Hi beloved guest/subscriber, you have discovered my amazing TH-cam Channel tailored specifically for you and other beginners and intermediate users. Please do not keep me to yourself (lol). Kindly share my videos and links with your students, colleagues and academic community so that they too can SUBSCRIBE and learn with ease….and for the global community to be aware that applied econometrics can be simplified. My teaching approach is very practical. I adopt a do-as-I-do style. Many thanks to those who have supported me by telling others. Once again, CrunchEconometrix loves to teach, support my Channel with your subscription, your likes, feedbacks and sharing my videos with your cohorts. Follow me on Facebook, Twitter and Reddit. Love you all, greatly!!! :)
CrunchEconometrix thank you so much for the eye opening videos. I posted a question on your blog about an ARDL model and a financial crisis. Hopefully you will be able to attend to at the convenience of your schedule. Once again thank you 🙏🏿
Hi Andrea, my sincere apologies. Time constraints has not allowed me to visit my Blog as I should. So, please always post your queries on the comment section of the respective TH-cam video. I'll respond from there. Thanks.
tThank you for your video. I have a question. My thesis topic is predictability of stock return ( alexakis. et.al. 2010)Use GMM for estimation. I have read from journal. It show that they did unit root before. In this case if it possible to use d.l.x. (Both lag and first difference) when running xtabond 2. Could you please help me?
Hi Ma'am, Your videos are of great help ma'am. They have simplified GMM so much. Ma'am, I am a PhD scholar at IIT Delhi, India in the area of finance and have been using your videos for my PhD. They have proved to be beyond helpful. I am in the last leg of my PhD and am facing some issue in the result. I am getting no values for AR, Sargan and Hansen test. I will be immensely grateful for the help.
@@CrunchEconometrix Thank you for the response ma'am, but all the variables except 3 have 100% data. the 3 variables jave 99% data. Dependent Variable too has 100% data
Dear professor, thanks for your videos. They are very helpful! I already followed your steps and made estimations with 8 different specifications: System GMM and Difference GMM, one and two steps, and ortoghonal and not ortoghonal for my master degree investigtion, because I wanna know if inequality and redistribution affected the colombian economic growth, and I wanned to have robust results. I kindly ask, besides Sargan, Hansen, AR(1) and AR(2), which other tests can i apply to asure that the estimations are free from any other problem. Thank you very much for your attention.
I am learning alot thanks to you Miss, @ 3:38 you say "AR2 not significant tells you that your model does not suffer from second order serial correlation" you mean it does suffer (no?)
Once the AR(2) > 0.05, the model does NOT suffer from 2nd-order serial correlation. That null hypothesis of 2nd-order serial correlation CANNOT be rejected.
Thank you very much for really helpful videos! I have a question. For a panel data estimation with t (11 years) and n (19) I used GMM like you taught us. Besides, what method can I apply in order to support my results? Thanks.
Thanks, Romanus for the positive feedback on my video, deeply appreciated! Kindly share my TH-cam Channel link with your friends, students and academic community on social media for awareness. They'll learn some useful tips and skills too...thanks 😊
Dear Dr.Adeleye : Thank you for your wonderful videos. I am learning a lot from you. I estimated 1-step Sys GMM and 2-step sys GMM but some coefficients are significant in 1-step but not significant in 2-step and both have good Hansen and Sargan results. Which should I report in my manuscript?
Great videos mam.Your videos helped me a lot in understanding gmm and with your help, I was able to write a research paper. Thank you very much. I was hoping if you could make some videos on estimating simultaneous equation using system gmm.I.m unable to find the commands for the same on stata.
Hi Naina, I'm encouraged by your commendation and positive feedback. Deeply appreciated! 😀 I'll do my best to make the videos once I fully understand the techniques. Please may I know from where (location) you are reaching me?
