Thanks for very for the video. It is very helpful. It has also raised one question. Could you please explain what is structural VAR and how to estimate SVARs in Eviews? That would be of tremendous help. Many thanks
Great video. Thank you Ralf! Before running the VAR, don't we need to check for the presence of a cointegrating relationship? Wouldn't it may be possible that the variables have a long-run relationship and therefore VECM would be more suitable? Thanks!
Hello Ralf, Your video on VAR has been very helpful to me in many ways. Could you please do a video of two stage and three stage least squares. I trust your fans would find it interesting too.
Hello Ralf, Based on your example, your Cholesky ordering was us-german-french. It means, as you said, that the US rate would be contemporaneously affected by the change in German/French rates. Then, in the IRFs, why do shocks in German/French rates not affect the US rate instantaneously? i mean, the graphs show that the US rate was impacted on the 1st period, not the zero-period. Thanks
Hey! When you used Cholesky ordering you pointed out that the ordering matters. Since by changing the ordering you get different results, how do you choose which ordering is appropriate, i.e. is there any test or we should rely on intuition?
There isn't a really easy answer to this. You should certainly consider the economic nature of the series and decide for which variables it makes most sense to restrict the contemporaneous effect to other variables 0. There are also other ways to solve the conundrum. They are often called structural VAR models. But you will have to read up on these in a textbook.
permission asked. is there a video tutorial For troubleshooting heteroscedasticity and autocorrelation simultaneously in eviews? I've read that the method also can be used, among others Seemingly Unrelated Regression (SUR), Panel Corrected Standard Error (PCSE), and GLS. but I am confused For apply these methods in eviews. Thank you for the help. Always success for you.
Dear Becker. Thank you for the video. A quick question that remains. Why did you add the first differences to the variable " D( ) "? Is it possible to do that with any type of variable ? or does that change the interpretation ? thank you for your consideration and your help!
Hi Ralf. Excellent video, thank you so much. Can you you please tell me what to do in the case where all my variables are stationary in levels (by majority of unit root tests) but, when I do my AR Root Test, one or 2 dots fall outside of the unit circle implying my VAR is not stable. Why is this the case and how can I solve this problem so I can interpret my IRFs and Variance Decomposition. Thank you very much.
Hi Ralf. Thank you for your helpful video. I have a question about how to know the effect presented in IRF (with error bands) is statistically significant or not. Thank you and I really appreciate your help.
Hello Ralf !! Hope you're great. A quick question please. In VAR analysis, do the actual VAR estimate coefficients have any empirical meaning, I ask because you went straight to the impulse response. I am currently writing a dissertation using VARs and was wondering if it's important to dwell on VAR estimates. Thank you
Hello Sir....I can not thank you enough. This is excellent job,, so helpful. I have a Question, how does this analysis take care of endogeneity? in other words does is it address spurious correlation from related variables?
Really like your videos. They are so helpful to me... However, have just a small question to ask you coz I am struggling a lot wit an exercise that I have but don't know who on earth to turn to... I have found you, please help
Hello, Ralph, I am trying to make a forecast of the VECM in EViews, and I am stuck because it does not do that directly or I simply cannot find the function (it's the first time I am working with EViews), can you help me?
great tutorial, do you have any tutorial about how to estimate long memory by using GPH test and also Z value (error term), Sign Bias tests & ICSS in eviews?
Hi, Ralf. I have a basic question concerning the VAR model. I wonder whether we need to take out the autocorrelation of the series to make it a white noise before putting it into a VAR model. Or we just put the original stationary series (maybe with time trend and autocorrelation) into the VAR model. Thanks so much for your help in advance!!!
Autocorrelated series are fine. you just need to ensure that the order of your VAR is high enough to eliminate the autocorrelation from the VAR residuals. Time-trending series are a slightly different issue. You may want to include a time trend, if they are trend stationary or may want to different of they are difference stationary. So you need to ensure that you read a textbook on how to test for these things.
Hi Ralf, I was wondering how to determine whether our impulse response estimates are statistically significant? I have computed the IRF table with reported standard errors in Eviews, but I'm assuming it isn't as straight forward as simply dividing the IRF coefficient estimates by the standard errors (as you would for simple OLS regression analysis)? Any help would be greatly appreciated. Thanks, James
+James Ward One often checks whether the confidence interval [IRF - 2* se , IRF + 2* se] includes 0 or not. If 0 is not included you sort of conclude that it is significant.
