch my ARDL videos.
hi Prof, I just want to confirm, to determine lag length for ARDL, it's differ with this video right? because I watched the ARDL video you, include the dependent as endo and the others as exogenous. Thank you
I am working with a series of 1981 - 2019. After testing for the optimal lag length, lag 3 was chosen as optimal. Should I go ahead with this selection or should I reduce it to lag2
Dear Professor I want to specify the lag length for the Johansen test. I saw that the results are sensitive to the lag lengths selected.(sometimes it rejects other times it fails to reject the null of no cointegration) VAR model would be misspecified for I(1) series. However, can i use VAR model for the purpose of lag length selection so that I can proceed to the Johansen test?
I forgot to mention, my series are not I(0) only I(1). I want to perform the Johansen test but I stuck with the lag length selection because the professor suggests to use VAR for that.
@@prepvisualsOk, watch my video on Johansen Conitegration Test. It's detailed enough to guide you through...and thanks for watching my videos. Please share with your colleagues.
@@prepvisualsOk, watch my video on Johansen Conitegration Test. It's detailed enough to guide you through...and thanks for watching my videos. Please share with your colleagues.
Madam, one basic question - if I have monthly data, I would select periodicity in auto arima as 12. However, if I have daily data then what should be the periodicity ? Should it be 365 or 0(default value) or 1?
Good day mam. Just a simple question. Can we reduce the maximum lags of the ardl model from 4 to 3? Because the maxlags 3 gives the better results. Thank you, hoping for your response.
Justine, it is better to use the optimum lags. But if 3 lags produce the best results and passes diagnostics, then use it and made a note to that effect in your work.
Dear Professor, I have a questions on the Log transformation; In a model let’s say VAR or VECM I have four variables. Can I use two in Log form and the other two on it is base/normal form? I mean is it OK to use mixed of Log and non-log in one model?
No, you can't do this in VAR for several reasons: the system is composed of endogenous variables. Using different functional forms will create interpretation problem...ok, if estimating ARDL model.
@@CrunchEconometrix Thanks u. I wanna ask about the book that u suggest to read to cover all of the statistics in eviews? Because I see that eviews old version was difference that the new version. That have many changes like the method
Hi Daniel, thanks for reaching out. Due to time constraints and my busy schedule, I limit interactions to TH-cam and refrain from personalized tutoring... unless it's for paid consultation.
Hi Alpha, thanks for the confidence reposed in my tutorials. Be rest assured that once I understand NARDL, I will make videos on the procedure. May I know from where (location) you are reaching me?
Mam, Really Thanks a lot,..! It is only because of these videos that made time series analysis easier for my projects. I would definitely share your videos and make your youtube channel more popular. I have a small doubt mam. For yearly data we are using lag 1. For monthly we are using 12,24. My data set is from 2012 - 2018. Both hourly data and prices for every 15 minutes is available here. Please do tell me what lag length should we choose for 15 minute block interval. Since it is a high frequency data, what lag length should we use? When I used lag length above 2000, my Eviews started getting stuck. Also in one of the videos it was mentioned that higher lag lengths would lead to loss in degrees of freedom. What shall I do mam? Thanks in advance, mam!
U estimate with 0 lag but exceptions can be made for certain models like VAR which must be dynamic and not static. May I know from where (location) you are reaching me?
@@shivally2077 Awesome Dr. Siva! 💕 Kindly spread the word about my videos to your friends, students and academic community in Bangalore and on social media for awareness. They'll learn some useful tips and skills too! 😊
Hi Saima, that's fine but before you do, scroll through my 107 videos and make sure you watch those videos on panel data. If you still have queries after watching, post them on the comment section of the respective TH-cam video so that other online audience can benefit from the discussion.
Hi Saima, that's fine but before you do, scroll through my 107 videos and make sure you watch those videos on panel data. If you still have queries after watching, post them on the comment section of the respective TH-cam video so that other online audience can benefit from the discussion.
Sure Haider I'm willing to assist in any little way that I can. But first, you subscribe to my Channel, then browse through the list of 102 videos to familiarise yourselves with the topics covered, afterwards you load in your data and practice along. That's the best and fastest way to learn...you'll be fine. Take care.
