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.
@@CrunchEconometrix mam in this video you have shown that all variables are stationary at 1st difference. but you have conducted the VAR on raw variables not on the first difference variable. kindly clarify....
Thanks Prof. I have learnt how to perform econometric data analysis from watching your channels. I currently have I(1) and I (0) variables in my system of equations. My objective is to perform a causality test on the macroeconomic variables including oil revenue. Which would suffice: ARDL or VAR.
Great. Why you calculated the optimal lag for each variable individually? Why not all variables at the same time? As VAR model will treat all variables as endogenous?? Thank you
Good tutorial mam. Have only one doubt. You explain that VAR logarithm automatically difference the series to first difference. Therefore, you compute the result with only natural log variables in spite of series is stationery at first difference. But please clarify if our variables are stationery at level after taking natural logarithm, is VAR model still differentiate them to first difference or not. Please clarify. Also tell a value from your results from which we can verify that results are automatically diffenced to first level by VAR Alogarithm.
Shubham, there are two Schools of Thought to estimating the VAR model. I followed one of them. Watch the INTRODUCTORY video on VAR specification to know which one. Thanks
Hi Maria, thanks for the encouraging feedback. Deeply appreciated! I appreciate the suggestion and once I understand the procedure I will create simple tutorials on it. Please may I know from where (location) you are reaching me?
Jones Dahlback Thanks Jo, I'm encouraged by your comments. I'll appreciate if you can help share my link with your social media community and academic networks so that others can know too... 😉
Thanks Masaid for the positive feedback on my videos. Deeply appreciated. But no promises on Bayesian VAR. Too complicated...may I know from where (location) you are reaching me?
Wow, good to know Masaidj...kindly share my YT Channel with your students and the Belgian academic community. They'll find my videos very useful...thanks!
Thankyou for your clear explanantion. Prof, what should we do if in our variable there are serial correlation and heteroscedasticity? Is it okay to run the data with VAR?
Thanks Professor. Very useful video. Please let me know if I can use a non-linear equation (A polynomial or a quadratic equation) in a VAR system. Will there be any change in the way I interpret?
I'll say yes, Nithya though I've never used the procedure before but I've come across some papers online. Do a Google search for a couple in that area. Also, Chapter 14 of Gujarati Basic Econometrics is devoted to Non-linear models with worked examples to guide you. Hope these tips are helpful. Take care.
Hello ,your videos are really helpful .nobody explains it your way .I have a question , you have only tested the stationarity at difference one but you didnt make a new object to reflect this change u want before running forecast . If I have a non stationary and non normal situation at level , I make a new object like d(log(variable)) and i continue my study based on that . Please correct me if what i am doing is wrong!!
Thanks for video. I want to ask something. What should we do when we have both stationary and non stationary series ? Is it enough to just take differences of non stationary ones, or do I have to take first differences of all variables ?
@@CrunchEconometrix Thank you for answer. But there is something I am lost. I have 5 variables and 3 of them are non stationary. When I take just three of them, since the non stationary series become shorter than than stationary ones, I am not able to run the regression.
@@zekiogulcansengul3593 Yes, you lose a year when you take the first difference and also when lags are taken. These reduce the degrees of freedom. How many years of observations do you have originally?
CrunchEconometrix it is monthly data regarding the inflation. I havr change in fx, commodity price inflation and output gap. Then , I omit the first row and starting from the second row that all variables are available , right ?
Hi Dr. I have the same question as Chathuri had that one of the preconditions of the VAR Model/test that data should be stationary. In this respect, your real data/Level form is non-stationary and became stationary at 1st Difference but you tested the VAR on data without converting in 1st difference ..... So my question is either VAR automatically converts the data into first difference and shows the results after taking 1st Difference?
@@muhammadshehryar5749 Thanks pal, you're one of those people who inspired me every day to turn out quality videos. Can you please help me share my videos with your friends, students and colleagues. Tell them they have a lot to learn by subscribering to my TH-cam Channel. Thanks in anticipation...gracias! 💕 😊
Maam you have checked lag of each variable individually before running unrestricted VAR. But we know lag test has to be done taking all endogenous variables together. What is the correct procedure? Another fact you used 1 lag while running VAR with all endogenous variables, but selection criteria require 1 lag for each variable. In case of VECM, one less lag is taken as data is transformed into 1st difference automatically. If out of three variables, say Y shows 2 lags, X shows 1 lags and Z shows 3 lags, in that case what to do if you test lag individaully before running VAR?
