RUclips 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.
I don't know how much I should thank to you for your effort and clear explanation!! I'm whipping myself to finish my thesis during self-quarantine and thanks to you I can proceed the analysis that I have been stuck. Your videos are not only helpful in terms of using stata technically but also understanding fundamental theoretically!
This video may have just been the key to my bachelor thesis! Thanks a lot for the awesome content and the clear explanation! (coming from London) ;) will share in my university cause I know a lot of people struggling with this sensitive subject
Thanks for the positive feedback, Maelian...deeply appreciated! Happy to hear that you are willing to share the clip with your colleagues...gracias!!! :)
This video is a life-saver! When i run the ADF test, I dont know why the absolute t-statistic value is always smaller than the absolute critical value(no matter 1% 5% 10%), which is disappointing since i wanna reject Ho. Your video explained this quite clear and i got the ideal output(reject Ho and the variable is stationary) after using difference of log form, according to this video. Thank you !
Hi Ngozi, I am refreshing my time series econometrics of non-stationary series and I can't stop watching your videos :-) Keep up the great work! Greetings from Switzerland - Claudio
Hi, I want to as you a question, why in 5:36, you only do the regression towards two variable lnpce and lnpdi, is there any reason why you don't include gdp on the regression?
Searching for a channel like this long time. Really good video though it's too brief. Like I didn't understand what lag is and why we use it here. Also, that trend and drift thing and how you generate the difference of the log variable.. I suggest you to please explain a little more in-depth as it really helps us beginners a lot. This channel is however the best one I found so far and I will definitely recommend it to others.
Hi Dipen, I'm humbled by your comments, thanks a lot. Please browse through my Channel of 102 videos and you'll see videos on "Optimal lag selection". By principle, I create brief and "straight-to-the-point" videos in series/parts....so as not to bore my viewers by limit most of my clips to below 15mins....keep watching and please share with your students and academic community...gracias!
Thank you dear Dr for the wonderful presentation. I have a question regarding the preliminary tests, when performing stationarity test, cointegration or cross-sectional dependence. From the empirical point of view, which is better to perform them before taking the log for the variables or after and why? Regards
Hi Mohd, I use the log of the variables for these...on the "why"', I wil advise you scour the literature on any econometrics resource on the relevance of using the natural logarithm of a variable. Thanks.
But I have problem with my data, the Spurious reg. with 4 variables under study Indicate that R-squre is 0.5804 and durbin-watson is 1.7294. Which means that the variables are stationary but I tried to verify by conducting ADF and After conducting the ADF test, one of the variables are not stationary. What should I do?
@@CrunchEconometrix oops! But that variable (OPEC annual oil Price over 30 years) is really my interest. Or can I substitute it with monthly or daily price?
As usual one if your greatest videos. I have a question plz which i could not understand, why test dfuller with lag 1? Does the sample size matter here? Thanx
Thanks for your great video. I am writing from Germany. Pls kindly advise me a question as below: My model has 8 variables (1 dependent and 7 independent) with annual data in the period of 31 years. When I draw the graph of variables, some of them show a clear trend, so I use command: dfuller [varname], trend lags(1) => okay. But I have 2 cases that I don't know which command I should apply: (1) some variables don't show a clear trend, at first they declined within the first 20 years, then they stayed quite stable in the last 10 years. (2) some variables don't show any trend, they fluctuated over time (stochastic). Which command should I apply for 2 cases above pls? dfuller [varname], lags(1) Or dfuller [varname], drift lags(1) Or dfuller [varname], nonconstant lags(1) (I mean the option: suppress constant term in regression) Hope to hear from you soon.
Discovered you yesterday. You are great. Nice to have a brilliant (Naija) sister doing us proud. I will tell my Econometrics class mate/colleagues about this site. Helped me out alot. Have you thought about doing such videos using open source programs like R and Python?
Thanks Ricky, at the moment I niche on the mostly used software. I will appreciate the publicity as many Nigerians are unaware of my RUclips Channel. My purpose is to let students and researchers know that econometrics is not as difficult as it seems. May I know from where (location) you are reaching me?
@@CrunchEconometrix I live in Abuja. I am more into Python and some R. Presently the Econometrics class is biased towards R. However your videos still opened my eyes to how to interprete the ACF, PACF, Augmented Dickey Fuller and so on. A real handy channel. Next week we Are moving into GARCH and the rest.
Thank you for this informative and educative video. In my data analysis, I have three variables; two are stationary after first differencing but one variable is stationary at level but not at first difference. What should I do ?
Hi! Many thanks for the video. It was immensely informative. Please may I ask you a question? I have an unbalanced panel dataset. Do i need to convert it to a balanced dataset before running panel regressions? Thank you so much and any advice you offer is highly appreciated.
Thank you very much for the amazing video, it is helping me a lot for my master thesis. A question I cannot aswer: what is the difference between L1 and LD''?Thanks again
Thank you so so much! I'll definitely include you in my prayers! I just had one question, when i tested for the first time, it was non stationary, but when i tested with the 1st differences of the variables, it showed me that it was stationary. So do I conclude that the variables are indeed stationary?
Hi Ali, please know that all I(1) series are NONSTATIONARY. Reason: you had to difference it before it became a stationary series. You can also say it's nonstationary but stationary after 1st difference.
Hi Doc, Thanks loads for your indepth but simple to understand tutorials. Doc, can you kindly help. So I am using to 3 variables for an analysis. For 2 of the variables, the raw time series data are stationary without differencing. But the 3rd, variable is stationary at first difference. Can go ahead and run the model on the raw values of 2 the two variable together with the first difference of the 3rd variable??
Hi Elorm, on what to do, kindly watch my video on "This is how to specify ARDL models". Afterwards, watch my videos on the Bounds Test followed by other ARDL/ECM videos.
Dear Ngozi, thank you for the great video. It's very helpful. I have question about how to transform the data into difference form? the difference is transformed from the original data or from the log data? Thank you!!
I see. I tried to run my data again. Now the t-statistic is greater than the 5% critical value. But the constant value still not significant. How to deal with this? 🙏🙏
Constant is the intercept of the model. My advise is that you support video tutorials with reading econometrics textbooks with Sections 3 and 4 of articles that used the technique. It is the best way to understand and interpret results.
Hi, thank you for this video and very easy to comprehend. Please, I have a quick question to ask you. If I do the stationarity test and my variable becomes stationary at the first difference, am I going use the data in the first difference to run my regressions or I can still use data at the levels to run the regressions
Thanks Bolarinwa for the kind comments. For your query, I'll refer you to my videos on VAR and ARDL Models. If you are using the VAR and ARDL algorithms, you enter the variables in their level forms but if you're using the OLS algorithm, then you use the 1st difference of the series.
I am writting from the UK, I wanna ask you something, how do you know how many lags you have to include? Do you test that through varsoc? One of my X in my regression is stationary at lag 1,2 but if I see through varsoc the optimal lag is 3, so when I am testing with 3 lags and no trend/drift it seems to be no stationary. but when I include a trend/drift with lag(3) it seems that it is stationary. From the graph I think the variable is stationary because I think it looks good. What it your opinion I hope I was clear, I wish I could send you the pictures of my panel from stata but I cannot here. Thank you for your time, I really appreciate your content.
Thanks madam for such a pretty presentation. Please, assuming this ADF is already assured as you end up. Then my question. If i needed to perform estimates with the series what data are going to be employed in the model? Are the one differenced for ADF test or the original data. To put it abit clear: estimate people's consumptions using cpi as independent as your data dictates. So, you hav successifuly checked for stationarity, then which data set will be used in your regression model? Please?
Hi Pete, there's still a lot of contention as to whether you use the raw or differenced variables for the regression. But I've always maintained that you used the raw variables if you are using the VAR or ARDL algorithm to estimate the models and you use the differenced series if you're using the OLS algorithm. Others may contest this, but this is my approach. You can always check Gujarati and Wooldridge or any econometrics text to see how time series models are expressed.