Hello dr When i do my ardl model i have the optimal lag is for the minimum aic but the model choosen using aic IS not really good. So i have two question: first, Can i choose HQ criteria if it was significant even if aic IS the minimum value? Second, i obtain ardl (1230) like a model, the 0 variable IS dropped on the short run. So what should i do to fix the problem? What should be the value of this variable in ecm? Thanks in advance
Mam, Really Thanks a lot,..! It is only because of these videos that made time series analysis easier for my projects. I would definitely share your videos and make your youtube channel more popular. I have a small doubt mam. For yearly data we are using lag 1. For monthly we are using 12,24. My data set is from 2012 - 2018. Both hourly data and prices for every 15 minutes is available here. Please do tell me what lag length should we choose for 15 minute block interval. Since it is a high frequency data, what lag length should we use? When I used lag length above 2000, my Eviews started getting stuck. Also in one of the videos it was mentioned that higher lag lengths would lead to loss in degrees of freedom. What shall I do mam? Thanks in advance, mam!
@@CrunchEconometrix Sorry for disturbing you again mam. My doubt is still not cleared. We can use lags 60, 120,..., 360 when it is a purely minute data (i.e., when the price is for every minute) but this is a data which tells price for every 15 minutes. Is it ok if we use the same lags 60, 120,..., 360.
@@arrthisampath3000 Relax, I said that lag selection is simply an empirical issue which requires a combination of intuition and reasoning. So, estimate your model and see the outcome.
@@CrunchEconometrix Dear mam, I really thank you whole-heartedly for clearing my doubts. Because none of the channels have taught us to do these step by step. It is only because of you that made me to complete my works sooner and clearer. Mam, I still have few more doubts and I still really sorry to disturb you again. For correlogram I got 96 as lag in both ACF and PACF plot on my first differenced data. I didn't get any other lags. So I kept my ARIMA as (96,1,96). Since this is the only model should I select this for further
R-squared 0.287638 Mean dependent var -0.001527 Adjusted R-squared 0.287629 S.D. dependent var 257.2794 S.E. of regression 217.1492 Akaike info criterion 13.59929 Sum squared resid 1.10E+10 Schwarz criterion 13.59947 Log likelihood -1582294. Hannan-Quinn criter. 13.59934 F-statistic 31319.83 Durbin-Watson stat 2.187677 Prob(F-statistic) 0.000000 Is it right, mam? and what shall I do to proceed further.
@@CrunchEconometrix I'm from Morocco, I'm looking for the application of the distributed lag model (almon) in eviews, what you explained is the first part of the adjustment , isn't it ?
@@naoufalelalaouielbahi8453 I have several videos on the ARDL application. Browse through the Playlists. I'll appreciate it if you can share the link to my TH-cam Channel with your students and academic community in Morocco 🇲🇦 ...May God bless you as you do, amen 🙏
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, help me stay online.
How can I subscribe?
@@tajudeenodetayo2333 Simply click on the SUBSCRIBE button showing in any of my videos or better still click on the SUBSCRIBE button on my TH-cam Channel and the BELL so that you can get my videos whenever I upload them.
You have 18k+ now lol and it's still growing bcz I'm subscribing. Keep up the good work
you r so smart yet so kind bcz sharing the knowledge here😭😭😭, keep healthy ma'am, may the lord bless ur soul and ur family 🙏🙏🙏
Amen and amen, Fajr. Thanks for the kind words.
Thanks ma'am
Thank you so much for your explanation in the video!! I legit found the answer I was hoping for - to understand how lags work within the selection criteria of VAR!! Very good work and analysis, keep it up!!! THANK YOU again, God bless you!!!🙏❤❤
You're very welcome, Nicolitsa...thanks!
In fact you are the greatest tutor ever
will be expecting you to cover more topics
Hahahaha Adedeji, not the greatest just teaching the little I know😊. Thanks for the positive feedback, deeply appreciated! Please may I know from where (location) you are reaching me?
Obafemi Awolowo University (Economics Department)
@@samueladedeji6883 Wow! The Great Ife!🙌 I'll appreciate it if you can share the link to my TH-cam Channel with your friends and academic community on social media. They will find the content helpful too. Thanks!
Selecting Optimal lags for Annual, Quarterly and Monthly data is a rare insight that comes only through experience. Thanks for providing this lag selection insights.
Yeah u're right, Bhaskar...and thanks for watching my videos!
Madam, I had one question - I am running an ARIMA on Bitcoin Close price data with daily data of (12,12) when I select Max AR as 12. Is this still fine to have such high lag values considering I have daily data.
@@bhaskartripathi Yes, u're still on track since it's a daily data.
Thank you so much professor. It helps to see your videos
Wow! It's been three days and I already love econometrics, thanks to your extremely helpful materials. My question is on the way forward once have established that there is no co-integration.
Hi Margaret, thanks for the positive feedback and remarks on my TH-cam videos. Deeply appreciated! Depending on the outcome of your unit root test, you perform either a VAR or ARDL technique. May I know from where (location) you are reaching me?
Your videos are always clear. Thank you!