Hello Dr. again! How are you? I hope you 're doing well. I have few questions waiting for your light answers to reflect on. No.1 Is it mandatory to include robust option when running xtabond2 since most of my variables of interest are not significant? No.2 When I include y*, the same thing happen again including y*. So do we expect y* to be significant? Thank you so much!
Hi Farid, as I explained in the video the "robust" option controls for heteroscedasticity. Please watch the clip on "GMM Year Dummies" to understand the relevance. Thanks
I have watched most of your TH-cam videos on GMM, and they are beneficial videos. I have a question to ask you, and I hope you will provide your expert advice on my query.
I have panel data with N>T case. After estimating the results using the two-step system GMM, should we test for cross-sectional dependency? As I understand, the existing tests proposed by the Breusch and Pagan (1980) LM test, Pesaran (2004) scaled LM test, Pesaran (2004) CD test, and Baltagi et al. (2012) bias-corrected scaled LM test is to be used after Fixed effect/Random effect test. They do not refer to the two-step system GMM case. I highly appreciate your advice in this regard. Thank you very much, prof.
Hi Sri, I gave the background to the GMM technique. You may need to watch it or check other online resources for more information regarding your queries. Thanks.
Thank you ma'am for your wonderful lessons, I am trying to apply the two-step system GMM, using the xtabond2 command, but unfortunately not able to get the result. Instead, I got this error "////// invalid. r(198); could you pls shed some light in this regard.
I have unbalanced panel data of 218 companies and I taken lagged dependent variable. Also taking Interactive Term, it is the product of financial crisis(dummy variable) and Div.payout( independent variable). At first I run simple OLS and after that I test for heteroscedasticity and then VIF, Correlation for testing multicollinearity. I also test Dicky-fuller augmented test for checking stationarity.After that I run two step system Dynamic Panel Regression with Robust of Arellano-Bover and Blundell -Bond estimation. Now my question is that whether auto correlation is must for panel data anlysis? and also whether Sargen test is compulsory for panel data..
Dear Professor, I am using two step SYS GMM for my analysis. I have a regression in Stata like this: xtabond2 y l.y x1 x2 x3 where x1 is an endogeneous variable meanwhile x2 x3 are exogeneous. Should it be right this line code in Stata? xtabond2 y l.y x1 x2 x3i.year, gmm( l.y l.x1, collapse) iv( x2 x3 i.year, eq(level) robust ecc... Should I include the lag of my endogenous variable in twice ( ie xtabond2 y l.y x1 l.x1 x2 x3...) or it is okay if I include the lag of endogenous variable ( l.x1) just in gmm()?
Thank you for all your videos. They were very useful and helpful. Keep shining. I have a question please: Some references put the option lag () in the xtabond2 command synthax, for example lag (4 7). Could you please explain what's the meaning of this option and when we should use it? Thank you in advance.
Prof. I would like to know if it is as well possible to run this estimation with the use of EVIEWS ? I have actually checked EVIEWS but I couldn't find the system GMM option.
Hello mam... your doing great job... i tried this method on my data but AR (1) , AR(2), sargan test and hasen test are significant... now what should i do?
Thank you for the video! It was very helpful. Nevertheless, while listening some questions crossed my mind and I would really appreciate if you can answer: 1) Is Windmeijer correction automatically applied when we use xtabond2 command? And if not what additional command should be used to account for it? 2) If AR(2) test is not significant that means that the model does not suffer from second order serial correlation. But what if AR(1) is also not significant (example: 0,200) does that indicate a problem ? 3) I think in every two-step regression a warning appeared saying "Two-step estimated covariance of moments in singular. Does that also indicate a problem and what does it mean exactly? 4) Related to the 3rd question if the program tells you "The difference in Sargan/Hansen test may be negative is that a problem? Thank you once again !
Hi Akim, thanks for the positive feedback. Deeply appreciated. I gave some clear explanations in the video, but where in doubt, kindly refer to the references at the end of the video for more information about the GMM technique. Thanks.
