great Video. thank you so much for excellent explenation i was wondering what can we do if we failed the tests what if our data has serial correlation, heteroskedasticity or non normality ? how can we fix it ? is there a video about that ? thank you
Yo Cheers for the video Justin! however i have a question about the tests, how do you choose the lags.pt and lags.multi to be 12? does it depend on the original var() estimation ? Thanks in advance and keep it up with your videos
Hi! Regarding choosing the number of lags.pt, you can specify any number of lags you'd like to test for. I just chose 12 as this was the most 'default' order used. However, you may opt to match the lag length selected or go above it. In practice, we typically choose a higher lags.pt or lags.multi over the lag order we ended up using. Hope this helps!
Thank you so much for the great videos, they really helped me a lot! Could elaborate a little more on the stability test? What it exactly does it calculate? Thanks in advance :)
Hi Justin, I really like your videos. Great as always! I was wondering whether there is a possibility to conduct a ramsey RESET test in vars or any other kind of omitted variables test? Thanks in advance for a short reply! Best, André
Hi Justin, These videos are amazing! I have enjoyed following you step by step. I have a question though, how do I add moderating variables to my model. Say in addition to the "Unemployment" and "GDP" above, I wanted to add two moderating variables "Consumer Price Index" and "Exchange Rate", how may I do this. Looking forward to your response. Thank you, Justin. /Feyi
Feyitimi Anthony Hi Feyi! Merely add them as variables by adding both of these when you bind the variables using cbind(). Just be sure to also declare both variables as a time series variable and you should be good to go. Hope this helps.
Hi Justin, great video!! Thanks for that. I got a message trying to do some stability testing. "Could not find function stability?" Any solution on that?
Hi, I would suggest to try and maybe transform the variables first through the use natural log, log difference, or difference. For the significance, you may want to look at the IRFs too, as even if the VAR coefficients may be insignificant, something may be concluded from the IRFs.
Hi again ! So I now completed this video with my data (and already started the next one, it's insane how clear you are compared to other youtubers btw) and it appears all the diagnostic tests failed (p-value < 0,05 each time). Does this mean the results I'll get for the applications in the next video will have no meaning ? And is there anything I could do about this ? Thanks !!
Hi! Maybe you could try to take the log of some variables or maybe add a few more controls. You can still be able to do the results of the one in the next video but yes, the accuracy will be called into question since the diagnostics failed. Best of luck!
Hi! Justin thankyou so much for your videos I have a doubt for estimating VAR don't we require the series to be stationary ? Here when you plotted the two series they were clearly non stationary , can you please help me out with it?
hi sir i my variables are non stationary im making stationary I(1) by using ADF test differencing estimate var model with difference series now Rseq is very low and digonostic checking are faild non normal residuals hetrocesdesticity ans serial correlation now i do kindly response to me ????
hi, thank you for video,i run the model VAR,but the test for Correlation and Hetrochdasticity both reject in my model and my model suffer from both ,now what could i do for solving the problem?thank you in advance.
Dear Justin, when executing the code of ARCH appears: Error in solve.default(omega0) : system is computationally singular: reciprocal condition number = 6.32316e-18. I searched in google but there is no answer on omega0! Thanks in advance
Stability is a univariate time series concept that is a stronger condition than stationarity. If a time series is stable then it is also stationary (vice versa is not necessarily true). Naturally, stability can be considered in a multivariate time series model. The simple implication is: if your model is unstable then one or more of the time series within your model is non-stationary, i.e. one or more time series in your model was generated by a process with unit roots. The simplest way to tackle this problem is to find out which one(s) are non-stationary and difference them and test for stability once more and hope that you rectify it. Theoretically, it's not necessarily going to work, but in reality, almost all time series we use are well-behaved, so treating stability as equivalent to stationarity and using differencing will likely fix the non-stationarity and instability problem. To understand stability at a basic level: if the innovations (error shocks) of your time series have a permanent effect on the future value then your series is unstable, whereas if they decay to zero over time then you have stability. There is a unique situation of having instability with stationarity - it's a weird theoretical situation that requires having a time series that goes infinitely into the past: it's abstract and not really something you're going to deal with real life data, so just treat stability and stationarity conditions as equivalent for practical purposes.
![ปฏิเสธไม่ไหว (Crush on you) - LIPTA feat. No One Else [Official MV]](http://i.ytimg.com/vi/NBw1jF342vw/mqdefault.jpg)
great Video. thank you so much for excellent explenation
i was wondering what can we do if we failed the tests
what if our data has serial correlation, heteroskedasticity or non normality ? how can we fix it ? is there a video about that ?
thank you
Thank You for these videos, Justin. They are helping me a lot!
Yo Cheers for the video Justin! however i have a question about the tests, how do you choose the lags.pt and lags.multi to be 12? does it depend on the original var() estimation ? Thanks in advance and keep it up with your videos
Hi! Regarding choosing the number of lags.pt, you can specify any number of lags you'd like to test for. I just chose 12 as this was the most 'default' order used. However, you may opt to match the lag length selected or go above it. In practice, we typically choose a higher lags.pt or lags.multi over the lag order we ended up using. Hope this helps!
