(Stata13):Estimate Bounds Cointegration Test

[@CrunchEconometrix]

KZbin 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.

[@sabreenkhan3498]

❤

[@RememberRund]

For understanding all knowledge that you did explain, I must to study by myself minimum in a week, luckily I understood in 7 minutes, thanks

[@CrunchEconometrix]

Excellent, Allayor...thanks for the encouraging feeback. Appreciated!

[@asefabelay617]

thanks so much for your very helpful presentations. YOU HAVE FEW VIEWERS SO please please DO NOT BE HOPPLESS and dont stop doing this we are following you......

[@CrunchEconometrix]

Thanks for the positive feedback and encouragement, Asefa...Deeply appreciated! 💕 😊

[@tewodrosdargie81]

interesting and help full. thank you.

[@CrunchEconometrix]

Glad it was helpful Desalegn...thanks!

[@voittamaton9881]

Thank you, I really needed the clarification.

[@CrunchEconometrix]

Voitta Maton U're welcome Voitta. I'll appreciate if you can help share my KZbin channel link with your social media community and academic networks. It will really help, thanks!

[@voittamaton9881]

Thanks will do like wise.

[@audreyconnick720]

Thank you for the detailed video. I used the varsoc command to estimate the number of lags and later used these lags to estimate the error correction model which reflected that all the explanatory variables are insignificant. Using the same lag values for the bound test and the error correction model produced significant explanatory variables. What do you suggest?

[@CrunchEconometrix]

Hi Audrey, this is not an issue. You can put a note in your work that the model is not robust to using different lag lengths. Hence, you used the same lag to estimate the model.

[@tesfayewgwolde3549]

i am very happy when i see your presentation. Thank for your great contribution. Having this, i am not clear why you use maxlags (2) even if the varsoc command give optimal lag 1? thank you.

[@CrunchEconometrix]

Hi Tesfaye, thanks for the positive feedback. Using maxlags(2) is informing Stata to use the "optimal" lags instead of going the way of varsoc. Either approach is fine.

[@mdnaiemhossain6074]

I am looking for a bound test tutorial for panel data. I request you to make a video about the panel cointegration test in stata.

[@CrunchEconometrix]

Bounds test is not applicable to panel ARDL.

[@mdnaiemhossain6074]

@@CrunchEconometrix thanks a lot. but how can I decide about short run or long run and ECM for panel ARDL? thanks in advance.

[@CrunchEconometrix]

Kindly watch my panel ARDL videos.

[@btessa2dc]

Thanks for this video. Can we perform the bounds cointegration test with panel data? I am working with panel data and I have a combination of I(0) and I(1) variables. From your video, bounds cointegration test is more appropriate in this case. The flip side in this case is that I have panel data and not time series, as in your example. So, I am confused if I can use Pedroni/ Westerlund/Kao or if there is bounds cointegration test available for panel data.

[@CrunchEconometrix]

No Tess, you cannot use the Bounds test for cointegration for panel ARDL. In my opinion, performing cointegration test is optional because the significance of the ECT evidences cointegration in the model. But no harm in performing Pedroni or Westerlund tests if you choose to. You may want to watch my videos on panel ARDL posted about 2weeks ago. They'll guide you on how to go about it. Hope these tips help.

[@ahlemouhibi3582]

thanks a lot for your video. I want to ask you a question, how do you are choosing the maximum lag 2?

[@CrunchEconometrix]

Hi Ahlem, I showed and explained in almost all my time series videos.

[@josephazumah7032]

I followed the command for performing bound test for ardl but stata keeps referring command not recognized. What could be the problem

[@CrunchEconometrix]

It implies that you are yet to install the synax. Type this in the command window: "help ardl" and follow the Stata prompts.

[@prabirghosh5015]

At first, I want to thank you for your beautiful work on time series econometrics in STATA. I just want to know, when the variables are I (1) i.e. stationary after taking their first difference, in this case which Co-integration test is appropriate?

[@CrunchEconometrix]

Hi Prabir, thanks for watching my videos...deeply appreciated! Regarding your query, both Johansen (if VAR) and Bounds (if ARDL) are applicable. I'll appreciate if you can share my Channel link with your students and academic networks...gracias!💕

[@sabreenkhan3498]

Thank u so much for such great work. I just want to confirm about the lags to be used if our bound testing suggests taking say 1 1 0 lags as in case of the above mentioned eg shall we take same lag length when we further go for ARDL PMG mg model analysis or do we have to test the lag length by versoc

[@CrunchEconometrix]

Sabreem, Bounds testing is NOT applicable to PMG/ARDL analysis.

