I just want to say... thank you so so much for your videos..... my professor would talk for 3 hours and not be explain what you layout beautifully crystal clear in 5 minutes..... thank you so much for the effort.
Hi, thanks for your comment. A process is stationary if delta is less than zero. If delta is equal to zero, this is indicative of a unit root (non-stationarity). Hope that helps! Thanks, Ben
Ben, I think you should make a video about what exactly a unit root means. I am taking an undergraduate course in ecnoometrics and so many of my coursemates including me, wish there was some really nice and high level explanation of it.
I have a question regarding the computation of the test: I include the constant by subtracting all y(i) with y(0) such that y(0) = 0 (is that correct?) and compute the delta as (sum of y(i)*y(i-1) / sum of y(i)^2) - 1 (is that correct?). How do I account for the error terms? Cheers!
Hi, thanks for your message. My apologies for the late reply - I have been quite busy over the past week. I would recommend you try the free software package Gretl - if you have any specific questions then please feel free to send me a message, and I will endeavour to reply. Thanks, Ben
Hi Ben, I'm watching both videos (Unit root and ADF Test) again and again but I really miss the very last point: how to compare the T-statistic of regression over a Dickey Fuller Test in practice. Do you think you could make a video using an Excel file? I think it would help me dissipate my last doubts about the Pairs Trading stationary assumptions. Thanks
brother i am working on time series and i am in great problem especially beacuse of lack of statistical software and in epth knowledge of time series. i have to complete my desertation with in a very short time period. i need your valuable and precious help. can you help plzz ?
Great vid! What is exactly the difference between AR(1) and AR(2). I am working with Eviews. Is it simply a lag increase from 1 to 2? How should I type it in in Eviews? Simply change lag to 2, or ... c ...(-1) ...(-2)?
does it mean by producing the ADF could also decide how many lags that i should include in the model by looking into the significance of the sum of lag?
Hi, thanks for your message. No, that is not the case. The reason for including lags in the ADF is to 'mop up' for any other factors which are important in determining that particular variable. In a regression once you have included these factors, it may not be necessary to include any lags of variables. Hope that helps, Ben
Hi, No it should be delta=1 in this example because we have first-differenced the series, meaning that a unit root will have delta=0. Hope that helps! Best, Ben
Hi Ben, when you try to determine the number of \beta_i\Delta\y_{t-i} to include in the regression, shouldn't you test from a general specification to a specific one? Otherwise the estimate of \beta_i don't converge because there's a omited variable bias if too few lags are included and your t-test or F-test are invalid?
+tom h Hi Tom, thanks for your comment. Not sure I understand though. The null is rho = 1 (non-stationarity), meaning that delta = rho - 1, is zero. The alternative is that rho < 1, in other words delta < 0. Hope that makes sense. Best, Ben
+Ben Lambert One doubt: take the equation: deltaYt= alpha + TetaYt-1 +BetaDeltaYt-1.... I thought that the ADF test was about the distribution of Teta and not Beta. I mean, I thought that the equation was to be re-written according to Teta (i.e TetaYt-1=deltaYt-alpha-betaDeltaYt-1...) and then take and average value for all these Tetas found and look at its distribution on a ADF table value. Am I wrong? Finally, why I often hear of ADF test made on residuals? isn't that for just Engle Granger tets?
hey ben, I am not sure whether you are going to check this comment, firstly, the videos are really informative and helpful comparing to all other videos I've ever watched. here is my question. if the series has a or two unit roots, does that mean the series is non-stationary? thanks a lot.
Thank you so much for your presentations. I learned and improved my knowledge. Just one thing I want to mention is that maybe the term "i=1" under the Sigma sign should be "i=2". Am I correct? Regards.
No problem Ben. I also want to tell you it would be so nice if you could develop a video explaining how VECM model, the long term cointegration and it's short term variables are interpreted. Also why the minus term means that there is cointegration and how the mechanism of convergence to the long term mean actually works on an example. There is no such video about VECM and Cointegration.
+Gert Thielemans Thanks for your comment. No, all processes where there is a non-zero delta (in the above notation) are AR(1). All AR(1) means is 'Autoregressive of order 1'. In other words, a variable is being regressed on its first order lag. It doesn't make any qualifications on the constant of that regression. Hope that helps! Best, Ben
1:16 "it turns out that the correct thing to do it THIS *does random thing*" cool but could you please explain why it is that it "turns out" that you have to put the Delta on the y(t-2)? otherwise this video is just as bad as my book that doesn't explain anything
I just want to say... thank you so so much for your videos..... my professor would talk for 3 hours and not be explain what you layout beautifully crystal clear in 5 minutes.....
thank you so much for the effort.
Ben Lambert for president!
for real
Hi, thanks for your comment. A process is stationary if delta is less than zero. If delta is equal to zero, this is indicative of a unit root (non-stationarity). Hope that helps! Thanks, Ben
Ben, I think you should make a video about what exactly a unit root means. I am taking an undergraduate course in ecnoometrics and so many of my coursemates including me, wish there was some really nice and high level explanation of it.
Hi, Thanks for your message and idea. I will add your suggestion to my list of videos I intend to make. Best, Ben
such a useful video. trying to teach myself this for my thesis and this saved me literally hours, maybe even days, of reading journals, thanks.
Why is the trend stationary if delta = 0? And why is the alternative hypothesis that delta < 0 and not that delta is not equal to 0? Thank you!
Please can you make a video series on Panel Data Analysis...?
Cant you use AIC or BIC for using the lagged variables?
How to understand that doing Augmented Dickey Fuller test with lag = 0?