Thanks Nur for the encouraging feedback. Deeply appreciated! 💕 Please tell others too! 😊
nice tutorial..thanks for sharing
Hi Abhijit, thanks for the encouraging feedback...deeply appreciated! May I know from where (location) you are reaching me?
@@CrunchEconometrix me from India...... whatsapp no. 91-9706027177.. you can add me if u like..
Thanks for the video, what are most suitable lag value to set when dealing with monthly data?
Kindly watch my video on OPTIMAL LAG SELECTION for guide.
Please, what how do you further make use of optimal lags. Like what test is optimal lags relevant
Hi Mehinsan, please watch my Time Series videos to know the relevance and application of optimal lags. Thanks
thank you so much CrunchEconometrix for explaining clearly. this is much helpful.
U're very welcome, Naw. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix, i am from Nepal, country from the south Asian region.
thank you.
hi thanks; but criteria for var give me 0 lags for all criteria
how can i choose between 1 or 2 lags ?
the 1 and 2 are both stable model with all tests( normality, AR, homoscedasity, LM) for VAR
thanks
Go ahead and use the lag length that produces good results and diagnostics. Ensure you explain why in your work.
You're an amazing instructor.
Thanks, for the encouraging feedback Sir... appreciated!
Hello, Professor Adeleye. Thank you for this informative video. I hope you can shed light to my problem.
I am working on a panel data analysis, spanning 32 years (annual). I followed your instruction on finding the optimal lag, and the results show 8. However, upon integrating it in the regression, the model suffered a normality problem.
I tried experimenting with lags for the dependent (GDP), and when I set the lag to 30, the normality and autocorrelation problem was fixed. May I use a lag of 30 or shall I continue with the optimal lag given by the lag length criteria results (8)?
Using lags is subject to understanding your dataset, the time dimension and loosing too many degrees of freedom.
Thank you so much for this video. If one of my series is stationary at first difference then we should use that series at level or at first difference while calculating lag for the model.
Hi Richi, 1st difference of the variable is not used for lag selection. It is either the raw or log form.
I learnt all my research skills from Crunch
Thanks Nchimunya, for the encouraging feedback. Deeply appreciated!
what is your equation after choosing lag? is it GDP=aGDPt+bGDP(t-1)+cGDP(t-2)+dpde(t)+....+pdpe(t-4)+qpci(t)+rpci(t-1)+error
Naman, the equation you specify depends on the estimation technique.
Can you tell me how to find individual lags in above mentioned equation? I am regression ARDL(p,q1,q2) i want to find p,q1,q2
Kindly follow my ARDL models on what to do. Thanks.
I follow you since Malaysia...
Wow! 👏🏽 👏🏽 👏🏽 Thanks! 😊 🙏
Thank you so much ma, is it better I use the unlogged varriables
Hi Favour, I'd suggest you read more about log transformation and decide if you want to use log or level-variables.
I am using an ARDL model because of co-integration. Please do these steps apply to choose an optimal lag length for my model?
Yes, they do.
What a fantastic video to explain ma'am! May I kindly ask however, how would you choose lags for mixed-frequency data such as Quarterly And Monthly data?
Thanks Babar, for the encouraging feedback. Deeply appreciated! But I have no idea about your query. You may want to check other online resources.
I’m doing johansen cointegration and try to find optimal lag. My data are stationary at I(1), so in finding optimal lag, which type of data to use whether to use level data or 1st difference data? Thank you
Watch my VAR or ARDL videos to know what to do.
in case of daily averages, what would be the lag intervals endogenous ? 1 and ???
Amina, this clip explains the lags to deploy with different time structures. You may want to watch it again and apply the information to your study. Thanks.
thanks for the video! if I've got daily data, what is the best lag to include?
Alvise, up to 365 lags.
How to then run VAR with the variables of different lag length in eviews? Thank you
Iryna, not possible to the best of my knowledge.
Dr Ngozi, can you make a video for censored and truncated regression models? I have a difficulty to understand these models, so I need your explanations.
Hi Patrick, thanks for the suggestions. Are you referring to TOBIT, LOGIT and PROBIT models? If yes, they are already on my to-do-list. Once I clear up my backlogs, I'll read up on them and create the videos. Thanks for believing in me. Keep watching, keep sharing too! :)
Thank you so much for your videos they have really helped us in our research and everything is clear. I just have a question regarding the optimal lag length, we are running an ARDL model on annual 30 observations and the optimal lag selection according to VAR was 5 is this ok or should it be reduced to 1 or 2?
Hi Malak, using 5 will reduce your degrees of freedom. Reduce to 2 and make a note in your work explaining why.