Thank you! amazing videos with a clear explanation! :D One question tho. I appreciate you could help me, please. D. Watson test confirms it's a stationary model but ADF shows the different answers. So after the first differential, some variables are stationary and others keep being non-stationary. Should I differentiate them once more? And then, should I run all the model with these new second differentiated variables?
Hi Lilliana, thanks for the positive feedback. Deeply appreciated! DW is not used to determine stationarity. Kindly watch the clip again and follow my interpretation. Please may I know from where (location) you are reaching me?
Hello Doc. Is it possible to ignore the trend and(or) the drift when running the ADF test.... So that the Command will just be like this =》 dfuller Var1, lags(1).
Dear Dr. Ngozi. I am using Feasible Generalized Least Square (FGLS) for my panel data. Do I need to do ADF test? because Panel data consist of time-series data
I am not so sure if DW test of serial correlation has anything to do with R-square and let alone Unit root and DF distribution, please may I ask where can I find the existing source of such information?
thanks for the video. A quick question - is performing a Dickey-Fuller test enough to confirm a variable is stationary? can we assume a variable is stationary in mean, variance, and covariance if it doesn't have a unit root?
Hi Dominic, convert to numeric variables stored as "byte, int, long, float or double". I advise you check out other online resources on how to do this. Thanks.
Hi Cruch Econometix, I have a question: The final regression table shows the constant coefficients. May I know do I need to input this constant coefficients later on at the ARIMA estimation stage (the other tutorial video), where there is a table to input p,d,q and there is also a cell to input constant. May I know is the constant coefficients need to be keyed into the cell? Thank you!
Hello CrunchEconometrix! The video is quite helpful, thank you for that! I have few questions: 1. When I check for trend, my graph is quite chaotic but the values don't intercept. is that a problem? 2. When I try to perform the DW test, an error with this warning occurs "sample may not include multiple panels". In the error code stata provide it says i'm trying to test something that doesn't make sense, but doesn't tell me what exactly. What exactly its trying to tell me and hw should I fix it? Thank you in advance! :)
Hello Gyuld, I have read your query over and over but still have no idea as to what the issue is. This video comment is for the augmented Dickey-Fuller (ADF) test, so can you be more explicit?
Thank you for posting this video! I have two questions: Do I have to test if all my variables (3 independent and 1 dependent variables) with the ADF test? If some of my variables are stationary and some are non-stationary, can I still use time-series analysis or do they all have to be stationary?
Yes Taekyun, perform ADF test on each variable and YES only stationary variables are allowed in TS analysis....may I know from where you are reaching (location) me?
Hi, thank you for your videos; I have few questions: - Do i need to identify the lag lenght (using VAR diagnostics and test->lag order selection (pre-estim.)) before the ADF test? - what really is the difference btw drift and trend? when i consider drift in my adf test (unit root in exch. rate) I find that my crit. value is less than 1% c.v. (1.682 < 2.343) but > than 5% (1.682 > 1.652), that's not happens without drift, what should be my policy? - What do L1. and LD. stand for and how do i consider them? - In the video I don't really catch why I need to impose lagged differences at 1 instead 0 (otherwise, it shouldn't be an ADF test?) Thank you again!
Alessandro Cremaschini HI Alessi, to your queries: (1) there's no particular order but it's advisable to test your stationarity b4 deciding on the optimal lags just to be sure of the certainty of using the variables; (2) drift is the same as the constant, it is the stochastic trend. So use it in performing the test. Unless you have a very good reason of including the trend (called the deterministic trend); (3) L1 means first lag and LD means lagged difference; (4) yes, if you indicate 0, then you're performing the DF test and not ADF. So always indicate lags depending on the structure of the data. Hope I didn't leave any unanswered. Thanks for watching my videos and the encouragement. Please keep sharing people need to know that it's not as difficult as it seems👍🏽😉
Hi! I just found this video, it's amazing and your way of explaining things is impeccable. Just one doubt though, at 9:30, you said the absolute value of test statistics is not greater than any of the available options of the absolute value of critical values, it didn't seem right to me. 1.436>1.292(10%critical value) So this should mean that lnpce is stationary at 10% critical value or is 90% stationary? Please let me know if I am thinking right. Thanks!
Thanks Shlagha, for the positive feedback. Deeply appreciated! I mentioned that the series is nonstationary at the 5% level. However, it is stationary at 10%. May I know from where (location) you are reaching me?s
@@ShlaghaRastogi311 Awesome! I'll appreciate it if you can share the link to my RUclips Channel with your friends and academic community in India 🇮🇳 for awareness...thanks 😊.
Hahahaha, thanks Gabe for giving me an excellent pass and for honoring me with your subscription. Grateful! 😀 May I know from where (location) you are reaching me?
@@CrunchEconometrix Wow, thank you!!! I'm from Brazil. Thanks for sharing your knowledge. I'm gonna indicate your channel to anyone who needs to implement an econometric model. Greetings! Obrigado
The series at first difference level becomes stationary only when we included the drift term. It is still non stationary when the constant term is suppressed or the trend is concluded. What does it say for the series over all - stationary or non stationary?
Hi, thanks for the very informative and easy to understand video. I am trying to conduct the ADF test on some time series data. I have two questions. 1. The literature emphasises the importance of choosing the optimal lag length. I have used varsoc to help me for each variable, how do I then add this to the test? 2. Say my data has a clear time trend at level after looking at the plot. If it is non stationary at level, and then I use instead the first difference, do I still select the trend when doing the ADF test for the first difference. Kind Regards, Spencer
Hi Spencer, thanks for the kind words and I am indeed glad that I have the opportunity to teach the little I know using this medium. For the 1st question, use this code: dfuller d.y, trend lags(1)....for 1st difference stationarity test. For the 2nd question, yes you can include the trend option if graphical plot of the series shows that it has a trend. Use this command: dfuller d.y, trend lags(1)....for 1st difference stationarity test. Remember, the choice of lags is from the outcome of "varsoc". Initially run the test using 1 lag, then 2. But not more than 2 so as not to bias your test. Result is acceptable If the series is stationary at 1st difference with 1 lag even if the optimal level is 2 or more. Hope this explanation helps.
CrunchEconometrix Thank you for the reply, I am trying that out now! Just for clarification. First with just the level data. Say that the recommended lag is 2. I conduct the ADF test with following code dfuller var1, trend lags(2) ?
And if I cannot reject the null of a unit root, I then conduct the ADF test on the first difference. So I find using varsoc for that first difference variable the lag is now 1. Do I use code Dfuller dvar1, trend lag(1) ?
Hello I have been watching your video series on GMM and its been a great help ..........well could you guide if unit test is a requirement of the variables when i intend to use GMM model and in case required, which test i should go for? .......thanks
Hi Tashila, thanks for the encouraging feedback. Deeply appreciated! Kindly watch my 9 GMM series for more information on the procedure. I covered most of the basics. Please may I know from where (location) you are reaching me?
Hi, kindly may I ask you for the academic references for your rule of thumb that since the absolute value of the Test Statistic (1.726) is lower than the absolute value of the 3 options (4.071, 3.464, 3.158) ...we can not reject the null-hypothesis (|phi| =1) or as you say . Nice vid, suscribed time ago. Greetings.
Hi Aless, you can always refer to any econometrics textbook on the rudiments of stationary where you read about whether a series is stationary or not....in addition to several journal papers for the interpretation of stationarity. Hope these suggestions are helpful. Thanks!