Hi Anan, thank you for that sharp observation....and you are right for a VAR model all the variables should be put together while running the unrestricted VAR to obtain the optimal lag lengths and not individually determined. I think I did it that way just to teach/show some people how to determine the optimal lags. Otherwise, it's not ideal. I think I was just fortunate that all the variables have 1 lag (lol)...thanks for the valid observation!
Hello my model is suffering from serial correlation and heteroscedasticity. I am dealing with daily stock price data, and i have 4 variables in the system. However when i am stesting the model for the stability test, in that case is OK. Can i conduct VAR and SVAR in that case?
Hello maam , i already got the answer from the previouse questions. Anyway thank you. I will take care tk share your videos with my friends. Wonderful explanations.
hello, can you help me on how to compute the forecast error of government spending? that is, forecasts of government spending are taken from October publications of the IMF’s WEO. Then, the fiscal spending shocks are identified as the forecast errors of government spending. Thus, FEi,t=gi,t(actual)-gi,t(forecast) where gi,t= Gi,t/Yi,t is government spending as a share of GDP. The actual government spending comes from the October WEO of the following year. i-refers to country and t refers to time.
Hello. May I ask why did you estimate a VAR model of the variables at level? Correct me if I am wrong, but I thought that running a VAR model for nonstationary series would yield a spurious regression. So, the variables to be estimated in the model should be the first difference of the three variables.
can this method be applied to a VECM model as well? for checking the lags individual variables need to be taken or can we take all the variables together as well?
Hi Dalisto, the website is being upgraded. Here's the link to obtain the datasets (some are free while some are available on purchase) 👇 cruncheconometrix.com/view/datashop.php
Please what should be done when the null hypothesis of joint normality test for VAR residuals is rejected? Will the results still be significant in the absence of normally distributed errors?
Dino, for me (others may disagree) I worry less about normality once my model passes the serial correlation, heteroscedasticity, stability and multicollinearity (VIF) test.
@@CrunchEconometrix Thank you for your fast reply.i'm from Malaysia. my variables are already in natural logs when i got the heteroskedasticity. May i know how do i get the option to use robust standard error to run var?
Hahahaha Adrienne, I don't about that but we learn everyday. The reality is that, in a VAR model all variables are endogenous. That is, that explain one another.
Hello Mam: here is a ref what I have got:www.personal.psu.edu/hxb11/EasyRegTours/VAR_Tourfiles/VARX.PDF. Also if you look over Eviews popup window there is a box for exogenous var box. That's why I wanted to know what about this exog variable option as you said all must be endo and no exog. Thanks
Since the selected time series are stationary at level 1, why don't you run the VAR with stationary data without going for level data VAR model?? Pls give quick reply
Hi Chathuri, the VAR model is estimated using either the VAR algorithm or the OLS algorithm. If the former, the level of the series is used but the 1st difference if using the latter. I used the VAR algorithm and if you see the outcome, the results are in 1st difference. Hope this clarifies, thanks.
I confused that the VAR model should be called with the stationary data series. You mean the results should be in 1st difference(stationary data)? ok thanks dr
Hi Dr. I have the same question as Chathuri had that one of the preconditions of the VAR Model/test that data should be stationary. In this respect, your real data/Level form is non-stationary and became stationary at 1st Difference but you tested the VAR on data without converting in 1st difference ..... So my question is either VAR automatically converts the data into first difference and shows the results after taking 1st Difference?
@@muhammadshehryar5749 Yes, the VAR algorithm differences the series during estimation. That's why you use the level not 1st difference. Verify this with your own data and observe your results.
you say that variables must be stationary ....so why you put them in VAR with their first level (lnrgdp, lngovex, lneds) and not d(lnrgdp) d(lngovex) d(lneds)
Because the VAR algorithm automatically treats them as FD variables. But if you are using the OLS estimator, you must include the "d" for the variables.
TH-cam recently changed the way my content will be monetised. My channel now needs 1,000 subscribers. So it would be amazing if you show your support by both watching my videos and subscribing to my channel if you haven’t done so already. Monetising my videos allows me to invest back into the channel with some new equipment so this small gesture from you will be extremely huge for me. Many thanks for your support….CrunchEconometrix loves to teach, help me stay online.
@Vaibhav Aggarwal Yes, you can. VAR is a system of equation and more often than some of the equations in the model may not be normally distributed.