I have a question regarding the computation of the test: I include the constant by subtracting all y(i) with y(0) such that y(0) = 0 (is that correct?) and compute the delta as (sum of y(i)*y(i-1) / sum of y(i)^2) - 1 (is that correct?). How do I account for the error terms? Cheers!
Hi, thanks for your message. My apologies for the late reply - I have been quite busy over the past week. I would recommend you try the free software package Gretl - if you have any specific questions then please feel free to send me a message, and I will endeavour to reply. Thanks, Ben
Hi Ben, I'm watching both videos (Unit root and ADF Test) again and again but I really miss the very last point: how to compare the T-statistic of regression over a Dickey Fuller Test in practice. Do you think you could make a video using an Excel file? I think it would help me dissipate my last doubts about the Pairs Trading stationary assumptions. Thanks
sir,, good noon. can u give an example using dickey fuller test and augmented dickey fuller test by manual competition
Hello! Why do we add the lag of the delta term instead of the second lag of Yt?
Thank you for your video, is it not the case that the hypothese has to be H0: Ro = 1 and H1: Ro < 1 ???
I was thinking the exact same thing.
No, you cannot easily test that hypothesis on PC. The PC estimates delta, which gets the same result.
brother i am working on time series and i am in great problem especially beacuse of lack of statistical software and in epth knowledge of time series. i have to complete my desertation with in a very short time period. i need your valuable and precious help. can you help plzz ?
Great vid!
What is exactly the difference between AR(1) and AR(2). I am working with Eviews. Is it simply a lag increase from 1 to 2?
How should I type it in in Eviews? Simply change lag to 2, or ... c ...(-1) ...(-2)?
does it mean by producing the ADF could also decide how many lags that i should include in the model by looking into the significance of the sum of lag?
Hi, thanks for your message. No, that is not the case. The reason for including lags in the ADF is to 'mop up' for any other factors which are important in determining that particular variable. In a regression once you have included these factors, it may not be necessary to include any lags of variables. Hope that helps, Ben
hi ben shouldnt the null in the ADF be delta=1 rather than 0? i was comparing notes and got confused.
Hi, No it should be delta=1 in this example because we have first-differenced the series, meaning that a unit root will have delta=0. Hope that helps! Best, Ben
+Ben Lambert so the null is one or zero? got confused here!!
@@mennaelhefnawy6040 SHOULD BE ZERO
Hi Ben, when you try to determine the number of \beta_i\Delta\y_{t-i} to include in the regression, shouldn't you test from a general specification to a specific one? Otherwise the estimate of \beta_i don't converge because there's a omited variable bias if too few lags are included and your t-test or F-test are invalid?
under the null hypothesis, p
+tom h Hi Tom, thanks for your comment. Not sure I understand though. The null is rho = 1 (non-stationarity), meaning that delta = rho - 1, is zero. The alternative is that rho < 1, in other words delta < 0. Hope that makes sense. Best, Ben
+Ben Lambert One doubt: take the equation: deltaYt= alpha + TetaYt-1 +BetaDeltaYt-1....
I thought that the ADF test was about the distribution of Teta and not Beta. I mean, I thought that the equation was to be re-written according to Teta (i.e TetaYt-1=deltaYt-alpha-betaDeltaYt-1...) and then take and average value for all these Tetas found and look at its distribution on a ADF table value. Am I wrong? Finally, why I often hear of ADF test made on residuals? isn't that for just Engle Granger tets?
Amazing video as always, very clear and concise. Keep up the helpful work :-)
If delta =0 ( we accept the null hypothesis ) that mean the time serie is a random walk ???
hey ben, I am not sure whether you are going to check this comment,
firstly, the videos are really informative and helpful comparing to all other videos I've ever watched.
here is my question.
if the series has a or two unit roots, does that mean the series is non-stationary?
thanks a lot.
yes
Thank you so much for your presentations. I learned and improved my knowledge. Just one thing I want to mention is that maybe the term "i=1" under the Sigma sign should be "i=2". Am I correct? Regards.
+Okan Aybar Hi, sorry just saw this - it depends on whether you start counting at zero or 1. I have assumed zero here; perhaps confusingly! Best, Ben
No problem Ben. I also want to tell you it would be so nice if you could develop a video explaining how VECM model, the long term cointegration and it's short term variables are interpreted. Also why the minus term means that there is cointegration and how the mechanism of convergence to the long term mean actually works on an example. There is no such video about VECM and Cointegration.
+Okan Aybar Thanks for your message. Yes, those are on my list of things to do. Hopefully, I can get to those pretty soon. Best, Ben
Thank you for your videos! They are really helplful!
AR(1)? I thought that only processes where there is no alpha are AR(1)?
+Gert Thielemans Thanks for your comment. No, all processes where there is a non-zero delta (in the above notation) are AR(1). All AR(1) means is 'Autoregressive of order 1'. In other words, a variable is being regressed on its first order lag. It doesn't make any qualifications on the constant of that regression. Hope that helps! Best, Ben
allright, thank you very much for your swift reply.
very clear explanation, thank you for sharing ^^
Great presentation, thanks bro it was helpful!!!
Can you explain this with a proper question? With a generic formula - it becomes hard to apply to a question.
1:16 "it turns out that the correct thing to do it THIS *does random thing*"
cool but could you please explain why it is that it "turns out" that you have to put the Delta on the y(t-2)? otherwise this video is just as bad as my book that doesn't explain anything
This confused me aswell😂😂. I'm a bit late.
Good video -
This doesn't actually help one do it in practise :(
too compicated