Hello Ngozi. Thank you so much for your very useful and helpful video. After trying this i have questions. My R square is less than Durbin-Watson d-statistic showing that i have stationary series. However i try to confirm this using ADF and find that the absolute value of tstat is less than critical values thereby meaning data is non stationery for my dependent and 3 out of 5 independt variables. when i perform ADF using first difference the absolute value for the dependent is larger than all critical values, for one of my independents, its only larger at 5 and 10% and for another independent variable, the absolute value is not larger than any critical values. for 3 of my control variables the absolute values of t sata is larger than all critical values. Also, when i first difference one of my dependent variable i get negative values. Can you tell me how to proceed on 1. the differences in ADF tests and how to handle them. I plan to use VAR, VECM or ARDL 2. i worry that the negative values in differenced variables will affect analysis
@@CrunchEconometrix Hi Ngozi. I have watched the ARDL and VAR videos and re run my ADF. i find that for GDP it is till non stationary after first difference while the remaining variables become stationary after first difference. from your videos, I can only run VAR if variables are integrated at order 1 and for ARDL i cannot run it if variables are integrated at order 2. in this case how do i proceed with the gdp variable and what model do i use. please help
Hi Felix, thanks for watching my videos. The ADF test can be used on as many variables as possible because you are basically testing each variable for the presence of a unit root. The Durbin-Watson test is a post-estimation test obtained after you have carried out a regression analysis (whether univariate, bi-variate or multi-variate). If my videos have been helpful, I'll appreciate if you tell others about my Channel...thanks!
Thanks so much for your videos, they are so helpful. I have a question - Does the rule of when the R^2 is greater than the Durbin Watson d-statistic the variables are non-stationary still hold when using multiple variables. For example, if I were to perform a regression with 6 variables. Would it be okay to draw the same conclusion? If not, how would I prove the variables are stationary when there are multiple variables.
Hi Victoria, thanks for your encouraging feedback. Deeply appreciated! Yes, same rule applies. Please may I know from where (location) you are reaching me?
Hi Burak, thanks for the positive feedback on my video. The coefficients for trend and intercept are indicated in the ADF Table as shown and explained in the video. Please may I know from where (location) you are reaching me?
Greetings Dr. Ngozi First and foremost thank you for the video,its very informative. Am humbly requesting for a clarification on running a spurious regression. What happens in the event that my R square is less than Durbin watson statistic? Will i still have to do a stationarity test if the R square is less than the DW statistic? Am using stata 15.1 Thank you in advance
Hi Kevin, thanks for reaching out and for the encouraging feedback. Deeply appreciated! For the most part, stationarity test is mandatory in time series analysis. There is no getting around that until developments in econometric analysis say otherwise.
@@CrunchEconometrix So professor when we first make the regression and then the estat dwatson, if the result is R^2< DW then can we conclude that all the variables are stationary? Or we must test each variable individually? From what I understood from your video, when R^2> DW then we must check for sure, but what happens at R^2
Thank you for your explanation. It was easy to undestarnd. Just one question: What could I do if my data had a clear seasonal pattern or/and strutural breaks?
Thanks Carlos for the compliments. For structural breaks, the Zivot and Andrews (1992) test for unit root will be the most applicable as their test is an improvement of Perron (1989)’s. They assume that the exact time of the break-point is unknown so their method endogenously identifies the breakpoint. Likewise, the Gregory and Hansen (1996) test is designed for cointegration testing when controlling for structural breaks. Do you use Stata?
Hello, In my case, all variables are stationary at level when I select drift term, but they are not stationary when I select trend term Also D.Watson is smaller then r square. Now, can I proceed with this data?
Hi thank you for this fantastic video i am very happy for this tutorial but i have a question about how to create differences in order to perform the unit roots test when it is not stationary at level, thank very much
Hi Valens, thanks for the positive feedback. Deeply appreciated. Watch the video to the end. I did just that. Please may I know from where (location) you are reaching me?
Hello, In case I use ADF including 1 or 2 lags, none of my first difference variables become stationary. Is it ok if I use classic DF test and phillips-Perron stationarity test? In case I do not include lags in stationarity test, shall I use lags when checking for coiintegration and testing VECM ? Thanks for the answer.
Hi Vane, if you use lags then you are performing the ADF. Without lags it is just the DF test. It is also ok to use the PP test. Yes, include lags when testing for cointegration and reduce the lags by 1 for the VECM (watch my videos on that).
Dear Ngozi, Thanks very much for the guidance. I followed your method and found: For level form (1 lag), the T-Statistic values of some variables are lesser than 1% Critical Value. However, below the table, P value is greater than 5%. Such conflicting results show that we can't conclude that this variable is non-stationary. Am I right?
Cherry, you are mixing things up a bit. Please go over the video tutorial again supported with readings from (1) Gujarati and Porter and (2) Wooldridge. That will clear up how to read the Table.
Respected professor, Thanks for your nice tutorial. Please kindly can you inform me that to choose the optimal lag length in the ADF test whether I need to use ''VARSOC'' command or I can choose the lag randomly. (From Xian Jiaotong University, China)
Kindly watch my video on "Optimal Lag Selection" and that of the applicable technique you want to use because I often show how to obtain the optimal lag. Thanks.
@@CrunchEconometrix I think I have a similar question. You found the optimal lag for the variable but you did not use it for the test of stationarity. You only said that if you leave it at level zero, it will have no difference with the normal Dickey-Fuller test. Was there any need for the optimal lag length selection?
Hi Catherine, using the "optimal lag" gives the model a perfect fit. However, there are situations where such models fail diagnostics. In that case, you can adjust the respective lags.
Hi.. thank you for this video. I am having some issues in identifying the model and lags should use to do the ADF tests. I am doing a test for 8 variables for a period of 30 years. I am failing to understand which model to use whether intercept trend or no intercept. Besides i used lag of 1 for all. For example for GDP growth annual in its level form , both trend , intercept only says its stationary. But supress constant says its not. So what should I take as good??
Hi Ramsamy, with annual data it is advisable to use 2 lags and a model with "intercept". You can include trend if the graphical plot indicates a trending series or if the trend coefficient is significant. I explained these in detail in my videos on "optimal lag selection" and this particularly video. Hope these tips are helpful.
Thank you for your prompt reply! So basically you are saying i can use both lag of 1 and 2 irrespective of the variables am doing is that right?? In my Gdp growth annual graph The graph is not around O and its showing some sort of downward trend. Going by the graph only i would have said its not stationary. However doing the test with trend the test statistics is greater than Critical value meaning it is stationary. And if i still do the differencing the graph now becomes around O and stationary. So, which one should I be going for: the level or first differencing?
Hi thank you for your amazing videos. They're helping me with my dissertation that I have to submit in a week. Can you please help me with one thing. When I perform the ADF test my constant always appears to be insignificant i.e. the p value is greater than 0.05 for the constant. What does that mean? can I not use my series for forecasting?
Aleena, thanks for the positive feedback. Deeply appreciated! Keep it simple and use the ADF statistic/pvalue to determine the stationarity of the variables.
Hi, dear Ngozi, what's the meaning of "include drift term in regression?" does it mean that we are now using the first-differenced form to see whether the variable is stationary or not?
Hi I have a panel unbalanced data, i used fisher type test only on dependent variables, like female labour force. I can see I have unit root. what am I supposed to do to remove it. Also do I need to check all my independent control variables as well ??
Hi, I have a question on how to determine the lag numbers in ADF test. May I confirm to determine the lag number, I perform the Varcos [var] in STATA, correct? I performed varcos [var] in STATA, and chose AIC as a criteria. My data is monthly data, and I got an optimal lag number 4. Is this reasonable, as you have mentioned that for monthly data, the lag number is usually 6, 12 etc. This lag number is different from the ARIMA (p,d,q) am I correct? Thank you so much!
Thank you very much for this video. I do have a question though. Is it a problem if my variable is still nonstationary when including the trend but stationary when including the drift?(i am using the first difference so, I'm not doing that part wrong) Edit: my variable is ln of real GDP in quarterly data
Hi Silva, thanks for the positive feedback on my video. Deeply appreciated! This is not a problem. Simply indicate on the Table of results that the series is stationarity with a drift then proceed to estimation. Please may I know from where (location) you are reaching me?