Hello dr, im using daily data from 2018 to 2022 and i have hard time to finding optimal lag since alot of video explain quarterly/ monthly/ yearly data only . do you have suggestion on what the optimal lag for daily data ? my observation is 1443 data
Hi Syazz, the time span is TOO small. You need at least 30 years observations.
@@CrunchEconometrix yes dr, I'm currently learning basic econometrics, so my prof gave me some data which is the number of observations 1443. This is for my assignment. I try to use 365 lag but the data is too small so I try to use 50 lag, and it shows my optimal lag for the data which is the smallest aic is 50. Do you have any suggestions for the lag range? I'm trying to figure out the lag because I want to do LM test in eviews. Thank you in advance !
Syazz, you said your data is 2018-2022. So, how come you have 1443 observations. I am not understanding. I suggest you ask your Prof for clarifications.
Hi madam, If the residual autocorrelation test results indicate that the p-value for the chosen lag is 0.000, which is less than 0.05, suggesting the presence of autocorrelation, should I change the number of lags and re-estimate the VAR model with this new lag?
Yes, adjust the lag length.
@@CrunchEconometrix Thank you for your reponse. I have another question when I estimated the model ,i found the R-squared coefficient is low for both variables, even though I tested the model's stability ,and the significance test of the coefficients indicates a probability less than 5%, so the model is significant .Does this low coefficient pose a problem?
Hello, I'm using E-views 11 for student and there is no option for "Lag Lenght Criteria" in Lag Structure option. My data is an annual panel data of 27 countries with a total of 810 observations. Since I can not determine the length of the Lag, is it appropriate to go with the automatic option which is for annual is 1-2 lag? all data series are cointegrated at 1st difference, I applied Johansen and VECM with lag 1-2.. please give me suggestion. Thank you.
Your channel is very educative and informative :)
Hi Sheila, thanks for the positive feedback....and yes allow the EViews determine the optimal lags.
Thank you very much for this! What if we use daily series ?
Hi Andrian, thanks for the encouraging feedback. Deeply appreciated! You can use up to 365 days. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix thank you very much! I'm currently studying in Edinburgh and it's the first time I'm learning econometrics and we are asked to run an analysis that requires quantitative research method. Your vids are really helpful.
I am using eview10 and can't find lag structure in the View drop down menu. Prof ma please assist me
Lucy, kindly follow any of the steps shown in my ARDL EViews videos.
Thank you for your video, if I understand you correctlym does it mean that if AIC shows ,for example, to use 2 lags for regressor, then to make the model better we need to add lag1 and lag2 of this regressor to the model?
Hi Islam, I show how to use optimal lags in my time series videos. I'll encourage you to watch them for clearer understanding. Thanks.
Hi! Thanks for ur explanation!! How about daily data? what is the range of lags for daily data? Thanks
Hi Ng, thanks for the encouraging feedback. Deeply appreciated! You can take up to 180 lags.
Very informative video.
What should the lag number daily or hourly data?
365 for daily. Use that to decipher for hourly.
I have to thank you first for your helpful videos. I just have a technical question. You said we use unrestricted VAR to estimate the optimal lag length because we assume that the variables are not co-integrated. Why we assume "not co-integrated"? Thank you again.
Hi Mona, because the cointegration test has not been performed to establish that. Hence, determine the optimal lag and other steps follow. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hello to you. I am in Italy. I am a student and I am preparing a project with Eviews. If I may I would like to ask is it better to run the co-integration test or unrestricted VAR to find the optimal lag. Because I have run the co-integration test and it exist among my variables. But interesting thing is that when I run the co-integration test the lag that it suggest me is the same when I run the unrestricted VAR.
Can you do a video on spillover effect using BEKK GARCH model?
Hi Ann, the suggestion is noted. Thanks!
@@CrunchEconometrixIt would really grateful.. Thank you so much..
I am running ARDL ECM model here and which method should I apply to choose the optimal lag length? Including all variables in the VAR or run VAR on each variables ? what are the difference between including all variables in VAR and running VAR on each variables?
Hi there, for optimal lags selection for ARDL models, VAR is performed on each variable. But if a VAR model, all the variables are listed. Since, you are performing an ARDL model, I'll advice that you watch all my videos (Stata or EViews, check out the playlists) on how to perform the procedures from start-to-finish. My videos are quite easy to understand. Thanks and good luck with your research.
well done, thanks prof.
Thanks Hamid for the positive feedback on my video, deeply appreciated! 💕 😊 May I know from where (location) you are reaching me?
Suppose there are 4 variables and in cace of ARDL bound test approach we get 3 Cointegrating equation by F test. Now my question is what is the best equation to show short and long run impact and why? and how the lag of the equation like (1,0,2,1) or (2,1,0,1) is selected. Thnx
Hi Bikash, I covered quite a lot on ARDL models and the Bounds test. Find the time to watch them. Practice along, listen to the explanations and jot some notes. You'll get better if you do these.