@@CrunchEconometrix mam in this video you have shown that all variables are stationary at 1st difference. but you have conducted the VAR on raw variables not on the first difference variable. kindly clarify....
Yes Meena, I clarified VAR specifications in the foundation video. You may need to watch the clip. Regards.
Very good teacher. Being a student I benefited from your videos a lot. Allah may keep you happy to the utmost level.
Glad to hear that
Very nice and clear presentation. Many thanks dear Dr
Thanks for the positive feedback, Dr. Alnour.
yor are a great professor, i benefited a lot from your channel
Thanks, Abderrazaq for the encouraging feedback. Appreciated!
Thanks Prof. I have learnt how to perform econometric data analysis from watching your channels. I currently have I(1) and I (0) variables in my system of equations. My objective is to perform a causality test on the macroeconomic variables including oil revenue. Which would suffice: ARDL or VAR.
Hi Lucy, please watch my videos on ARDL starting with "This is how to specify ARDL models". Thanks.
@@CrunchEconometrix I did but my professor has argued that VAR can handle both I(1) & I (0) variables...
Lucy, your Professor is wrong. But you may choose to take his opinion. Read the literature yourself and be convinced on what to do.
@@CrunchEconometrix I will read ma'am. Thanks Prof
really very good indeed. request if one on sign restricted VAR can be added
Hi Manbhanjanrbi, I honestly have no idea what you are referring to.
Great. Why you calculated the optimal lag for each variable individually? Why not all variables at the same time? As VAR model will treat all variables as endogenous?? Thank you
Hi Amina, I just chose to do that. There are other views where I selected the lag length at once for all the variables.
Bless you
Good tutorial mam. Have only one doubt. You explain that VAR logarithm automatically difference the series to first difference. Therefore, you compute the result with only natural log variables in spite of series is stationery at first difference. But please clarify if our variables are stationery at level after taking natural logarithm, is VAR model still differentiate them to first difference or not. Please clarify. Also tell a value from your results from which we can verify that results are automatically diffenced to first level by VAR Alogarithm.
Shubham, there are two Schools of Thought to estimating the VAR model. I followed one of them. Watch the INTRODUCTORY video on VAR specification to know which one. Thanks
Excellent work please make vedio on GARCH MIDAS
Hi Maria, thanks for the encouraging feedback. Deeply appreciated! I appreciate the suggestion and once I understand the procedure I will create simple tutorials on it. Please may I know from where (location) you are reaching me?
CrunchEconometrix I am from China
Thank you so much for your response your vedio are very helpful for researcher keep doing great work
thank you for your videos they are realy helpful
Jones Dahlback Thanks Jo, I'm encouraged by your comments. I'll appreciate if you can help share my link with your social media community and academic networks so that others can know too... 😉
already done
Hi,
thank you for all your videos. Is it possible to make a video for a Baseyian Var analyse?
Thank you!
Thanks Masaid for the positive feedback on my videos. Deeply appreciated. But no promises on Bayesian VAR. Too complicated...may I know from where (location) you are reaching me?
@@CrunchEconometrix Hi, I'm from Belgium.
Wow, good to know Masaidj...kindly share my YT Channel with your students and the Belgian academic community. They'll find my videos very useful...thanks!
@@CrunchEconometrix Do you have knowledge of a good econometric paper for this matter? Thank you in advance!
Masaidj, there are several papers that used FEVD. Simply do a Google search.
Thankyou for your clear explanantion. Prof, what should we do if in our variable there are serial correlation and heteroscedasticity? Is it okay to run the data with VAR?
Hi Ardiani, I suggest you re-estimate the model using higher-order lags.
Thanks Professor. Very useful video. Please let me know if I can use a non-linear equation (A polynomial or a quadratic equation) in a VAR system. Will there be any change in the way I interpret?
I'll say yes, Nithya though I've never used the procedure before but I've come across some papers online. Do a Google search for a couple in that area. Also, Chapter 14 of Gujarati Basic Econometrics is devoted to Non-linear models with worked examples to guide you. Hope these tips are helpful. Take care.
Hello ,your videos are really helpful .nobody explains it your way .I have a question , you have only tested the stationarity at difference one but you didnt make a new object to reflect this change u want before running forecast . If I have a non stationary and non normal situation at level , I make a new object like d(log(variable)) and i continue my study based on that . Please correct me if what i am doing is wrong!!
Hi Ahmad, thanks for the encouraging feedback. Deeply appreciated! Your query is not quite clear. Kindly recast.