[@NhiLe-lr9es]

Thank u so much for your great video. I just have a small question. If I don't write the section "maxlags(...)" for the command "ardl var1 var2..., maxlags(...) aic", will Stata automatically choose optimal lags for my model? Because I have already done that and Stata really chose optimal lags for me, but I don't know whether I should believe that result or not. Thanks again ♥

[@CrunchEconometrix]

You are on track, Nhi. Well done.

[@nausheensodhi803]

Thank you for this insightful video! It has been helpful in understanding the concepts. However, I needed to ask if the Bounds test can also be applied on panel data? I have a balanced panel with 4 variables and 19 units (for 15 years), comprising two I(0) and two I(1) sets. Could I apply bounds test on this? Thanks

[@CrunchEconometrix]

Hi Nausheen, thanks for the encouraging feedback. Bounds test is only applicable for time series analysis. You can deploy Pedroni or other panel cointegration tests.

[@maleselasalvation]

How do we know the maximum lag on the code for "ardl y x1 x2, maxlags (!!!) aic"?

[@CrunchEconometrix]

You may want to watch my video on OPTIMUM LAG SELECTION.

[@dominicwong8797]

Thank you for the video. I have a question regarding the ardl command: For the maxlags section, is this determined by the varsoc command on all variables or each individual variable?

[@CrunchEconometrix]

Hi Dominic, the "maxlag" option produces the optimal lags for the variables listed in the equation.

[@sabreenkhan3498]

ALSO, i want to know is Bound testing applicable to Panel data or can we use it only in case of time series

[@CrunchEconometrix]

Sabreem, please know that Bounds testing is NOT applicable to panel data analysis ONLY time series.

[@MauLov22]

Thank you. The presentation is quite clear. Would it be possible to have a video on panel data cointegration test?

[@CrunchEconometrix]

Thanks Mo, it's on my to-do-list. Once I clear my backlogs, I'll do a video on it. Kindly share my YT Channel videos and links with your colleagues...:)

[@zanyatwa]

Thanks, for your clear presentation! would you mind sending the do-file for ardl and cointegration to my email: [e-mail]

[@CrunchEconometrix]

Access all Stata dofiles from my website. The link is at the end of the video.

[@joselopes4588]

Dear Dr. Ngozi, I hope you are doing well. I am working on an unbalanced panel data and my series are stationary at I(0) and I(1). For this reason I am using the ARDL. I understand also that I cannot use bounds test for co-integration because this can be used only for time series. What co-integration test should I use? Thank you in advance.

[@CrunchEconometrix]

Hi José, kindly watch my videos on panel ARDL. Well explained.

[@mehmetakyol4332]

Dear professor, may I ask you a question. is there any command for panel ardl bound test ?

[@CrunchEconometrix]

Hi Mehmet, Bounds test is not applicable to panel ARDL.

[@davidmoreno5539]

Thanks for sharing your knowledge. I just have one question, can I use the bounce test if one variable is I(1) and the other is I(2). thanks for your help

[@CrunchEconometrix]

Hi David, thanks for your encouraging feedback. Bounds Test is not applicable but the Toda-Yamamoto test. I have no video on it yet so you may want to check out other online resources. Best regards.

[@davidmoreno5539]

@@CrunchEconometrix Thank you very much for answering, your videos have helped me a lot in college. Have a good day

[@khemaram3790]

Hello Ma'am.. I have obtained F statistic of 30.75. It is quite high compared to my critical values.. is it ok to conclude co integration or there is an issue here? please help urgently

[@CrunchEconometrix]

Hi Khema, the higher the F-stat, the better. It convincingly provides evidence of cointegration

[@MrLothman]

Excellent explanation! Just a question... Can we perform the bounds test for cointegration when all the variables are I(1)? Thanks in advance

[@CrunchEconometrix]

Yes!

[@fuzhufeifei]

This video is helpful. Thanks very much, dear Professor. My model has a few variables. May I know how to add one more variable, which is the initial income level? My proxy is log GDP for the beginning year, namely Log GDP at year 1982. When I ran ardl v1 v2 v3, maxlags(2) aic, I had to drop this variable. This is because there is collinearity issue. But I read a few papers, researchers put the initial income level to test its impact on economic growth. Dear Professor, may I know how to put initial income level, proxied by log GDP in the first year in Stata?