Great videos series. it really help me on my dissertation. one question please. do we need to do diagnostic data test such as multicollinearity test, normality test, heteroskedasticity test, outlier test before do two step system gmm? thank you very much
Hi Hamba, not at all. Only correlation analysis is required which is a PRE-ESTIMATION test. Kindly watch my video on MULTICOLLINEARITY on why this is essential in ALL empirical analysis. Thanks.
@@CrunchEconometrix Thank you very much for your response. definitely will go through your video. One more question please.... For two step system gmm, is it necessary to include "robust" option, since when I applied it, many of my variables become not significant? when I am not using the "robust" nearly all my variable is significant and AR(2) and hansen test is good. What is your opinion? thank you very much
Hi, Thanks for your video. What can be the cause of the problem with the following warning? (condition with my estimation: n of instruments < n of groups) "Warning: Two-step estimated covariance matrix of moments is singular." Using a generalized inverse to calculate optimal weighting matrix for two-step estimation.
Hi Xez, thanks for the positive feedback. Deeply appreciated! I'll suggest you don't worry about it. It appears in all Stata GMM output. Concentrate on the results. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hi there and thanks for your answer. I reached you from Germany. Why do you think that using a generalized inverse W Matrix is nothing one should worry about? I read so far different explanations why this could occur, and some are indeed worrying to me.
Thank you maam for your great lessons, I have 10 indep.variables and 4 control variables out of them six are endogenous. Now should I include the lag of only this six or all as the Instruments? Dear pls reply it's one month my work is stuck at this point.
Hi Ms Adeleye, many thanks for your swift respone. I am running the Two-Step System GMM. In some specifications, the outcomes for time-invariant variables (e.g. coefficients) are reported as omitted. When I change the specifications (e.g.by removing some variables), the outcomes for the time-invariant variables are normally presented. Why is that?
Dr Adeleye, your videos have been so useful for me! I wonder if you could help me with two questions. Is it possible to include many dummy variables in the two-step system GMM? To be more precise, I want to include a dummy for the year (y), region (h) and technological intensity (g). Following your example: xtabond2 lngdp l.lngdp lngfcfgdp lnlabr infl lnintuse lnmob lnfts y* h* g*, gmm(l.lngdp lnmob, collapse) iv(lngfcfgdp lnlabr lnintuse lnfts y* h* g*) twostep robust orthogonal small
Hi Elizabeth, thanks for the positive feedback. Deeply appreciated! I suggest you use the dummies sparingly. That is, one at a time to avoid your model breaking down due to insufficient instruments. Please may I know from where (location) you are reaching me?
Dr. I take lag 3 for my dv in two step system gmm. Iif i take that then i can make the AR1 AND AR2 INSIGNIFICANT ONLY. IS IT OK IF I TAKE LAG 3 OF DV? HOW TO JUSTIFY THIS DR? PLEASE, I NEED TO KNOW DR. IF AR1 is significant then is it ok to report on thesis?then what is the justification? Thank you dr.
Hi Ms. Adeleye, thnx very much these very useful tutorial videos. if I include limited dummies years as instruments ONLY (NOT in xtabond2 part), shall I put in the output GMM table "time dummies" YES or NO?
@UCK9hD254JKbCZ4Bf8Iz1s7g Hi dear Adeleye. Firstly thnx very much for your prompt response. Secondly I do apologies if I posted my question in inappropriate way (I am not an English speaker and beginner in GMM). I was trying to come direct to my question. However, let me re-posted as follow: I am running system GMM, my explanatory variable is dummy (xtabond2 lny L1.lny FTA ORECs lnOP lnICEXT lnPCDVA SEPG DCPSGD PSAV , gmm(lny,lag(2 2) collapse)iv(L2.lny l1.FTA OREC s lnOP lnICEXT lnPCDVA DCPSGD y9-y13, equation (level) ) nodiffsargan twostep robust orthogonal small. While I was shuffling variables I found the best model result as when I use dummy years as instruments , that I do by "ONLY" to best of my knowledge. if is that true, shall I put yes or no for time dummies in output table?...hopefully I could manage to make it clear.