Good morning my professor Sorry professor, I used the technique of Toda-Yamamoto (by which I have a variable it I (2)), so I estimated two models VAR (1), (1st model (the variables in line) and the 2nd model (the variables in logarithm)), After I chose the model in log, but I found that the results of the causality a bizarre). Are the results of causality (from the VAR model (1 + 2)) a role in the choice of the best model? Best
Hello, How do you decide whether the variable is significant or not? You say, for example that the values on FR are low, but when are the values too low?
As a rough rule of thumb, when looking at an individual coefficient, you want its t-stat [in square brackets] to be either smaller than -2 or larger than 2. But make sure you know how exactly t-tests work.
If I get it right, I can't estimate a VAR with 0 lag. Then, how can I estimate a VAR on Eviews, if I want to see the simultaneous effects? For instance, I want to see the immediate relationship between interest rates and monetary mass. The effect should be immediate, so I need 0 lag. Could you help me please?
Hi, Ralf! Can you make VAR model in Excel? It would be nice if you could show how to make this model for one dependent variable and four independent variables. Thank you very much for your help!
+Ella Wess What you describe (one dependent and four indep vars) isn't really a VAR model as the latter involve multiple dependent vars. It is rather what we call a multiple regression model and there should be plenty of resources to guide you on how to do that in Excel.
+Ralf Becker Hi! Thank you for your answer! What I wrote was just an example. Actually, I was told to use VAR model to evaluate the dependence between returns to 11 indices. Now I got access to Eviews, so it will go well! Thanks!
Your video was really informative for my thesis, however I have few questions: My variables are inflation (CPI), growth(real Gdp) and interest rate (federal fund rate). Should I make the data stationary by removing time and seasonal trend? Also how should I order the three variable? would cholesky dof will order the variable by itself? Waiting for your reply
see my previous comment on whether you want stationary data. regarding the ordering you may have some view from econ theory as to how the variables ought to be ordered. Otherwise you may want to try different orderings and see how sensitive your results are to the ordering.
Dear Ralf, I need to use the variables at level in my VAR as required by my professor. However, the variables are not stationary and have trend in them. Is there some way other than differencing to remove trend?
Thank You for your reply. How can I include deterministic time trend? Do you have some video on it? When I am making VAR should I just include @trend along with other variables?
Hey Omer! Could you include a deterministic trend in your model? In my thesis I need to use the serie in levels. In case you did it, how did you do? Thanks in advance.
+Godsaw you want the level series to have a linear tend, I.e. not look explosive. Sometimes this will only be the case after taking logs and that is when you use the log of a series
hi! how do I remove serial correlation in the VAR environment? I have differenced the variables and I obtained serial correlation problem in the model. I have then take the log values of the variables which I have differenced also because of unit roots but the models still have serail correlation issues. How do I practically removed serial correlation in a VAR model using Eviews software?
bonjour monsieur je suis étudiante, et je suis entrain de faire un mémoire de fin d'étude qui porte sur le niveau d'adéquation des réserves de changes, je veux savoir que ce que ça veut dire l'estimation à long terme? , le but de test de cointégration?
In some sense you are estimating a simultaneous equation model as you are allowing error terms to be correlated at time t. But of course that by itself tells you noting about the contemporaneous relationships as you cannot say anything about directions of influences. That is why, when you create impulse response functions you need to impose an ordering which is essentially imposing some structure. And you can make different type of restrictions to identify contemporaneous info (structural VARs) but in the end you need to make some a priori restrictions.
i have a voice of a baby,thank your share.but,why must to difference the variable，some teacher told me,it maybe lost the variable information if you do it .others also told me not care the non-stationary of variable.i donot know,yet.so i hope you can help me.
This is a not straightforward question and opinions are divided. Chris Sims for instance argues that it is ok to do so in particular perhaps if forecasting is your main purpose. but then you ought to consider whether forecasting in the non-stationary form is really the best. It may not be as robust to structural changes as a model in stationary variables.
Hi, I run a VAR model and it turns out the R2 is .96. I feel so weird about this super huge R2, but I do all the tests before to make sure the two series are stationary, the appropriate length of lags, the stationarity of the VAR system. Can anyone help to solve my puzzle here? Thanks a lot in advance!!!