[@CrunchEconometrix]

It means log GDP will enter your model with 0 lag. The question is the optimal lag =0? If yes, then you can re-specify your model using this code (also in my ARDL dofiles): ardl y x1 x2 x3, lags(1 0 0 1) aic. This tells Stata to estimate the model with 1 lag of the depvar, 0 for the first 2 explvars and 0 for the 3rd explvar. Always remember that your model, research scope and coverage is not the same as used/conducted by the papers/authors read, so never expect the same outcome.

[@fuzhufeifei]

Got it, thanks very much! @@CrunchEconometrix

[@kanchandatta4668]

mam, if a series is I(2) and others are I(1) which model should be appropriate ? ECM, ARDL or VAR? Please response

[@CrunchEconometrix]

Kanchan, I suggest the Toda-Yamamoto technique in that case. Please check other online resources on how to go about this since I have no video on it, at the moment.

[@kanchandatta4668]

@@CrunchEconometrix thank you so much mam. One thing can I use 1st difference data of I(2) and others are I(1) variables to run VAR or VEC ECM?

[@CrunchEconometrix]

Kanchan, "1st difference of I(2)"? What does that even mean?

[@kanchandatta4668]

@@CrunchEconometrix it means I know that a variable (say unemployment) is a I(2) variable. Now if I take 1st difference data of unemployment that is rate of change of unemployment as one variable which is now became I(1) variable and other variables of the study which are say I(1). Now the two series are I(1). We can proceed for cointegration or other tests. If they are not coinegrated then VAR model. Mam these are my assumptions, I don’t know is there any theoretical support exists in literature or not?

[@CrunchEconometrix]

Kanchan, please read about STATIONARITY from any basic econometrics textbooks. 1st differencing is NOT rate of change.

[@bav1362]

Thank you for a very clear explanation! I'm using 3 variables and my F-statistic is greater than I(1) in all cases except the last case when L_01! In that case F-statistic is between the I(0) and I(1). How to interpret this result? Thank you very much for these videos!!

[@CrunchEconometrix]

Bav, use the OVERALL F-stat as I explained in the video. Thanks.

[@najeebahjoomun813]

Please help.. i have 1 dependent and 6 independent variables, what do i put in the "maxlags(2)" part of my command?

[@CrunchEconometrix]

Najeebah, watch the video again and my other ARDL videos on what to do. Thanks.

[@jayshreeacharaz9023]

Why do we put max lag 2

[@CrunchEconometrix]

I have limited observations, so I constrained the maximum lag length to 2.

[@jayshreeacharaz9023]

@CrunchEconometrix for example if I did varsoc for each variables I have and the highest is 3 can I put maxlags(3)

[@CrunchEconometrix]

Please watch the video again and adapt to your analysis. That's the essence of creating the tutorial.

[@peshalamadhuwanthi217]

Thank you madam for the lesson and your exellent explanation.. If my data series have combination of i(0)and i(1) stationary .when doing bound test what should i select as my variables . It's ok to get both level form variables and 1st different form variables. Can you explain me..

[@CrunchEconometrix]

Hi Peshala, please watch the video again with other ARDL videos. Follow the guides shown. Thanks.

[@thetruth4712]

Dear prof. Thank you for this video. This was very useful. BTW, what can we do if bound tests results show inclusive? Best wishes.

[@CrunchEconometrix]

"Inclusive"?

[@thetruth4712]

@@CrunchEconometrix Sorry prof., I mean inconclusive (the F value was in between the value of I(0) and I(1)). In this regards, what i can do. Best wishes.

[@CrunchEconometrix]

Estimate the unrestricted model.

[@thetruth4712]

@@CrunchEconometrix Dear Prof., does this mean just using the short-run ARDL model? Regards.

[@CrunchEconometrix]

Watch "This is how to specify ARDL models". Well explained.

[@sibylla553]

Thank you for providing a video tutorial. I'm running a cointegration test and one variable is omitted due to collinearity i don't know why. I have 8 observations.

[@CrunchEconometrix]

Bylla, but you just stated why: due to collinearity.