@@CrunchEconometrix no need as similar question was addressed and answered in the comments , Thanks a lot .
Alright, thanks.
Thanks for video. I want to ask something. What should we do when we have both stationary and non stationary series ? Is it enough to just take differences of non stationary ones, or do I have to take first differences of all variables ?
Hi Zeki, take the 1st diff of only the NS series.
@@CrunchEconometrix Thank you for answer. But there is something I am lost. I have 5 variables and 3 of them are non stationary. When I take just three of them, since the non stationary series become shorter than than stationary ones, I am not able to run the regression.
@@zekiogulcansengul3593 Yes, you lose a year when you take the first difference and also when lags are taken. These reduce the degrees of freedom. How many years of observations do you have originally?
CrunchEconometrix it is monthly data regarding the inflation. I havr change in fx, commodity price inflation and output gap. Then , I omit the first row and starting from the second row that all variables are available , right ?
You did not respond to my question.
Hi Dr.
I have the same question as Chathuri had that one of the preconditions of the VAR Model/test that data should be stationary. In this respect, your real data/Level form is non-stationary and became stationary at 1st Difference but you tested the VAR on data without converting in 1st difference ..... So my question is either VAR automatically converts the data into first difference and shows the results after taking 1st Difference?
Hi MS, that's exactly what happens! Verify this with your data and see the outcome.
Thanks, Dr. for this clarification as well as for your quick response. You are doing a great job, please keep it up..
Stay high!
@@muhammadshehryar5749 Thanks pal, you're one of those people who inspired me every day to turn out quality videos. Can you please help me share my videos with your friends, students and colleagues. Tell them they have a lot to learn by subscribering to my TH-cam Channel. Thanks in anticipation...gracias! 💕 😊
@@muhammadshehryar5749 hello , have you done that?
Yes, I've.
Can you please teach threshold var and nonlinear impulse response function...pls...
Suggestion noted, thanks.
Maam you have checked lag of each variable individually before running unrestricted VAR. But we know lag test has to be done taking all endogenous variables together. What is the correct procedure?
Another fact you used 1 lag while running VAR with all endogenous variables, but selection criteria require 1 lag for each variable. In case of VECM, one less lag is taken as data is transformed into 1st difference automatically. If out of three variables, say Y shows 2 lags, X shows 1 lags and Z shows 3 lags, in that case what to do if you test lag individaully before running VAR?
Hi Anan, thank you for that sharp observation....and you are right for a VAR model all the variables should be put together while running the unrestricted VAR to obtain the optimal lag lengths and not individually determined. I think I did it that way just to teach/show some people how to determine the optimal lags. Otherwise, it's not ideal. I think I was just fortunate that all the variables have 1 lag (lol)...thanks for the valid observation!
I'll also upload the interpretations of the results very soon....any moment from now.
I'll do my best, Prof...that's a promise!
Hi Dr, in 8:30 you have said that there is no serial correlation based on a probability of 0.8643, what about 0.29?
Thank you
Hi Hamza, no serial correlation cos pvalue > 0.05
@@CrunchEconometrix Hai Prof, what about there is serial correlation in variable? what should we do next?
Re-estimate the model with higher-order lags.
Hello my model is suffering from serial correlation and heteroscedasticity. I am dealing with daily stock price data, and i have 4 variables in the system. However when i am stesting the model for the stability test, in that case is OK. Can i conduct VAR and SVAR in that case?
Hi Florin, I'm unclear about your query.
Hello maam , i already got the answer from the previouse questions. Anyway thank you. I will take care tk share your videos with my friends. Wonderful explanations.
Good to hear, Florin... appreciated! 💗🙏
hello,
can you help me on how to compute the forecast error of government spending?
that is, forecasts of government spending are taken from October publications of the IMF’s WEO. Then, the fiscal spending shocks are identified as the forecast errors of government spending. Thus,
FEi,t=gi,t(actual)-gi,t(forecast)
where gi,t= Gi,t/Yi,t is government spending as a share of GDP. The actual government spending comes from the October WEO of the following year. i-refers to country and t refers to time.
No idea.
HI THANK YOU...
U're welcome, Sa. Please may I know from where (location) you are reaching me?
Is it posiible to run the VAR, if some endogenous variables are integrated of I(0) and other are integrated of I(1) ???
Nope Javeria, but ARDL will fit in with that. May I know from where (location) you are reaching me?
Can VECM be carried out on two variables (one dependent and one independent)?