@@yassinyahia2453 Well, a reviewer will want to know why you did not control for time variations. Please watch my video on that (if you haven't) to know the essence of time dummies in GMM procedure.
Hello Ngozi, Thanks for your commitment and effort. I have a question, in my 2 step sys GMM result, my first GMM regression AR(1) =0.380 while AR(2) =0.775. AR(1) is obviously out of place. how can i reduce my AR(1), or can i ignore my AR(1) since AR(2) is more important(as you said in one of your videos). The hansen test is 0.807 and sargan test is 1.000, how can i reduce the sargan result or can i ignore sargan result and use just hansen?? In my 2nd regression, the AR(1) and AR(2) was good, but none of the variable was significant. What could be the reason??? when i try to change the variable and lag, I got omitted result(blank) for all the variables, can i accept this result??. best wishes
Dr Adeleve thank you for this video. I have some questions I would be grateful to answer me. In my model I have 1 IV and 6 moderators and 6 control variables. So can I I apply all moderators in one commands like (xtabond2 dv l.dv IV IV*modert1 IV*modrat2.... IV*modrat6 control variables y*, gmm(l.DV) iv(IV IV*modrat1.... IV*modrat6 control variables y*, equation(level))
Caacay, I am very certain that you won't get any meaningful results due to the many regressors. You will have little or degrees of freedom left. Trim down your regressors and follow the steps shown in my video to engage the analysis.
@@CrunchEconometrix okay dr. my question is, depending on what we select the iv (instrument variables)? Regarding your replay "trim down your regressors" how i do that? is it by deleting some of the control variables or by applying only one moderator then the other ...etc. thank you dr
hi dear doctor，just wanted to ask if the coefficient of lagged dependent variable（instrumental variable）is not significant，shall we continue the test，or the model its not valid？thanks
@@CrunchEconometrix dear doctor，thanks for your fast reply，sorry to ask one or two more questions， 1. usually i do correlation and coefficient test plus VIF test for the variables，and then i start GMM test，is there any other test before executing GMM test？ 2. as you said，even the coefficient of lagged depended variable（IV）is not significant，so model is still valid? then，another question is，after running GMM test，I do run Sargan test and Abond test, but in many cases, even I go for two-step and VCE test, though AR(2) is more than 5%,( so we accept the results?), but AR(1) sometimes still less than 5%, so in this case shall I accept the VCE results or should go for other test? Thank you for your kind answer, I am really eager to capture GMM model especially for longitudinal research work,thank you so much dear doctor.
Dear Ngozi ADELEYE hope you are doing? how to fix ivstyle invalid issue in stata while estimating GMM estimation. after understanding your model Explanatory, control, and endogenous variable. I have put exactly mine model in the same sequence for estimation. Please guide me.
Tahirah, I'm not familiar with the error. But you can click on the Stata error code and follow the prompt on identifying the problem which will guide you on fixing it.
Hi Seyede, dofiles are no longer free but available on my website upon payment of a token fee after which you are allowed a one-time download. Kindly find the link cruncheconometrix.com.ng/shop/
Hi there, y* is the short form of including all the year dummies into the analysis. Also, due to abuse and unethical conduct Stata dofiles used in my videos are no longer free but available on my website upon payment. Here's the link cruncheconometrix.com/view/datashop.php
Very informative video ma'am.do we include lagged values of independent variables as instruments or other variables outside the model not taken initially?
Dear Ngozi ADELEYE hope you are doing? I am really inspired by your work. I am using System GMM for my Ph.D. dissertation but most of the variables have negative values even dependent variables also containing most of the negative values. I am unable to take the log of dependent variables when I take the log this made the panel unbalanced and the sample size will also be reduced? how do I fix this issue please guide me?