[@sibylla553]

​@@CrunchEconometrix My Durbin-Watsons are higher than R-squared. Is that the cause of collinearity?

[@CrunchEconometrix]

Watch my video on MULTICOLLINEARITY. I pointed out things to note. You will find it helpful.

[@sibylla553]

Thank you

[@rachelbabs9951]

Hello, thank you so much for all your videos, they are honestly so helpful. I'm currently trying to run a VAR model with 6 variables. 4/6 of the variables are non-stationary after ADF test, but become stationary at first difference. 2/6 of the are stationary after ADF test. One of these 2 variables remains stationary at first difference. However the other variable (which is in log form) becomes non-stationary at first difference. Can I estimate a VAR model with 2 variables at levels and the rest at first difference. If not, do you have any further advice? Thanks in advance for your help.

[@CrunchEconometrix]

Hi Rachel, thanks for the encouraging feedback on my videos. Deeply appreciated. Firstly, all variables must be I(1) to estimate VAR and secondly, reduce your variables to 3 or 4. Thanks.

[@sumayatasnim782]

Why did you use the in form on one variable but not in others?

[@CrunchEconometrix]

Taking natural logarithms depends on the variable and at the discretion of the researcher.

[@NusratJahan-eb6ox]

Thank you for the video. But in my data set of 4 variables(including dependent) I am not able to use bound test. When I enter command stata shows "command ardl is unrecognized". "Matrix e(lags) not found" Can you please tell me why this is happening?

[@CrunchEconometrix]

Nusrat, you are yet to install ARDL. Type "help ardl" in the Command window to get guides on what to do.

[@rosebanda3134]

Hi. Thanks again for the videos. 6 variables (dependent included). The bounds test wont run with more than 4 independent variables. when i run it excluding 1 variables, i am successful in 1. checking for break in the constant and 2. checking for break in the constant and trend. However, i cant check the break in constant and slope. i get the error that says adf(): 3301 subscript invalid main(): - function returned error : - function returned error what does this mean and how do i resolve it. Also how do i explain the exclusion of one variable. Thanks

[@CrunchEconometrix]

How many observations do you have? If you have initial 30 years with 6 variables, their lags and differences will result in loss of degrees of freedom. So, you need more time span.

[@addaada321]

If the test is inconclusive do we have to estimate only a short-run ARDL? Best

[@CrunchEconometrix]

Yes. Or you can decide to change control variables and re-estimate the model for a better outcome.

[@yanuozhou6028]

Dear sir, does the number in 'maxlags()' need to be the same for all bound testing models? I'm doing an ARDL. For one specific model, using the same maxlags number as the others gave me the problem of serial correlation. in this case, can i use a different maxlags() number that does not give me such a problem?

[@CrunchEconometrix]

Yanuo, I answered your question in the introductory ARDL video: "How to Specify ARDL Models". Kindly watch it and jot doing some note while at it. Thanks.

[@sharminkeya46]

Sir, I am facing a problem when I use (matrix list e(lags)) command. Stata shows that matrix e not found.How could I overcome this problem?

[@CrunchEconometrix]

Hi Sharmin, click on the error code and follow the prompts. Thanks.

[@dianalouisecatap9546]

What if the obtained f-value falls below each of the I_0 bounds except at the 10% level where it is between I_0 and I_1? Is it still inconclusive?

[@CrunchEconometrix]

Yes.

[@mrzadocknanyaro7584]