CrunchEconometrix hello again! I am writing to you from Lisbon but am currently studying at the university of Nottingham! Let me take this chance to once again thank you for the clarity that you provide with your step by step analysis in your videos. It really is a game changer and has already answered a few of my many questions related to my thesis 😅. Thank you and God bless you
Hi, Why did you use 'trend' in the first differenced series? since the plot of the first difference series does not show any trend, isn't it wrong to use 'trend' option in the ADF test for that first differenced series?
Hello, thank you for such an insightful video. I run my data just as you have shown in the video but after using the data at first difference to run the augmented dickey fuller test my test statistic value is still lower than the 5% critical value. I know this means that my data is still non stationary at first difference but i don't know what to do now to get it to be stationary. Can you help please? I'm stuck with my research now.
Hi Judy, don't panic. You can either change that variable to a closer proxy or estimate the Toda-Yamamoto technique. I've not had cause to use it but you can read up on it. You'll be fine. May I know from where (location) you are reaching me?
Yes please, I'll read about it and see how I can use it in my work. Thank you very much for your reply. I'm sending this email from China. I'm currently pursuing my Master's degree..
@@judylawson2588 Alright, Judy. Wish you the best. I'll appreciate it if you can share the link to my RUclips Channel with your friends and academic community in China for awareness. Thanks 😊
Can you please tell me what kind of variable is qtrly? It's howing me as a string(as it contains letters and numbers) hence not able to use it along with tsset command. Would appreciate if you could tell me where I am going wrong. Thank you.
Dw= 1.976556 and r^2=0.4179 but 3 out of 4 independent variables are not stationary according to augmented Dickey fuller test. What is the problem here?
Hello sir, does lags(1) correspond to "betaLNPCE t-1" in the regression model? varsoc gave me the optimal lag length of 4, should i put in "dfuller, X1 lags(4)" ? Also, would my model look like this: "betaX1 t-4" if i want to do ARDL modeling? (assuming X1 is stationary at level with lags(4)). Thank you!
RUclips 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.
Thank you
Hello Ma how can I get to have a chat with you. Really struggling with my data
Hi Dipen, kindly search online for resources on this as I'm yet to have videos on this procedure. Thanks.
This video gives the basis on my master dissertation! For real is perfect, it have clear content and illustration. THANKS A LOT
Thanks for the encouraging feedback, Elide. Deeply appreciated! Please may I know from where (location) you are reaching me?
I don't know how much I should thank to you for your effort and clear explanation!! I'm whipping myself to finish my thesis during self-quarantine and thanks to you I can proceed the analysis that I have been stuck. Your videos are not only helpful in terms of using stata technically but also understanding fundamental theoretically!
Thanks for the positive feedback and kind remarks. Deeply appreciated! Please keep safe! ❤️ May I know from where (location) you are reaching me?
@@CrunchEconometrix Thanks for your comment back! I'm from South Korea, doing my degree in Turkey now :)
This video may have just been the key to my bachelor thesis! Thanks a lot for the awesome content and the clear explanation! (coming from London) ;) will share in my university cause I know a lot of people struggling with this sensitive subject
Thanks for the positive feedback, Maelian...deeply appreciated! Happy to hear that you are willing to share the clip with your colleagues...gracias!!! :)
This video is a life-saver! When i run the ADF test, I dont know why the absolute t-statistic value is always smaller than the absolute critical value(no matter 1% 5% 10%), which is disappointing since i wanna reject Ho. Your video explained this quite clear and i got the ideal output(reject Ho and the variable is stationary) after using difference of log form, according to this video. Thank you !
U're welcome, WZ😊
Hi Ngozi, I am refreshing my time series econometrics of non-stationary series and I can't stop watching your videos :-) Keep up the great work! Greetings from Switzerland - Claudio
Hi Claudio, thanks for the encouraging feedback. Deeply appreciated! 🙏 ❤️
You just have a way of teaching and ensuring we get
Dear Ngozi, amazing for the explanation of the ADF Test through the youtube and very useful for my article. Thank you so much
You're most welcome Fernando!
your accent is very soothing to the ears
Hahahaha, Rao. Thanks for the warm feedback 😄
Thank you for your perfect contribution. However, please clarify a little bit, why we use lag one instead of zero while performing ADF
Hi Kubilay, 0 lag = DF; 1 lag = ADF.
your explanation is perfect!!!!! love it so much
Hi, I want to as you a question, why in 5:36, you only do the regression towards two variable lnpce and lnpdi, is there any reason why you don't include gdp on the regression?
Compliment is humbly taken, Fry! Thanks. May I know from where (location) you are reaching me?
Omission is deliberate. I could have included GDP if I wanted to. thanks for the observation, though.
Very interesting ma. You made me understand so very well 🎉
Glad to hear that, deeply appreciated! 🥰
Searching for a channel like this long time. Really good video though it's too brief. Like I didn't understand what lag is and why we use it here. Also, that trend and drift thing and how you generate the difference of the log variable.. I suggest you to please explain a little more in-depth as it really helps us beginners a lot. This channel is however the best one I found so far and I will definitely recommend it to others.
Hi Dipen, I'm humbled by your comments, thanks a lot. Please browse through my Channel of 102 videos and you'll see videos on "Optimal lag selection". By principle, I create brief and "straight-to-the-point" videos in series/parts....so as not to bore my viewers by limit most of my clips to below 15mins....keep watching and please share with your students and academic community...gracias!
CrunchEconometrix yes... Definitely! And thank you so much. I'll watch all your videos in these few days.
love your detailed presentation,,,,go ahead
Thanks Sabiha for the positive feedback!!! May I know from where (location) you are reaching me?
Absolutely fantastic presentation of the content and explanation. Thank you so much
Thanks Jean-Philippe Chen for the positve feedback. Deeply appreciated! May I know from where (location) you are reaching me?
Thank you dear Dr for the wonderful presentation. I have a question regarding the preliminary tests, when performing stationarity test, cointegration or cross-sectional dependence. From the empirical point of view, which is better to perform them before taking the log for the variables or after and why?
Regards
Hi Mohd, I use the log of the variables for these...on the "why"', I wil advise you scour the literature on any econometrics resource on the relevance of using the natural logarithm of a variable. Thanks.
Thankyou so much for the wonderful videos
U're welcome, Ekta. Please may I know from where (location) you are reaching me?
Wow! I find this useful without ambiguity. Kudos to you.
Thanks Sikiru, for the positive feedback...deeply appreciated! May I know from where you are reaching me?
But I have problem with my data, the Spurious reg. with 4 variables under study Indicate that R-squre is 0.5804 and durbin-watson is 1.7294. Which means that the variables are stationary but I tried to verify by conducting ADF and After conducting the ADF test, one of the variables are not stationary. What should I do?
Yea, I am a Nigerian, currently Studying MBA in Beijing Normal University, China.
@@aliuomotayosikiru6154 Drop it and use another close proxy.
@@CrunchEconometrix oops! But that variable (OPEC annual oil Price over 30 years) is really my interest.
Or can I substitute it with monthly or daily price?
Many Thanks (From RWANDA - Kigali)
You are welcome, Sir 🙏. Much love from Nigeria ❤️.
As usual one if your greatest videos. I have a question plz which i could not understand, why test dfuller with lag 1? Does the sample size matter here? Thanx
Using lag makes it the ADF test. Without lag, it's the DF test.
Thanks for your great video. I am writing from Germany. Pls kindly advise me a question as below: My model has 8 variables (1 dependent and 7 independent) with annual data in the period of 31 years.
When I draw the graph of variables, some of them show a clear trend, so I use command: dfuller [varname], trend lags(1) => okay. But I have 2 cases that I don't know which command I should apply: (1) some variables don't show a clear trend, at first they declined within the first 20 years, then they stayed quite stable in the last 10 years. (2) some variables don't show any trend, they fluctuated over time (stochastic).