@@CrunchEconometrix Thnx
Dear Professor,
What is the difference between determining the optimal Lag Length
for a model and a variable?
Do they have different usage or different purpose?
In the video the optimal Lag Length for the model is 2, but
when you checked for the individual variable it showed different optimal Lag
Length; one of the variables showed two optimal lag, the other four and the
other one.
Please explain the purpose and usage of determining optimal
lag for a model and each variable.
Hi Shaf, as rightly said optimal lags for a model and for variables differ as shown in the video. This applies to ARDL modeling not VAR. But the process is simplified with EViews and Stata as these software can automatically select the appropriate lags for each variable during estimation. Thanks.
Thank you so much for the clarification.
U're welcome, John.
Hello Dr. I watched your video on ARDL estimation and specifically your demonstration on how to determine optimal lag. Its quite different from what you outlined here. My question is, what is the best way to determine appropriate lag structure.
Kenechukwu, while this clip gives a general overview on optimal lag selection, that of ARDL is specific to the estimation technique....and there is no "best way" as I explained it to be "an empirical issue" subject to several factors I have pointed out. I will also suggest that you read up other sources on how to determine optimal lags for deeper understanding.
I have a question.. If your AIC is negative, what does it means?
Most times, the information criteria (AIC, BIC, HQIC) are in the negative. So, it is best to consider the absolute figures to avoid confusion in making a decision.
Dr. Ngozi, i have been following your tutorials and they are really helpful. Now concerning Lag order selection creteria, what should you do when the creteria dont agree on the lag. For instance, when AIC shows minimum lag should be 1 while SBIC shows the minimum lag should be 2??? kindly explain
Dear Onesmus, edit the multiple question marks. Highly unnecessary. Thanks.
great explanation. however, i have a question. now i understood how to select the lags, but you've done it separately for each variable. my question is, how should select the appropriate lags for the whole model? say THE DEPENDENT VARIABLE IS GDP= PDI+ PCE?
Hi Yassin, same procedure only that you include all 3 variables together starting with the depvar.
@@CrunchEconometrix well understood Dr. thank you.
How about daily data? How may lag we should use?
up to 365
what if the model suing the obtained optimal lag suffer from heteroskedasticity? is it convent to take the lag after ? i dont want to take the log of my variables. Thanx
Amina, your query is confusing. Kindly rephrase. Thanks
Oh sorry . I calculated the optimal lag which is found to be 1, then re-estimate the PVAR(1), no serial correlation found, BUT Heteroskedasiticity was obtained. I tested again on lag 2 and no issue found. But As far as i understand increasing lag must be based on Theory or previous studies which I couldn’t find. I applied the GLS but couldn’t retest the Heteroskedasiticity. Any suggestions?
@bellisma77 Use lag 2 and support your analysis with the literature.
@@CrunchEconometrix i will try to find. Thank you for you advice 🙏
Mam is it possible that in our model there is no constant and trend and can we select under trend specification the option"none" as all the options are giving insignificant result..meaning when I am estimating the model by selecting both constant and trend the p value for both constant and trend is coming insignificant..so is my model ok if i estimate under 'none' option as this gives the correct results
Zoya, kindly look up other online resources on more information about using the "none" option. Thanks.
@@CrunchEconometrix thanks a lot mam for your speedy response.Its really very helpful for my research.I appreciate you for your so much concern towards the students and learners.
Good morning sir!! I want to choose a suitable las length of the variables (i have 4 variables) using unrestricted VAR. I found that lag length 3 is appropriate. but does it seem logical to find the optimal lag equal to 3 for annual data. Thanks for your help
Yes, if you have enough time span you can use 3 lags for annual data.
madam, how do I know when will I use log difference for variable in ARDL model? basically how to build ardl model?
Hi Fatin, log difference is the growth rate of a variable. Usage depends on the researcher.
crystal clear. thanks very much
Thanks Jefri, for the encouraging feedback. Deeply appreciated! 🙏 ❤️
Amazing video! thank you
You're welcome, dear...kindly tell your students and colleagues to subscribe. Gracias! 💕 😊
Hey mam, if you don’t mind I want to know how much lags do I need for a daily data? Thank you.
Hi Said, use up to 365.
Hi, Dr. if our optimal lag is 1 when we process VECM need to minus 1 lag it would be zero, what should I do? thank you.
Hi PIN, in that case, retain the 1 lag because VECM cannot be estimated as a static model. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix Thank you, Dr. for your answer. I'm from Cambodia.