Hello , i have followed your videos for my project where i study the determinants of foreign direct investment in a specific country(1990-2018) . the final model i run is shown below. does this model have too many variables? might suffer from autocorrelation? the average VIF is around 6. the model is ardl LNfdi dLNintr dLNreer LNgdp dLNtsub dLNnetoda dLNinfla dLNtrade, lags(2,1,1,2,1,1,1,2) ec ARDL(2,1,1,2,1,1,1,2) regression Sample: 1993 - 2018, but with gaps Number of obs = 20 R-squared = 0.8888 Adj R-squared = -1.1129 Log likelihood = 9.3552255 Root MSE = 0.6778 ------------------------------------------------------------------------------ D.LNfdi | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- ADJ | LNfdi | L1. | -.0966123 .655951 -0.15 0.907 -8.43126 8.238035 -------------+---------------------------------------------------------------- LR | dLNintr | -1.660472 28.00881 -0.06 0.962 -357.5462 354.2252 dLNreer | -9.389247 124.8792 -0.08 0.952 -1596.13 1577.352 LNgdp | 1.578758 18.44437 0.09 0.946 -232.7792 235.9367 dLNtsub | 29.84937 245.8256 0.12 0.923 -3093.66 3153.359 dLNnetoda | -15.98767 102.0813 -0.16 0.901 -1313.053 1281.078 dLNinfla | -12.32564 103.1859 -0.12 0.924 -1323.426 1298.775 dLNtrade | 18.97839 132.6266 0.14 0.910 -1666.202 1704.159 -------------+---------------------------------------------------------------- SR | LNfdi | LD. | -.151286 .764658 -0.20 0.876 -9.867187 9.564615 | dLNintr | D1. | .1963492 1.764978 0.11 0.929 -22.22982 22.62252 | dLNreer | D1. | -.9500715 3.607655 -0.26 0.836 -46.78968 44.88954 | LNgdp | D1. | .9667242 2.661247 0.36 0.778 -32.84763 34.78108 LD. | -.2588802 1.965706 -0.13 0.917 -25.23555 24.71779 | dLNtsub | D1. | 2.568221 6.826822 0.38 0.771 -84.17478 89.31122 | dLNnetoda | D1. | .6592227 2.432948 0.27 0.832 -30.25431 31.57276 | dLNinfla | D1. | .9288262 3.017507 0.31 0.810 -37.41223 39.26989 | dLNtrade | D1. | -1.94197 2.835643 -0.68 0.618 -37.97223 34.08829 LD. | .373256 3.065926 0.12 0.923 -38.58302 39.32954 | _cons | 1.847643 13.60531 0.14 0.914 -171.0242 174.7194 ------------------------------------------------------------------------------

[@CrunchEconometrix]

There's no need pasting your results. Due to time constraints, kindly briefly explain what the issues are.

[@officialmintt]

What model could I possibly use for a set of variables with the same difference level but different optimal lag lengths?

[@CrunchEconometrix]

Hi Mint, use ARDL.

[@officialmintt]

@@CrunchEconometrix You save me again. Thanks!

[@mrzadocknanyaro7584]

when running the models why dont you use the variables in their first difference? or is this just for ADF test?

[@CrunchEconometrix]

The ARDL algorithm allows the use of variables in their raw forms. Kindly read the refs at the end of the video for more information. Thanks.

[@suhrabkhan3696]

which model i can run can if the ARDL bounds Test is inconclusive about co-integration.?

[@CrunchEconometrix]

Hi Suhrab, either estimate the ARDL model or change your variables and redo your analysis.

[@joselopes4588]

Dear Dr. NGozi I am running the command varsoc var1 var2 and the output is: "repeated time values in sample". I checked for repeated values and there is none. What is wrong?

[@CrunchEconometrix]

Hi Jose, is your data in panel form or a time series?

[@joselopes4588]

@@CrunchEconometrix my data is in panel form.

[@joselopes4588]

If I cannot use varsoc for panel data, what do you suggest ?

[@CrunchEconometrix]

Ok. Watch my panel ARDL series on "Optimal Lags". You may find it useful... how many countries and years in the panel?

[@joselopes4588]

14 countries and 23 years.

[@jahnavipratishtha]

Hi so suppose my F-stat of 4.008 is > I(1) of 10%, but my t-stat (-2.506) is less than I(0)= -2.57 and I(1)= -3.46. Do I still have cointegration at 10% significance level or not?

[@CrunchEconometrix]

Jahnavi, my explanation is quite clear. Adapt and interpret your result.

[@jahnavipratishtha]

@@CrunchEconometrix So meaning we consider only the F-statistic, not really the t-statistic. Since my F-stat is greater than the 10% I(1) bound i have cointegration? (However, this is F-stat is only > I(1) bound at 10% is that okay?)

[@jahnavipratishtha]

@@CrunchEconometrix Thank you so much for your help! Your videos are really insightful.

[@MrStaron47]

I have a question, after i run my data using the code ardl lnmva y x1 x2, maxlags(2) aic, it is said that my ardl command is unrecognized, what should I do?

[@CrunchEconometrix]

Have you installed the ardl syntax? If not, then do so.

[@mdnaiemhossain6074]

@@CrunchEconometrix me also face this problem. so, how can I install this ardl syntax?