Perhaps they have time trends. When you tested for stationarity you may have included a time trend and then you found that they have no stochastic trend (beyond the time trend) but when you then use the data in a VAR you will find that the time trend explains most of the variation (even if you do not include an explicit time variable in your model). If that was the case you may want to detrend the data first.
Hi, Ralf. Thanks for your kind reply. Yes, I did include a time trend when checking the stationarity. I think the spurious high R2 might result from the fact I didn't detrend the data. I will try this solution. Again, thanks so much for your answering, Ralf!!!!!!!!
Hi, Dr. Becker. I followed your suggestions before. I have two stationary series, and one of them is stationary with a time trend. So I detrend this series, let's say series A, but don't do anything to the other series B. Then I run a VAR model, the R2 of the model with series B as dependent variable looks ok, 0.09, but the R2 of another model with detrended series A as dependent variable is still super big, 0.86. But later the Granger causality test shows that the series B granger causes the detrended series A. In this case, I am wondering whether such R2 is acceptable. Thanks for your attention in advance!
I want to add a piece of information. The unit root analysis shows the series B is stationary with a constant. So I am wondering whether I need to deduct this constant from the original values of series B before conducting any analysis. Thanks.
Thanks for very for the video. It is very helpful. It has also raised one question. Could you please explain what is structural VAR and how to estimate SVARs in Eviews? That would be of tremendous help. Many thanks
Great video. Thank you Ralf! Before running the VAR, don't we need to check for the presence of a cointegrating relationship? Wouldn't it may be possible that the variables have a long-run relationship and therefore VECM would be more suitable? Thanks!
Hello Ralf, Your video on VAR has been very helpful to me in many ways. Could you please do a video of two stage and three stage least squares. I trust your fans would find it interesting too.
Hello Ralf,
Based on your example, your Cholesky ordering was us-german-french. It means, as you said, that the US rate would be contemporaneously affected by the change in German/French rates. Then, in the IRFs, why do shocks in German/French rates not affect the US rate instantaneously? i mean, the graphs show that the US rate was impacted on the 1st period, not the zero-period.
Thanks
you should do a vid about estimating a SVAR
Hey! When you used Cholesky ordering you pointed out that the ordering matters. Since by changing the ordering you get different results, how do you choose which ordering is appropriate, i.e. is there any test or we should rely on intuition?
There isn't a really easy answer to this. You should certainly consider the economic nature of the series and decide for which variables it makes most sense to restrict the contemporaneous effect to other variables 0.
There are also other ways to solve the conundrum. They are often called structural VAR models. But you will have to read up on these in a textbook.
Thank you for your answer and my suggestion for you next video is SVAR. Best regards!
permission asked. is there a video tutorial For troubleshooting heteroscedasticity and autocorrelation simultaneously in eviews? I've read that the method also can be used, among others Seemingly Unrelated Regression (SUR), Panel Corrected Standard Error (PCSE), and GLS. but I am confused For apply these methods in eviews. Thank you for the help. Always success for you.
Many many thanks Ralf, for such an incredible informative video!!!!
Please keep up the good work!
Very good description, thank you for your helps and support
Dear Becker. Thank you for the video. A quick question that remains. Why did you add the first differences to the variable " D( ) "? Is it possible to do that with any type of variable ? or does that change the interpretation ? thank you for your consideration and your help!
Hi Ralf. Excellent video, thank you so much. Can you you please tell me what to do in the case where all my variables are stationary in levels (by majority of unit root tests) but, when I do my AR Root Test, one or 2 dots fall outside of the unit circle implying my VAR is not stable. Why is this the case and how can I solve this problem so I can interpret my IRFs and Variance Decomposition.
Thank you very much.
Good presentation @Ralf.
Hi Ralf. Thank you for your helpful video. I have a question about how to know the effect presented in IRF (with error bands) is statistically significant or not. Thank you and I really appreciate your help.
thank you so much, please keep doing eviews tutorials
Hello Ralf !! Hope you're great. A quick question please. In VAR analysis, do the actual VAR estimate coefficients have any empirical meaning, I ask because you went straight to the impulse response. I am currently writing a dissertation using VARs and was wondering if it's important to dwell on VAR estimates. Thank you
Hello Sir....I can not thank you enough. This is excellent job,, so helpful. I have a Question, how does this analysis take care of endogeneity? in other words does is it address spurious correlation from related variables?