Which command should I apply for 2 cases above pls?
dfuller [varname], lags(1)
Or
dfuller [varname], drift lags(1)
Or
dfuller [varname], nonconstant lags(1)
(I mean the option: suppress constant term in regression)
Hope to hear from you soon.
Hi Thurong, I always start with option 1.
@@CrunchEconometrix hi, do you mean that dfuller [varname], lags(1). Right?
@@thuongvu2503 Yes
Great video! But how to know whether to first-difference or de-trend? That should also be covered by your excellent videos
Thanks for the positive feedback, Achudakhim. You may check other online resources for "de-trend" videos. Thanks.
Discovered you yesterday. You are great. Nice to have a brilliant (Naija) sister doing us proud. I will tell my Econometrics class mate/colleagues about this site. Helped me out alot.
Have you thought about doing such videos using open source programs like R and Python?
Thanks Ricky, at the moment I niche on the mostly used software. I will appreciate the publicity as many Nigerians are unaware of my RUclips Channel. My purpose is to let students and researchers know that econometrics is not as difficult as it seems. May I know from where (location) you are reaching me?
@@CrunchEconometrix I live in Abuja. I am more into Python and some R. Presently the Econometrics class is biased towards R. However your videos still opened my eyes to how to interprete the ACF, PACF, Augmented Dickey Fuller and so on. A real handy channel. Next week we Are moving into GARCH and the rest.
Thank you for this informative and educative video. In my data analysis, I have three variables; two are stationary after first differencing but one variable is stationary at level but not at first difference. What should I do ?
Hi, Makungu, watch my ARDL videos. That's your next step.
Your videos are really helpful. Great work.
Thanks for the positive feedback, Pranay. Deeply appreciated!
Your video has helped me a lot with my thesis, I really appreciate your video
Hi Nhi, if you untransformed the variable then it means the series is stationary at level with a trend. That is, I(0).
@@CrunchEconometrix Thank u so much
Hi! Many thanks for the video. It was immensely informative. Please may I ask you a question? I have an unbalanced panel dataset. Do i need to convert it to a balanced dataset before running panel regressions? Thank you so much and any advice you offer is highly appreciated.
Hi Debosmita, you posted your query on a WRONG video. This is NOT a panel data video. Kindly re-post correctly. Thanks
@@CrunchEconometrix so sorry..
Thank you very much for the amazing video, it is helping me a lot for my master thesis. A question I cannot aswer: what is the difference between L1 and LD''?Thanks again
Thanks, Camillo for your encouraging feedback. Deeply appreciated.
L1: first lag
LD: lag difference
Soo helpful... am almost sybmiting my asignmnt, this z soo helpful thank yu
U're welcome, Grace! I'm glad you find this video very helpful. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix Im in Zimbabwe
thank you sooo much for this!! you are a lifesaver! btw do we need to take first difference in kpss test?
Thanks, Öykü for the encouraging feedback. Yeah, any stationary test will suffice but make sure you understand what its null hypothesis is.
Thank you so so much! I'll definitely include you in my prayers! I just had one question, when i tested for the first time, it was non stationary, but when i tested with the 1st differences of the variables, it showed me that it was stationary. So do I conclude that the variables are indeed stationary?
Hi Ali, please know that all I(1) series are NONSTATIONARY. Reason: you had to difference it before it became a stationary series. You can also say it's nonstationary but stationary after 1st difference.
Thank you so much!! It was clear and easy to understand!! 🎉💐
Glad it was helpful, Joe!
😍😍
Thank you with your teachings..... can you assist with How to test for cross sectional dependence in stata
Collin, the video will be uploaded to my Teachable platform in due course. Here is the link cruncheconometrix.teachable.com
Thnks. Can you advice ...is it possible to report ...t test statistics and wald tests together undr one model .... can you help
Yes.
Hi Doc,
Thanks loads for your indepth but simple to understand tutorials.
Doc, can you kindly help. So I am using to 3 variables for an analysis. For 2 of the variables, the raw time series data are stationary without differencing. But the 3rd, variable is stationary at first difference. Can go ahead and run the model on the raw values of 2 the two variable together with the first difference of the 3rd variable??
Hi Elorm, on what to do, kindly watch my video on "This is how to specify ARDL models". Afterwards, watch my videos on the Bounds Test followed by other ARDL/ECM videos.
You teach so well! ❤️
Thanks, Chelsea for the encouraging feedback... deeply appreciated!
Dear Ngozi, thank you for the great video. It's very helpful. I have question about how to transform the data into difference form? the difference is transformed from the original data or from the log data? Thank you!!
Shinta, that will depend on the form of the variable. Difference can be obtained both ways.
I see. I tried to run my data again. Now the t-statistic is greater than the 5% critical value. But the constant value still not significant. How to deal with this? 🙏🙏
Constant is the intercept of the model. My advise is that you support video tutorials with reading econometrics textbooks with Sections 3 and 4 of articles that used the technique. It is the best way to understand and interpret results.
Hi, thank you for this video and very easy to comprehend. Please, I have a quick question to ask you. If I do the stationarity test and my variable becomes stationary at the first difference, am I going use the data in the first difference to run my regressions or I can still use data at the levels to run the regressions
Thanks Bolarinwa for the kind comments. For your query, I'll refer you to my videos on VAR and ARDL Models. If you are using the VAR and ARDL algorithms, you enter the variables in their level forms but if you're using the OLS algorithm, then you use the 1st difference of the series.
Thank you very much for this lesson. In eviews, it looks a bit complicated going through many levels.
The Art of The Deal Not at all, watch the EViews clip on that and you'll realize that it's easier than you think😊
Absolutely. I will encourage others to subscribe.
I am writting from the UK, I wanna ask you something, how do you know how many lags you have to include? Do you test that through varsoc? One of my X in my regression is stationary at lag 1,2 but if I see through varsoc the optimal lag is 3, so when I am testing with 3 lags and no trend/drift it seems to be no stationary. but when I include a trend/drift with lag(3) it seems that it is stationary. From the graph I think the variable is stationary because I think it looks good. What it your opinion I hope I was clear, I wish I could send you the pictures of my panel from stata but I cannot here. Thank you for your time, I really appreciate your content.
Hi Apla, varsoc gives you optimal lags. No need to send the graph. Brief explanation will suffice. What technique are you engaging?
Thanks madam for such a pretty presentation. Please, assuming this ADF is already assured as you end up. Then my question. If i needed to perform estimates with the series what data are going to be employed in the model? Are the one differenced for ADF test or the original data. To put it abit clear: estimate people's consumptions using cpi as independent as your data dictates. So, you hav successifuly checked for stationarity, then which data set will be used in your regression model? Please?
Hi Pete, there's still a lot of contention as to whether you use the raw or differenced variables for the regression. But I've always maintained that you used the raw variables if you are using the VAR or ARDL algorithm to estimate the models and you use the differenced series if you're using the OLS algorithm. Others may contest this, but this is my approach. You can always check Gujarati and Wooldridge or any econometrics text to see how time series models are expressed.
CrunchEconometrix kindly, Thanks inadvance I wan check also in the proposed books on time series hints.
Thank you! amazing videos with a clear explanation! :D
One question tho. I appreciate you could help me, please. D. Watson test confirms it's a stationary model but ADF shows the different answers. So after the first differential, some variables are stationary and others keep being non-stationary. Should I differentiate them once more? And then, should I run all the model with these new second differentiated variables?
Hi Lilliana, thanks for the positive feedback. Deeply appreciated! DW is not used to determine stationarity. Kindly watch the clip again and follow my interpretation. Please may I know from where (location) you are reaching me?
Thanks so much for this. Really helpful
You're very welcome, Laura!
Hello Doc.
Is it possible to ignore the trend and(or) the drift when running the ADF test.... So that the Command will just be like this =》 dfuller Var1, lags(1).
Yes. I only add trend if the series is nonstationary.
Thank you.