Thank you so much for the great videos, they really helped me a lot! Could elaborate a little more on the stability test? What it exactly does it calculate? Thanks in advance :)
Hey Justin, I really enjoy your videos and learnt a lot from them. I do have a small question as well. Why did you take "OLS CUSUM" in stability test?
Thank you for your videos. Do you have any videos that show how to perform Threshold VARs and to derive the GIRFs?
Hi, Can you do a series of what to do if all the diagnostic test fails? Thanks!
Hi Justin, I really like your videos. Great as always! I was wondering whether there is a possibility to conduct a ramsey RESET test in vars or any other kind of omitted variables test? Thanks in advance for a short reply! Best, André
Hi Justin,
These videos are amazing! I have enjoyed following you step by step. I have a question though, how do I add moderating variables to my model. Say in addition to the "Unemployment" and "GDP" above, I wanted to add two moderating variables "Consumer Price Index" and "Exchange Rate", how may I do this. Looking forward to your response. Thank you, Justin.
/Feyi
Feyitimi Anthony Hi Feyi! Merely add them as variables by adding both of these when you bind the variables using cbind(). Just be sure to also declare both variables as a time series variable and you should be good to go. Hope this helps.
Excellent video. Could you please add the sequential video links in the description so that one can follow and use the links? Thanks.
The link to the playlist is here: th-cam.com/play/PLEuzmtv9IuT88IrQDHqz100twn85HxWfV.html
@@JustinEloriaga Thanks.
Hi Justin, great video!! Thanks for that. I got a message trying to do some stability testing. "Could not find function stability?" Any solution on that?
Really appreciate both the videos.. Can you tell me if there is completely no significance in Var result lags, how does it affect the model?
Hi, I would suggest to try and maybe transform the variables first through the use natural log, log difference, or difference. For the significance, you may want to look at the IRFs too, as even if the VAR coefficients may be insignificant, something may be concluded from the IRFs.
@@JustinEloriaga Thanks for your response, really appreciated. Can you tell me if I can run RMSE , MAE and MAPE in VAR?
@@mashalsuchwani5605 yes, i think you can use the accuracy function
How could I know hoy many lags should I use in the serial correlation test?
Hi again ! So I now completed this video with my data (and already started the next one, it's insane how clear you are compared to other youtubers btw) and it appears all the diagnostic tests failed (p-value < 0,05 each time).
Does this mean the results I'll get for the applications in the next video will have no meaning ? And is there anything I could do about this ? Thanks !!
Hi! Maybe you could try to take the log of some variables or maybe add a few more controls. You can still be able to do the results of the one in the next video but yes, the accuracy will be called into question since the diagnostics failed. Best of luck!
Ok thanks mate !
Hi ...sir I have 6 variables can O use VAR at a time on it?
Hi! Justin thankyou so much for your videos I have a doubt for estimating VAR don't we require the series to be stationary ? Here when you plotted the two series they were clearly non stationary , can you please help me out with it?
Take differential with the diff() command
What is the stability test for VECM?
hi sir i my variables are non stationary im making stationary I(1) by using ADF test differencing estimate var model with difference series now Rseq is very low and digonostic checking are faild non normal residuals hetrocesdesticity ans serial correlation now i do kindly response to me ????
How I do correlation analysis in VAR model?
hi,
thank you for video,i run the model VAR,but the test for Correlation and Hetrochdasticity both reject in my model and my model suffer from both ,now what could i do for solving the problem?thank you in advance.
Hello, thank you for your comment. Maybe consider taking the natural log or the differenced value of your variables. This might alleviate the problem
Dear Justin, when executing the code of ARCH appears: Error in solve.default(omega0) :
system is computationally singular: reciprocal condition number = 6.32316e-18.
I searched in google but there is no answer on omega0! Thanks in advance
reduce the number of lags.multi
If stability test fails what is the implication
Stability is a univariate time series concept that is a stronger condition than stationarity. If a time series is stable then it is also stationary (vice versa is not necessarily true). Naturally, stability can be considered in a multivariate time series model. The simple implication is: if your model is unstable then one or more of the time series within your model is non-stationary, i.e. one or more time series in your model was generated by a process with unit roots. The simplest way to tackle this problem is to find out which one(s) are non-stationary and difference them and test for stability once more and hope that you rectify it. Theoretically, it's not necessarily going to work, but in reality, almost all time series we use are well-behaved, so treating stability as equivalent to stationarity and using differencing will likely fix the non-stationarity and instability problem. To understand stability at a basic level: if the innovations (error shocks) of your time series have a permanent effect on the future value then your series is unstable, whereas if they decay to zero over time then you have stability.
There is a unique situation of having instability with stationarity - it's a weird theoretical situation that requires having a time series that goes infinitely into the past: it's abstract and not really something you're going to deal with real life data, so just treat stability and stationarity conditions as equivalent for practical purposes.