Kindly watch my videos on VAR and VECM for detailed information on the VECM procedure. Thanks.
Hello. May I ask why did you estimate a VAR model of the variables at level? Correct me if I am wrong, but I thought that running a VAR model for nonstationary series would yield a spurious regression. So, the variables to be estimated in the model should be the first difference of the three variables.
Hi JM, first clue on spurious regression is when DW
@@CrunchEconometrix ohh. Thank you so much!
No worries, JM...I'm happy to assist in every way possible. May I know from where (location) you are reaching me?
can this method be applied to a VECM model as well?
for checking the lags individual variables need to be taken or can we take all the variables together as well?
All variables together when deciding on optimal lags selection.
@@CrunchEconometrix Thank you soo much ma'am.
thanks prof. The data for xlx doesnt appear on the site
Hi Dalisto, the website is being upgraded. Here's the link to obtain the datasets (some are free while some are available on purchase)
👇 cruncheconometrix.com/view/datashop.php
Please what should be done when the null hypothesis of joint normality test for VAR residuals is rejected? Will the results still be significant in the absence of normally distributed errors?
Dino, for me (others may disagree) I worry less about normality once my model passes the serial correlation, heteroscedasticity, stability and multicollinearity (VIF) test.
@@CrunchEconometrix Hi maam. May i know what should i do if i have heteroskedasticity when i run heteroskedaticity test for VAR residuals? Thank you
Hi, transform the variables into their natural logs and use robust standard errors...may I know from where (location) you are reaching me?
@@CrunchEconometrix Thank you for your fast reply.i'm from Malaysia. my variables are already in natural logs when i got the heteroskedasticity. May i know how do i get the option to use robust standard error to run var?
@@thecrystalphantoms Unfortunately I have no idea with EViews. With Stata, including option "robust" takes care of the problem.
Can you give the command of variance decomposition command for stata??? I really need it..
Hi Deepty, you can purchase my dofiles cruncheconometrix.com.ng/shop/
Hello Mam: a VAR model can also include the exogenous variable on the right-hand side. cant it?
Hahahaha Adrienne, I don't about that but we learn everyday. The reality is that, in a VAR model all variables are endogenous. That is, that explain one another.
Hello Mam: here is a ref what I have got:www.personal.psu.edu/hxb11/EasyRegTours/VAR_Tourfiles/VARX.PDF. Also if you look over Eviews popup window there is a box for exogenous var box. That's why I wanted to know what about this exog variable option as you said all must be endo and no exog. Thanks
through optimal lag criteria if i am getting different lags for different period what to do
Your query is unclear, Rational. Please rephrase.
Hi, what if the optimal lags differs?
VAR model employs same number of lags for all the variables.
Since the selected time series are stationary at level 1, why don't you run the VAR with stationary data without going for level data VAR model?? Pls give quick reply
Hi Chathuri, the VAR model is estimated using either the VAR algorithm or the OLS algorithm. If the former, the level of the series is used but the 1st difference if using the latter. I used the VAR algorithm and if you see the outcome, the results are in 1st difference. Hope this clarifies, thanks.
I confused that the VAR model should be called with the stationary data series. You mean the results should be in 1st difference(stationary data)? ok thanks dr
Hi Dr.
I have the same question as Chathuri had that one of the preconditions of the VAR Model/test that data should be stationary. In this respect, your real data/Level form is non-stationary and became stationary at 1st Difference but you tested the VAR on data without converting in 1st difference ..... So my question is either VAR automatically converts the data into first difference and shows the results after taking 1st Difference?
@@muhammadshehryar5749 Yes, the VAR algorithm differences the series during estimation. That's why you use the level not 1st difference. Verify this with your own data and observe your results.
Thanks, Dr. for this clarification as well as for your quick response. You are doing a great job, please keep it up..
Stay high!
If all the variables have no the same optimal lag structure then plz tell me?
Hi Asma, use ARDL technique... Or use FMOLS, and DOLS.
@@CrunchEconometrix thanks mam and can you give me the coding of dynamic nelson Siegel yield curve model?
I have no idea about that technique, Asma.
PLEASE SAY ABOUT MICROFIT ..
Hi Sa, I don't know how to use Microfit.
you say that variables must be stationary ....so why you put them in VAR with their first level (lnrgdp, lngovex, lneds) and not d(lnrgdp) d(lngovex) d(lneds)
Because the VAR algorithm automatically treats them as FD variables. But if you are using the OLS estimator, you must include the "d" for the variables.