Hi ma'am, do I perform the optimal lag selection using AIC and BIC before the ADF test or after? Thank you in advance!
Better to select optimal lags before testing for a unit root.
Dear Dr. Ngozi. I am using Feasible Generalized Least Square (FGLS) for my panel data. Do I need to do ADF test? because Panel data consist of time-series data
Hi Shinta, yes.
I am not so sure if DW test of serial correlation has anything to do with R-square and let alone Unit root and DF distribution, please may I ask where can I find the existing source of such information?
Hi GG, I don't know what to make of your query but you search online for resources that addresses your quest. Thanks.
thanks for the video. A quick question - is performing a Dickey-Fuller test enough to confirm a variable is stationary? can we assume a variable is stationary in mean, variance, and covariance if it doesn't have a unit root?
You have just defined stationarity.
Great Video, however I am struggling with the time series part, as it says the dates I've used are 'string variables'.
Hi Dominic, convert to numeric variables stored as "byte, int, long, float or double". I advise you check out other online resources on how to do this. Thanks.
Hi Cruch Econometix,
I have a question: The final regression table shows the constant coefficients. May I know do I need to input this constant coefficients later on at the ARIMA estimation stage (the other tutorial video), where there is a table to input p,d,q and there is also a cell to input constant. May I know is the constant coefficients need to be keyed into the cell? Thank you!
He, your queries are confusing. Kindly post respective queries on the videos concerned not lumped together. This video is on ADF not ARIMA. Thanks.
Hello CrunchEconometrix!
The video is quite helpful, thank you for that!
I have few questions:
1. When I check for trend, my graph is quite chaotic but the values don't intercept. is that a problem?
2. When I try to perform the DW test, an error with this warning occurs "sample may not include multiple panels". In the error code stata provide it says i'm trying to test something that doesn't make sense, but doesn't tell me what exactly. What exactly its trying to tell me and hw should I fix it?
Thank you in advance! :)
Hello Gyuld, I have read your query over and over but still have no idea as to what the issue is. This video comment is for the augmented Dickey-Fuller (ADF) test, so can you be more explicit?
Thank you for posting this video! I have two questions: Do I have to test if all my variables (3 independent and 1 dependent variables) with the ADF test? If some of my variables are stationary and some are non-stationary, can I still use time-series analysis or do they all have to be stationary?
Yes Taekyun, perform ADF test on each variable and YES only stationary variables are allowed in TS analysis....may I know from where you are reaching (location) me?
Hi, thank you for your videos; I have few questions:
- Do i need to identify the lag lenght (using VAR diagnostics and test->lag order selection (pre-estim.)) before the ADF test?
- what really is the difference btw drift and trend? when i consider drift in my adf test (unit root in exch. rate) I find that my crit. value is less than 1% c.v. (1.682 < 2.343) but > than 5% (1.682 > 1.652), that's not happens without drift, what should be my policy?
- What do L1. and LD. stand for and how do i consider them?
- In the video I don't really catch why I need to impose lagged differences at 1 instead 0 (otherwise, it shouldn't be an ADF test?)
Thank you again!
Alessandro Cremaschini HI Alessi, to your queries: (1) there's no particular order but it's advisable to test your stationarity b4 deciding on the optimal lags just to be sure of the certainty of using the variables; (2) drift is the same as the constant, it is the stochastic trend. So use it in performing the test. Unless you have a very good reason of including the trend (called the deterministic trend); (3) L1 means first lag and LD means lagged difference; (4) yes, if you indicate 0, then you're performing the DF test and not ADF. So always indicate lags depending on the structure of the data. Hope I didn't leave any unanswered. Thanks for watching my videos and the encouragement. Please keep sharing people need to know that it's not as difficult as it seems👍🏽😉
so, since L1. = Y(t-1) and LD: = D.Y(t-1), my original model, with lagged difference = 1 in ADF test, would be Yt = m + Yt-1 + Yt-2 + Et ?
I'm not clear with your query.
I mean, having L1. = Yt-1 and LD. = DYt-1 and setting at 1 The lagged difference means that i'm considering an ar(2)?
Alessandro Cremaschini Are you testing for stationarity of the series?
Hi! I just found this video, it's amazing and your way of explaining things is impeccable. Just one doubt though, at 9:30, you said the absolute value of test statistics is not greater than any of the available options of the absolute value of critical values, it didn't seem right to me. 1.436>1.292(10%critical value) So this should mean that lnpce is stationary at 10% critical value or is 90% stationary? Please let me know if I am thinking right. Thanks!
Thanks Shlagha, for the positive feedback. Deeply appreciated! I mentioned that the series is nonstationary at the 5% level. However, it is stationary at 10%. May I know from where (location) you are reaching me?s
@@CrunchEconometrix Oh now I get it! Thanks! I am from India, by the way :)
@@ShlaghaRastogi311 Awesome! I'll appreciate it if you can share the link to my RUclips Channel with your friends and academic community in India 🇮🇳 for awareness...thanks 😊.
@@CrunchEconometrix Sure I will! :)
I just subscribed your channel... thank you for helping me out. Great job. By the way, I'm reading your website at this moment.
Hahahaha, thanks Gabe for giving me an excellent pass and for honoring me with your subscription. Grateful! 😀 May I know from where (location) you are reaching me?
@@CrunchEconometrix Wow, thank you!!! I'm from Brazil. Thanks for sharing your knowledge. I'm gonna indicate your channel to anyone who needs to implement an econometric model. Greetings! Obrigado
The series at first difference level becomes stationary only when we included the drift term. It is still non stationary when the constant term is suppressed or the trend is concluded. What does it say for the series over all - stationary or non stationary?
Neyalee, don't suppress the constant term. Series is stationary with a drift.
Hi, thanks for the very informative and easy to understand video. I am trying to conduct the ADF test on some time series data. I have two questions.
1. The literature emphasises the importance of choosing the optimal lag length. I have used varsoc to help me for each variable, how do I then add this to the test?
2. Say my data has a clear time trend at level after looking at the plot. If it is non stationary at level, and then I use instead the first difference, do I still select the trend when doing the ADF test for the first difference.
Kind Regards,
Spencer
Hi Spencer, thanks for the kind words and I am indeed glad that I have the opportunity to teach the little I know using this medium.
For the 1st question, use this code: dfuller d.y, trend lags(1)....for 1st difference stationarity test.
For the 2nd question, yes you can include the trend option if graphical plot of the series shows that it has a trend. Use this command: dfuller d.y, trend lags(1)....for 1st difference stationarity test.
Remember, the choice of lags is from the outcome of "varsoc". Initially run the test using 1 lag, then 2. But not more than 2 so as not to bias your test. Result is acceptable If the series is stationary at 1st difference with 1 lag even if the optimal level is 2 or more.
Hope this explanation helps.
CrunchEconometrix Thank you for the reply, I am trying that out now!
Just for clarification. First with just the level data. Say that the recommended lag is 2. I conduct the ADF test with following code
dfuller var1, trend lags(2) ?
And if I cannot reject the null of a unit root, I then conduct the ADF test on the first difference. So I find using varsoc for that first difference variable the lag is now 1. Do I use code
Dfuller dvar1, trend lag(1) ?
Correct...that's the syntax to use!
Hello
I have been watching your video series on GMM and its been a great help ..........well could you guide if unit test is a requirement of the variables when i intend to use GMM model and in case required, which test i should go for? .......thanks
Hi Tashila, thanks for the encouraging feedback. Deeply appreciated! Kindly watch my 9 GMM series for more information on the procedure. I covered most of the basics. Please may I know from where (location) you are reaching me?
Tashila Ethenpa I am studying in Tokyo,Japan but basically from India
Hi, kindly may I ask you for the academic references for your rule of thumb that since the absolute value of the Test Statistic (1.726) is lower than the absolute value of the 3 options (4.071, 3.464, 3.158) ...we can not reject the null-hypothesis (|phi| =1) or as you say . Nice vid, suscribed time ago. Greetings.