Really like your videos. They are so helpful to me... However, have just a small question to ask you coz I am struggling a lot wit an exercise that I have but don't know who on earth to turn to... I have found you, please help
Hello, Ralph, I am trying to make a forecast of the VECM in EViews, and I am stuck because it does not do that directly or I simply cannot find the function (it's the first time I am working with EViews), can you help me?
great tutorial, do you have any tutorial about how to estimate long memory by using GPH test and also Z value (error term), Sign Bias tests & ICSS in eviews?
+Jiner Mendoza no I don't
Hi, Ralf. I have a basic question concerning the VAR model. I wonder whether we need to take out the autocorrelation of the series to make it a white noise before putting it into a VAR model. Or we just put the original stationary series (maybe with time trend and autocorrelation) into the VAR model. Thanks so much for your help in advance!!!
Autocorrelated series are fine. you just need to ensure that the order of your VAR is high enough to eliminate the autocorrelation from the VAR residuals. Time-trending series are a slightly different issue. You may want to include a time trend, if they are trend stationary or may want to different of they are difference stationary. So you need to ensure that you read a textbook on how to test for these things.
Thanks for your suggestions! I will look at all these issues you mentioned. Thank you so much.
Hi Ralf,
I was wondering how to determine whether our impulse response estimates are statistically significant? I have computed the IRF table with reported standard errors in Eviews, but I'm assuming it isn't as straight forward as simply dividing the IRF coefficient estimates by the standard errors (as you would for simple OLS regression analysis)?
Any help would be greatly appreciated.
Thanks,
James
+James Ward One often checks whether the confidence interval [IRF - 2* se , IRF + 2* se] includes 0 or not. If 0 is not included you sort of conclude that it is significant.
Good morning my professor
Sorry professor, I used the technique of Toda-Yamamoto (by which I have a variable it I (2)), so I estimated two models VAR (1), (1st model (the variables in line) and the 2nd model (the variables in logarithm)), After I chose the model in log, but I found that the results of the causality a bizarre). Are the results of causality (from the VAR model (1 + 2)) a role in the choice of the best model?
Best
Hello,
How do you decide whether the variable is significant or not? You say, for example that the values on FR are low, but when are the values too low?
As a rough rule of thumb, when looking at an individual coefficient, you want its t-stat [in square brackets] to be either smaller than -2 or larger than 2. But make sure you know how exactly t-tests work.
Thank you for the quick answer :-)
Thank you Ralf, very useful.
are there any conditions that series must respect before running a var model?
Thank You! You saved my semester LOL
If I get it right, I can't estimate a VAR with 0 lag. Then, how can I estimate a VAR on Eviews, if I want to see the simultaneous effects?
For instance, I want to see the immediate relationship between interest rates and monetary mass. The effect should be immediate, so I need 0 lag. Could you help me please?
Hi, Ralf! Can you make VAR model in Excel? It would be nice if you could show how to make this model for one dependent variable and four independent variables.
Thank you very much for your help!
+Ella Wess What you describe (one dependent and four indep vars) isn't really a VAR model as the latter involve multiple dependent vars. It is rather what we call a multiple regression model and there should be plenty of resources to guide you on how to do that in Excel.
+Ralf Becker
Hi! Thank you for your answer! What I wrote was just an example. Actually, I was told to use VAR model to evaluate the dependence between returns to 11 indices. Now I got access to Eviews, so it will go well! Thanks!
Hello Ralf, thank you very much for your work! Do you know if there is a way to construct confidence bands for the bayesian var?
+Sergey Kozik I'm afraid, I don't.
Hi, Ralf .thank you for this video, I have a question. How can I remove serial correlation in the VAR model? What is the reason ?
+Ashkhen Nigoyan This typically means that you have to estimate a VAR with a higher order.
+Ralf Becker But when I increase lag order, in that case my model is not normally distributed. Is it important in the VAR model?
+Ashkhen Nigoyan Normality isn't really a big issue. It is more important to eliminate the autocorrelation.
Good presentation thank you
Your video was really informative for my thesis, however I have few questions:
My variables are inflation (CPI), growth(real Gdp) and interest rate (federal fund rate). Should I make the data stationary by removing time and seasonal trend? Also how should I order the three variable? would cholesky dof will order the variable by itself?
Waiting for your reply
see my previous comment on whether you want stationary data. regarding the ordering you may have some view from econ theory as to how the variables ought to be ordered. Otherwise you may want to try different orderings and see how sensitive your results are to the ordering.