@@pinlyhuy7221 Awesome! Kindly spread the word about my TH-cam Channel to your friends and academic community in Cambodia 🇰🇭. They will find the content helpful too 😊.
Again once, Samia from Algeria, I thank you for your videos i am fun of your useful and simple explanation, i've one question i found AIC -6.332499 and Schwarz criterion -5.970776 which should i choose?
Samia, you choose the criteria with the lowest value.
Amaze..
Thanks, Tayyaba!
THANK YOU!!!
You are welcome 🥰🙏
Madam Professor greetings....
Professor can we use these resulted "optimal lags" from VAR, into the Unit root tests for panel data?
I mean can we use that method of lag selection for unit root tests in panel data. e.g. Pessaran CADF or CIPS tests
No.
Hello Mme, do you have a video which explains how to determine the optimal lag when conducting panel VECM in eviews 10?
Not yet dear...
Ok Mme, I shall wait for the video. Thank you!
But what if after the Johansen test we get to know that there is cointegration? Does the lag through Unrestricted VAR still hold true?
Hi Rehmat, you reduce the lag by 1 to run the VECM. Kindly watch my videos on the procedure. Thanks.
To summarize: (i) we transform our variables to be stationary (log or d_log) and (ii) look for the optimal lag length on transformed variables. Is that correct?
Yes, Ksenija but variables transformation is not compulsory, though advisable.
@@CrunchEconometrix In comparison to IMF course related to VAR (that I followed prior to yours), you are clearer, more informative and more precise. Thank you!
For the optimal lag I don’t know whether to write the first lag and add lag 2 or just write lag 2
Patrick, it is vivid that my videos are practical. Watch what I did and adapt to your analysis.
Ma I'm running annual data and the lag for all my variables except one is 1 while the other is 2. Is it advisable to use lag 1?
Hi Rodiya, you have the information at your disposal. This clip explained the intricacies of selecting optimal lags.
terima kasih bu
hehehe
Ubay, may I know the amusement?
@@CrunchEconometrix just an expression of thanks :) , greeting from indonesia
@@irus7498 Deeply appreciated! My love to Indonesia 🇮🇩 and do share my videos with your friends and academic community. Thanks
Thanks madam
hello mam, can we take log of ratios and rate also. like variables trade to GDP ratio.
Opinions are divided on this but I will take the the log of trade openness.
Hello madam. What does mean if i got the optimal lag equal to zero? Looking forward for your reply thank you.
It means you won't include any lags. But for dynamic models, you need to include at least one lag. May I know from where (location) you are reaching me?
Normally in time series data we check stationarity of each variable. some series are stationary at level but some at first difference. so my question is that is it necessary to change the original variable data at first difference if the variables are stationary at first difference.
Hi Muhd, I think you mixed up this question: "is it necessary to change the original variable data at first difference if the variables are stationary at first difference"...are you asking whether to use 1st difference or level data?
@@CrunchEconometrix when the variables are stationary at level. Is it necessary to use data at 1st difference for using ARDL model or we have to use the original data for
ARDLs
Use differenced data if you are using the OLS algorithm and level data if using the ARDL algorithm. I have videos in both procedures. Please check thru the Playlist. Thanks.
@@CrunchEconometrix thanks
U're welcome, Muhd.
MAm how many lags should be used for daily stock price data?
Thank-you in advance ?
Hi Jay, as explained in the video, lag selection is not cast on stone. So, for daily data, you can use up to 30, 60, 120, ..., 360 lags. Also, consult other online resources. Thanks.
@@CrunchEconometrix Thank-you for your time, i will recommend you to my colleagues
hello mam, your videos are really very helpful and easy to understand. I have a query- if the variables under study have different lag length than how to go for cointegration test?
Thanks Boni, for the encouraging words and feedback. Deeply appreciated! Watch my video on "This is how to specify ARDL model" after which you watch my ARDL videos.
@@CrunchEconometrix ok mam. Thank you so much for your quick reply. Truly and deeply humbled.
You are welcome 😊.
hi Prof, I just want to confirm, to determine lag length for ARDL, it's differ with this video right? because I watched the ARDL video you, include the dependent as endo and the others as exogenous. Thank you
This is just a general video about lags selection.
I am working with a series of 1981 - 2019. After testing for the optimal lag length, lag 3 was chosen as optimal. Should I go ahead with this selection or should I reduce it to lag2
Go ahead.
@@CrunchEconometrix thanks a lot Ma, for your quick response
Dear Professor
I want to specify the lag length for the Johansen test. I saw that the results are sensitive to the lag lengths selected.(sometimes it rejects other times it fails to reject the null of no cointegration)
VAR model would be misspecified for I(1) series. However, can i use VAR model for the purpose of lag length selection so that I can proceed to the Johansen test?