Hi Aless, you can always refer to any econometrics textbook on the rudiments of stationary where you read about whether a series is stationary or not....in addition to several journal papers for the interpretation of stationarity. Hope these suggestions are helpful. Thanks!
Hello Ngozi. Thank you so much for your very useful and helpful video.
After trying this i have questions.
My R square is less than Durbin-Watson d-statistic showing that i have stationary series.
However i try to confirm this using ADF and find that the absolute value of tstat is less than critical values thereby meaning data is non stationery for my dependent and 3 out of 5 independt variables.
when i perform ADF using first difference the absolute value for the dependent is larger than all critical values, for one of my independents, its only larger at 5 and 10% and for another independent variable, the absolute value is not larger than any critical values. for 3 of my control variables the absolute values of t sata is larger than all critical values. Also, when i first difference one of my dependent variable i get negative values.
Can you tell me how to proceed on 1. the differences in ADF tests and how to handle them. I plan to use VAR, VECM or ARDL 2. i worry that the negative values in differenced variables will affect analysis
Kindly follow my guide on ARDL and VAR techniques to understand which approach best applies to your study.
@@CrunchEconometrix Hi Ngozi. I have watched the ARDL and VAR videos and re run my ADF. i find that for GDP it is till non stationary after first difference while the remaining variables become stationary after first difference. from your videos, I can only run VAR if variables are integrated at order 1 and for ARDL i cannot run it if variables are integrated at order 2. in this case how do i proceed with the gdp variable and what model do i use. please help
Thanks for the video, what if we are using more than two variables? How are we performing the durbin Watson and the adf tests?
Hi Felix, thanks for watching my videos. The ADF test can be used on as many variables as possible because you are basically testing each variable for the presence of a unit root. The Durbin-Watson test is a post-estimation test obtained after you have carried out a regression analysis (whether univariate, bi-variate or multi-variate). If my videos have been helpful, I'll appreciate if you tell others about my Channel...thanks!
Thanks so much for your videos, they are so helpful.
I have a question - Does the rule of when the R^2 is greater than the Durbin Watson d-statistic the variables are non-stationary still hold when using multiple variables. For example, if I were to perform a regression with 6 variables. Would it be okay to draw the same conclusion? If not, how would I prove the variables are stationary when there are multiple variables.
Hi Victoria, thanks for your encouraging feedback. Deeply appreciated! Yes, same rule applies. Please may I know from where (location) you are reaching me?
CrunchEconometrix I’m from the UK
Thanks for the reply and once again your videos are amazing and very helpful, looking forward to more.
No worries, I'll keep doing my best...thanks!
when it was not stationary at first, you created the difference. how did you get the figures for the difference
gen dlnpce=D.lnpce
Thank you Dr. Will try this and revert. And, please, do you do this for all data that exhibit similar outcome ma.
Yes. If the series is nonstationary at level.
hello, how do you make the first difference?
D.variable
How do you fix the 'repeated time values' error? Can't find the solution.
Chimuka, that happens when you are using a panel data to estimate a time series analysis.
Thank you for your lessons, but i have a problem. How can i determine intercept and trend value ?
Hi Burak, thanks for the positive feedback on my video. The coefficients for trend and intercept are indicated in the ADF Table as shown and explained in the video. Please may I know from where (location) you are reaching me?
Greetings Dr. Ngozi
First and foremost thank you for the video,its very informative.
Am humbly requesting for a clarification on running a spurious regression. What happens in the event that my R square is less than Durbin watson statistic? Will i still have to do a stationarity test if the R square is less than the DW statistic? Am using stata 15.1
Thank you in advance
Hi Kevin, thanks for reaching out and for the encouraging feedback. Deeply appreciated! For the most part, stationarity test is mandatory in time series analysis. There is no getting around that until developments in econometric analysis say otherwise.
@@CrunchEconometrix So professor when we first make the regression and then the estat dwatson, if the result is R^2< DW then can we conclude that all the variables are stationary? Or we must test each variable individually? From what I understood from your video, when R^2> DW then we must check for sure, but what happens at R^2
Thank you for your explanation. It was easy to undestarnd. Just one question: What could I do if my data had a clear seasonal pattern or/and strutural breaks?
Thanks Carlos for the compliments. For structural breaks, the Zivot and Andrews (1992) test for unit root will be the most applicable as their test is an improvement of Perron (1989)’s. They assume that the exact time of the break-point is unknown so their method endogenously identifies the breakpoint. Likewise, the Gregory and Hansen (1996) test is designed for cointegration testing when controlling for structural breaks. Do you use Stata?
Thanks for your quick reply. Yes I do!
No worries...please tell others about my channel!...gracias!!!!
You should make a dummy variable
thanks a lot
Your video is a life saver!
Thanks for the encouraging feedback. Deeply appreciated!
Hello,
In my case, all variables are stationary at level when I select drift term, but they are not stationary when I select trend term
Also D.Watson is smaller then r square. Now, can I proceed with this data?
Hi Walid, you decide if you want to include a trend or not and then take your analysis from there. Regards.
Hi thank you for this fantastic video i am very happy for this tutorial but i have a question about how to create differences in order to perform the unit roots test when it is not stationary at level, thank very much
Hi Valens, thanks for the positive feedback. Deeply appreciated. Watch the video to the end. I did just that. Please may I know from where (location) you are reaching me?
Hi, thanks for this useful video. I have one question to ask, do we need to run var model even our data are stationary at level form?
Hi Toba, thanks for the positive feedback. Not VAR, just the OLS technique.
Hello,
In case I use ADF including 1 or 2 lags, none of my first difference variables become stationary. Is it ok if I use classic DF test and phillips-Perron stationarity test?
In case I do not include lags in stationarity test, shall I use lags when checking for coiintegration and testing VECM ?
Thanks for the answer.
Hi Vane, if you use lags then you are performing the ADF. Without lags it is just the DF test. It is also ok to use the PP test. Yes, include lags when testing for cointegration and reduce the lags by 1 for the VECM (watch my videos on that).
Thanks for the video. Very useful
U're welcome Raifatou......may I know from where you (location) are reaching me?
@@CrunchEconometrix I'm in China. I'm doing my master's degree in agricultural economics and management.
Dear Ngozi,
Thanks very much for the guidance.
I followed your method and found:
For level form (1 lag), the T-Statistic values of some variables are lesser than 1% Critical Value. However, below the table, P value is greater than 5%. Such conflicting results show that we can't conclude that this variable is non-stationary. Am I right?
Cherry, you are mixing things up a bit. Please go over the video tutorial again supported with readings from (1) Gujarati and Porter and (2) Wooldridge. That will clear up how to read the Table.
@@CrunchEconometrix Got it! Thank you.
Respected professor, Thanks for your nice tutorial. Please kindly can you inform me that to choose the optimal lag length in the ADF test whether I need to use ''VARSOC'' command or I can choose the lag randomly. (From Xian Jiaotong University, China)
Kindly watch my video on "Optimal Lag Selection" and that of the applicable technique you want to use because I often show how to obtain the optimal lag. Thanks.
@@CrunchEconometrix I think I have a similar question. You found the optimal lag for the variable but you did not use it for the test of stationarity. You only said that if you leave it at level zero, it will have no difference with the normal Dickey-Fuller test. Was there any need for the optimal lag length selection?
Hi Catherine, using the "optimal lag" gives the model a perfect fit. However, there are situations where such models fail diagnostics. In that case, you can adjust the respective lags.
hello i have a question. what should i do if the variables are still not stationary even after using the first difference. thanks in advance :)
Hi Micheal, I will advise you drop it and use a closer proxy.
Thanks ma'am
Most welcome, Immaculate!😊
Hi, i have a question: How to create a dialogue form of a variable?
Hi Nga, I honestly have no idea what a "dialogue form" is.