Dear Ralf,
I need to use the variables at level in my VAR as required by my professor. However, the variables are not stationary and have trend in them. Is there some way other than differencing to remove trend?
if they have linear trends in time you could include a deterministic time trend.
Thank You for your reply. How can I include deterministic time trend? Do you have some video on it? When I am making VAR should I just include @trend along with other variables?
Hey Omer! Could you include a deterministic trend in your model? In my thesis I need to use the serie in levels. In case you did it, how did you do? Thanks in advance.
Thanks for this video, it helped me a lot !
So good, thank you.
How can run Panel VAR model using EVIEWS and compute impulse response and Variance Decomposition
Not related. I can hear a baby in the background. Too cute.
Thanks for the video! Helps a lot.
this is so good!!!
How would you choose between doing the first difference and the log difference of the data?
+Godsaw you want the level series to have a linear tend, I.e. not look explosive. Sometimes this will only be the case after taking logs and that is when you use the log of a series
Hi, thank you. How can I convert negative variables into natural log in eviews?
Shalom S You can't. Natural logs do not contain negative values, ever.
thank you
what if in autocorrelation test my Prob es not 0.0000 is 0.0007?? is not significative?
hi! how do I remove serial correlation in the VAR environment? I have differenced the variables and I obtained serial correlation problem in the model. I have then take the log values of the variables which I have differenced also because of unit roots but the models still have serail correlation issues. How do I practically removed serial correlation in a VAR model using Eviews software?
+berzanna seydou Ouattara increase the number of lags until any autocorrelation issues are resolved.
bonjour monsieur
je suis étudiante, et je suis entrain de faire un mémoire de fin d'étude qui porte sur le niveau d'adéquation des réserves de changes, je veux savoir que ce que ça veut dire l'estimation à long terme? , le but de test de cointégration?
Could anyone please send a link for the data ( US, German and French interest rates) used in the video?
check the notes for the video
can i use the VAR model to estimate simultaneous equationsthank you
In some sense you are estimating a simultaneous equation model as you are allowing error terms to be correlated at time t. But of course that by itself tells you noting about the contemporaneous relationships as you cannot say anything about directions of influences. That is why, when you create impulse response functions you need to impose an ordering which is essentially imposing some structure.
And you can make different type of restrictions to identify contemporaneous info (structural VARs) but in the end you need to make some a priori restrictions.
i have a voice of a baby,thank your share.but,why must to difference the variable，some teacher told me,it maybe lost the variable information if you do it .others also told me not care the non-stationary of variable.i donot know,yet.so i hope you can help me.
Can we estimate VAR even if we don't have a stationnary data
This is a not straightforward question and opinions are divided. Chris Sims for instance argues that it is ok to do so in particular perhaps if forecasting is your main purpose. but then you ought to consider whether forecasting in the non-stationary form is really the best. It may not be as robust to structural changes as a model in stationary variables.
Thank you sir ! :)
Hi, I run a VAR model and it turns out the R2 is .96. I feel so weird about this super huge R2, but I do all the tests before to make sure the two series are stationary, the appropriate length of lags, the stationarity of the VAR system. Can anyone help to solve my puzzle here? Thanks a lot in advance!!!
Perhaps they have time trends. When you tested for stationarity you may have included a time trend and then you found that they have no stochastic trend (beyond the time trend) but when you then use the data in a VAR you will find that the time trend explains most of the variation (even if you do not include an explicit time variable in your model). If that was the case you may want to detrend the data first.
Hi, Ralf. Thanks for your kind reply. Yes, I did include a time trend when checking the stationarity. I think the spurious high R2 might result from the fact I didn't detrend the data. I will try this solution. Again, thanks so much for your answering, Ralf!!!!!!!!
Hi, Dr. Becker. I followed your suggestions before. I have two stationary series, and one of them is stationary with a time trend. So I detrend this series, let's say series A, but don't do anything to the other series B. Then I run a VAR model, the R2 of the model with series B as dependent variable looks ok, 0.09, but the R2 of another model with detrended series A as dependent variable is still super big, 0.86. But later the Granger causality test shows that the series B granger causes the detrended series A. In this case, I am wondering whether such R2 is acceptable. Thanks for your attention in advance!
I want to add a piece of information. The unit root analysis shows the series B is stationary with a constant. So I am wondering whether I need to deduct this constant from the original values of series B before conducting any analysis. Thanks.