I forgot to mention, my series are not I(0) only I(1). I want to perform the Johansen test but I stuck with the lag length selection because the professor suggests to use VAR for that.
@@prepvisualsOk, watch my video on Johansen Conitegration Test. It's detailed enough to guide you through...and thanks for watching my videos. Please share with your colleagues.
@@prepvisualsOk, watch my video on Johansen Conitegration Test. It's detailed enough to guide you through...and thanks for watching my videos. Please share with your colleagues.
helpful ♥
Thanks Zine, for the positive feedback. May I know from where (location) you are reaching me?
Madam, one basic question - if I have monthly data, I would select periodicity in auto arima as 12. However, if I have daily data then what should be the periodicity ? Should it be 365 or 0(default value) or 1?
Yes, with daily data you can use 365days.
@@CrunchEconometrix for hourly data what should be the lag length mam?
@@arrthisampath3000 60, 120, 180,...,360
Good day mam. Just a simple question. Can we reduce the maximum lags of the ardl model from 4 to 3? Because the maxlags 3 gives the better results. Thank you, hoping for your response.
Justine, it is better to use the optimum lags. But if 3 lags produce the best results and passes diagnostics, then use it and made a note to that effect in your work.
@@CrunchEconometrix can u give me some references about this adjusting of optimal lags? Thank you.
Not everything require references. Simple justification about what you did will suffice.
Dear professor,
If the data is daily which lags is appropriate.
best regards
Hi Zeravan, I explained all you need to know about lags in this video. Kindly apply the knowledge.
Thank you
U're welcome, Bayu...kindly share my Channel link with your students and academic networks! 💕
Dear Professor, I have a questions on the Log transformation;
In a model let’s say VAR or VECM I have four variables. Can I use two in Log form and the other two on it is base/normal form? I mean is it OK to use mixed of Log and non-log in one model?
No, you can't do this in VAR for several reasons: the system is composed of endogenous variables. Using different functional forms will create interpretation problem...ok, if estimating ARDL model.
Thank you so much for the explanation!
My pleasure, John.
Please can you write how the model wit the lag will look like
Which model are you referring to please?
Hi madam, I wanna how about the data that we took daily. The suitable lags of the daily data for this situation?
Up to 365 lags.
@@CrunchEconometrix Thanks u. I wanna ask about the book that u suggest to read to cover all of the statistics in eviews? Because I see that eviews old version was difference that the new version. That have many changes like the method
Hey, how many lags can I include for daily data ?
But I explained that in the video.
No for daily frequency you didn't. You explained for annual and quarterly.
Alrite. You can use up to 365.
Dr, please is it possible to establish a contact with you. If yes please tell me how. My problem is peculiar
Hi Daniel, thanks for reaching out. Due to time constraints and my busy schedule, I limit interactions to TH-cam and refrain from personalized tutoring... unless it's for paid consultation.
can Akaike criterion be negative?
Yes
thank you so much.
U're welcome, Yasin...please share my videos with your friends and academic community on social media. Thanks! 💕 😊
@@CrunchEconometrix i will. cheers.
Good morning sir, can you please do a video on nonlinear ARDL? I really need. Thank you...
Hi Alpha, thanks for the confidence reposed in my tutorials. Be rest assured that once I understand NARDL, I will make videos on the procedure. May I know from where (location) you are reaching me?
If lag optimun in lag 0 , what should i do ? please respon to me, thank you
That will depend on the intended model.
Mam, Really Thanks a lot,..! It is only because of these videos that made time series analysis easier for my projects. I would definitely share your videos and make your youtube channel more popular.
I have a small doubt mam. For yearly data we are using lag 1. For monthly we are using 12,24.
My data set is from 2012 - 2018. Both hourly data and prices for every 15 minutes is available here. Please do tell me what lag length should we choose for 15 minute block interval. Since it is a high frequency data, what lag length should we use? When I used lag length above 2000, my Eviews started getting stuck. Also in one of the videos it was mentioned that higher lag lengths would lead to loss in degrees of freedom. What shall I do mam?
Thanks in advance, mam!
I have responded to this.
@@CrunchEconometrix Thanks a lot mam
what if i have daily data?
Hi Amina, this video explains what lags to use with different time frames of the data...so, you can adapt any that fits your data. Thanks.
what should i do if the optimal lag is 0 in all criteria?
U estimate with 0 lag but exceptions can be made for certain models like VAR which must be dynamic and not static. May I know from where (location) you are reaching me?
for daily data what is optimal lag?
I explained that in this video. You may need to watch it again.
thank you very much
You are welcome, Muhd!
How much lag to use for daily data ?