The video is applicable in case of time series data only or the same steps are applicable to panel data as well ? Plz clarify.
You can use it for panel analysis too.
Hi.. thank you for this video. I am having some issues in identifying the model and lags should use to do the ADF tests. I am doing a test for 8 variables for a period of 30 years. I am failing to understand which model to use whether intercept trend or no intercept. Besides i used lag of 1 for all. For example for GDP growth annual in its level form , both trend , intercept only says its stationary. But supress constant says its not. So what should I take as good??
Hi Ramsamy, with annual data it is advisable to use 2 lags and a model with "intercept". You can include trend if the graphical plot indicates a trending series or if the trend coefficient is significant. I explained these in detail in my videos on "optimal lag selection" and this particularly video. Hope these tips are helpful.
Thank you for your prompt reply! So basically you are saying i can use both lag of 1 and 2 irrespective of the variables am doing is that right?? In my Gdp growth annual graph The graph is not around O and its showing some sort of downward trend. Going by the graph only i would have said its not stationary. However doing the test with trend the test statistics is greater than Critical value meaning it is stationary. And if i still do the differencing the graph now becomes around O and stationary. So, which one should I be going for: the level or first differencing?
It's obvious that you didn't watch my videos on these issues because I explained these queries. Kindly do so, they'll really help you.
Hi thank you for your amazing videos. They're helping me with my dissertation that I have to submit in a week. Can you please help me with one thing. When I perform the ADF test my constant always appears to be insignificant i.e. the p value is greater than 0.05 for the constant. What does that mean? can I not use my series for forecasting?
if my constant is insignificant, shall i suppress it while running my ARIMA model? as that is what I'm using
both my trend and constant are insignificant when I perform the ADF test on the first difference of my variable.
Aleena, thanks for the positive feedback. Deeply appreciated! Keep it simple and use the ADF statistic/pvalue to determine the stationarity of the variables.
Hi, dear Ngozi, what's the meaning of "include drift term in regression?" does it mean that we are now using the first-differenced form to see whether the variable is stationary or not?
If you think your series exhibits a drift, then include it in your regression. Otherwise, not necessary.
Hi I have a panel unbalanced data, i used fisher type test only on dependent variables, like female labour force. I can see I have unit root. what am I supposed to do to remove it. Also do I need to check all my independent control variables as well ??
Kindly watch my videos on panel ARDL. I covered stationary test and other specifics. You will find them helpful. Thanks
@@CrunchEconometrix thank you very much, is there a chance I could ever chat with you?
I have used first difference but my R2 becomes very very low. 0.0015.
You didn't pay attention to my response. I indicated what you need to do. Watch my panel ARDL videos.
Is performing ADF Test for Panel data the same? If not, do you have any references?
Hi Michellia, the ADF algorithm for panel data is different. You can do a Google search on it for details and references. Thanks.
Hi,
I have a question on how to determine the lag numbers in ADF test. May I confirm to determine the lag number, I perform the Varcos [var] in STATA, correct? I performed varcos [var] in STATA, and chose AIC as a criteria. My data is monthly data, and I got an optimal lag number 4. Is this reasonable, as you have mentioned that for monthly data, the lag number is usually 6, 12 etc.
This lag number is different from the ARIMA (p,d,q) am I correct? Thank you so much!
Hi Liu, once any IC chooses the optimal lag, you stick with that. It is the same as the "p" referred to in ARIMA.
@@CrunchEconometrix hi, but my arima p is zero. my SARIMA is (0,1,1)(0,1,1,12). Is anything wrong?
Thank you very much for this video. I do have a question though. Is it a problem if my variable is still nonstationary when including the trend but stationary when including the drift?(i am using the first difference so, I'm not doing that part wrong)
Edit: my variable is ln of real GDP in quarterly data
Hi Silva, thanks for the positive feedback on my video. Deeply appreciated! This is not a problem. Simply indicate on the Table of results that the series is stationarity with a drift then proceed to estimation. Please may I know from where (location) you are reaching me?
CrunchEconometrix hello again! I am writing to you from Lisbon but am currently studying at the university of Nottingham! Let me take this chance to once again thank you for the clarity that you provide with your step by step analysis in your videos. It really is a game changer and has already answered a few of my many questions related to my thesis 😅. Thank you and God bless you
Hi, Why did you use 'trend' in the first differenced series? since the plot of the first difference series does not show any trend, isn't it wrong to use 'trend' option in the ADF test for that first differenced series?
Raihan, it is not wrong to use the trend option.
Hello, thank you for such an insightful video. I run my data just as you have shown in the video but after using the data at first difference to run the augmented dickey fuller test my test statistic value is still lower than the 5% critical value. I know this means that my data is still non stationary at first difference but i don't know what to do now to get it to be stationary. Can you help please? I'm stuck with my research now.
Hi Judy, don't panic. You can either change that variable to a closer proxy or estimate the Toda-Yamamoto technique. I've not had cause to use it but you can read up on it. You'll be fine. May I know from where (location) you are reaching me?
Yes please, I'll read about it and see how I can use it in my work. Thank you very much for your reply. I'm sending this email from China. I'm currently pursuing my Master's degree..
@@judylawson2588 Alright, Judy. Wish you the best. I'll appreciate it if you can share the link to my RUclips Channel with your friends and academic community in China for awareness. Thanks 😊
Hello, please does the fact of choosing 1 lag gives different results because in eviews 10 they set number of lag by defaut at 9 ?
Hi Fatima, modify the lag structure.
Mam can we increase the lag to2 while using ADF test in order to make my variable stationary or we have to maintain the lag value to 1
Yes, Marya, you can.
@@CrunchEconometrixmam can you please help me with narrative citation in ms word without using Mendeley.
Hi Marya, MS Word had an inbuilt citation menu that you can use. Also, do a Google search for citation software. There are thousands of them.
my R sq. value is 0.98 and watson statistic value is 1.92... does this means , i do not need to perform the stationarity test?
Hi Ekta, performing the stationarity test gives you the confidence required to proceed in time series analysis.
I appreciate u!!!
Thanks so much, Gubula! 🙏🥰
I noticed that lnpce has the letter D beside it making it D.lnpce. What does the letter D mean?
Hi Mark, it means "1st Difference " OR "Change".
@@CrunchEconometrix thank u so much for the reply. Love your channel as well as the content. Much love!!! ❤
Thanks, Mark for your encouraging feedback. Deeply appreciated!🥰🙏
Can you please tell me what kind of variable is qtrly? It's howing me as a string(as it contains letters and numbers) hence not able to use it along with tsset command. Would appreciate if you could tell me where I am going wrong. Thank you.
Got it how to be done. Thank you for these videos. Worth appreciating!
Thanks for the encouraging feedback, Priyesh. Deeply appreciated! Please may I know from where (location) you are reaching me?
Yeah sure:) It's New Delhi, India.
Please how do I generate the difference variables
Hi Joseph, use this syntax:
gen dy = D.y
Dw= 1.976556 and r^2=0.4179 but 3 out of 4 independent variables are not stationary according to augmented Dickey fuller test. What is the problem here?
Make the series stationary and re-estimate.
if only you had a unit root test tutorial on panel data in stata. Please upload one
Rahmata, watch my videos on panel ARDL. I covered 'stationarity test' on the series.
Hello sir, does lags(1) correspond to "betaLNPCE t-1" in the regression model? varsoc gave me the optimal lag length of 4, should i put in "dfuller, X1 lags(4)" ? Also, would my model look like this: "betaX1 t-4" if i want to do ARDL modeling? (assuming X1 is stationary at level with lags(4)). Thank you!
Yanuo, the answer to your 1st and 2nd queries is YES. For the 3rd, watch my video to see what I did and adapt your model.
@@CrunchEconometrix Thank you so much!
Why it is not correct to put Olags in the ADF test? I did not understand your explanation for that
0 lags*
Nicola, 0 lag = Dickey-Fuller test. With lag = ADF test.