Lags of 30, 60, 180,..., 360 days are appropriate. May I know from where (location) you are reaching me?
@@CrunchEconometrix thank you madam...from bangalore, india
@@shivally2077 Awesome Dr. Siva! 💕 Kindly spread the word about my videos to your friends, students and academic community in Bangalore and on social media for awareness. They'll learn some useful tips and skills too! 😊
mam i want to discuss some issues relevent to my panel data plz gide me
Hi Saima, that's fine but before you do, scroll through my 107 videos and make sure you watch those videos on panel data. If you still have queries after watching, post them on the comment section of the respective TH-cam video so that other online audience can benefit from the discussion.
Hi Saima, that's fine but before you do, scroll through my 107 videos and make sure you watch those videos on panel data. If you still have queries after watching, post them on the comment section of the respective TH-cam video so that other online audience can benefit from the discussion.
Madam plz guide me in my research....
Sure Haider I'm willing to assist in any little way that I can. But first, you subscribe to my Channel, then browse through the list of 102 videos to familiarise yourselves with the topics covered, afterwards you load in your data and practice along. That's the best and fastest way to learn...you'll be fine. Take care.
Hello dr
When i do my ardl model i have the optimal lag is for the minimum aic but the model choosen using aic IS not really good. So i have two question: first, Can i choose HQ criteria if it was significant even if aic IS the minimum value? Second, i obtain ardl (1230) like a model, the 0 variable IS dropped on the short run. So what should i do to fix the problem? What should be the value of this variable in ecm?
Thanks in advance
Waed, you can always use alternative IC but make sure you justify it. 0 lag variable ONLY appears in the LR part of the results.
Mam, Really Thanks a lot,..! It is only because of these videos that made time series analysis easier for my projects. I would definitely share your videos and make your youtube channel more popular.
I have a small doubt mam. For yearly data we are using lag 1. For monthly we are using 12,24.
My data set is from 2012 - 2018. Both hourly data and prices for every 15 minutes is available here. Please do tell me what lag length should we choose for 15 minute block interval. Since it is a high frequency data, what lag length should we use? When I used lag length above 2000, my Eviews started getting stuck. Also in one of the videos it was mentioned that higher lag lengths would lead to loss in degrees of freedom. What shall I do mam?
Thanks in advance, mam!
Since you have "minutes-data", try the following lags 60, 120,..., 360
@@CrunchEconometrix Sorry for disturbing you again mam. My doubt is still not cleared. We can use lags 60, 120,..., 360 when it is a purely minute data (i.e., when the price is for every minute) but this is a data which tells price for every 15 minutes. Is it ok if we use the same lags 60, 120,..., 360.
@@arrthisampath3000 Relax, I said that lag selection is simply an empirical issue which requires a combination of intuition and reasoning. So, estimate your model and see the outcome.
@@CrunchEconometrix ok mam. Thanks a lot.
@@CrunchEconometrix Dear mam, I really thank you whole-heartedly for clearing my doubts. Because none of the channels have taught us to do these step by step. It is only because of you that made me to complete my works sooner and clearer.
Mam, I still have few more doubts and I still really sorry to disturb you again.
For correlogram I got 96 as lag in both ACF and PACF plot on my first differenced data. I didn't get any other lags. So I kept my ARIMA as (96,1,96). Since this is the only model should I select this for further
Variable Coefficient Std. Error t-Statistic Prob.
C -0.010120 1.862233 -0.005434 0.9957
AR(96) 0.920059 0.000364 2528.116 0.0000
MA(96) -0.671310 0.000678 -990.7412 0.0000
SIGMASQ 47152.95 22.26408 2117.894 0.0000
R-squared 0.287638 Mean dependent var -0.001527
Adjusted R-squared 0.287629 S.D. dependent var 257.2794
S.E. of regression 217.1492 Akaike info criterion 13.59929
Sum squared resid 1.10E+10 Schwarz criterion 13.59947
Log likelihood -1582294. Hannan-Quinn criter. 13.59934
F-statistic 31319.83 Durbin-Watson stat 2.187677
Prob(F-statistic) 0.000000
Is it right, mam? and what shall I do to proceed further.
thank you so much
U're welcome, Naoufal😊. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix I'm from Morocco, I'm looking for the application of the distributed lag model (almon) in eviews, what you explained is the first part of the adjustment , isn't it ?
@@naoufalelalaouielbahi8453 I have several videos on the ARDL application. Browse through the Playlists. I'll appreciate it if you can share the link to my TH-cam Channel with your students and academic community in Morocco 🇲🇦 ...May God bless you as you do, amen 🙏
@@CrunchEconometrix Already done, i'm so glad, thank you professor