Good morning Professor, your videos are really helpful. I am following all the steps of the tutorial .But the problem is my Hansen Test is .341. It is the lowest among all the steps. Now what should I do? Will my analysis be correct if it is .341? Thanks in advance. Have a nice day.
i have two specific questions. I took data for 23 Pakistani banks for 12 quarters and using system GMM in this research using STATA. I am unable to understand what is meant by the term ‘instrument’ here since I just used two endogeneity variables but system is showing number of instruments to be 65. In my initial result, number of instruments is greater than number of groups; Would system GMM be affected by this? does 'collapse' is the only option to reduce it? I need help in this regard.
Thanks, Dawit for the encouraging feedback, deeply appreciated ❤️. But NO. GMM is suited only for N>T panel structure. I explained this in the introductory video.
Mam, When I run the xtabond2 commands in stata as a panel data analysis for the period of 2005-2018. then result omit two years due to collinearity problem i-e 2005 and 2015, I know that 2005 omit due to base year but 2015 omit due to collinearity how to solve this problem.
Hi Majid, I will tell you not to worry about this. Stata will usually drop several year dummies due to collinearity. What is important is that is you obtain good results and your models passes the mandatory diagnostics.
I applied two step system gmm, but most of my variables are not significant, while Sargan and Hansent tests hold a p-value of more than 0.05 and there is no serial correlation in the second order. Dear ma'am pls guide me in this regard.
Hi Farhan, this is not a problem. You can change your variables OR try all the 4 GMM variants: one-step/2step diff/system to know which one gives you the best results. All GMM variants are relevant to the literature. I have used the difference GMM in recent publications due to my data not being robust for system GMM analysis.
Good morning prof. Could you please (if possible) throw a bit of lights on this problem I encountered with system GMM? I followed the video on Difference and System GMM to do an analysis. For this work-balanced panel, I used the system GMM, however, the results have been very discouraging. In your videos, you emphasized on the collapse command which I have used, orthogonal, Robust as well, but the results are still not good. AR1 & AR2 both insignificant , Sargon and Hansen also are both insignificant. All my variables are insignificant as well. I am wondering what the problem could be.
Hi Ateh, from my videos you would have known that insignificant diagnostics are good. But your model failed the coefficients test. The silent rule is that at least 50% of coefficients must be statistically significant. Change your regressors, run several simulations ane check your outcomes.
Thank you for the videos, they're very clear and precise; I'm trying to use system GMM with my dataset but I'm not sure about its adequacy, because it has 17 periods but 11 variables. In your opinion could the results be unreliable? Should I use other techniques like PMG or similar mentioned in one of your videos? Thank you so much :)
Hi Ciak, thanks for the feedback... deeply appreciated. Make sure you watch the foundational GMM it's features and determine if your data fits the technique. Kindly do same with the panel ARDL foundational video. Thanks.
@@CrunchEconometrix I've watched your videos about the adequacy of SysGMM with my dataset and I know that I can use this technique. Just not sure about the meaning of Hansen excluding group and Difference tests for GMM instruments levels.
Thank you so much Ms. Adeleye... You really help me. I'm a beginner in stata and also on statistics. Do you have a tutorial or tips of how to plot margin effects after gmm. I tried to use margin and marginplot but stata shows "warning: cannot perform check for estimable functions"
U're welcome, Budi😄...I'm glad to be of help but I have no idea about plotting margin effect. Why not post the query on Statalist.org I'm sure you will get answers on what to do. May I know from where (location) you are reaching me?
@@CrunchEconometrix Oke Ms, I'll try it. I'm from Bali province, Indonesia. I am an undergraduate university student, I've seen many papers use GMM thus I searched about it and found your videos. I've watched all your GMM video series, those are really helpful 😊😊
@@budimudita5649 Awesome! Please spread the word about my videos to your colleagues and academic community in Indonesia 🇮🇩 for awareness. They'll learn some useful tips and skills too...thanks 😊
Hi, much appreciated content dear but I've a question. how can we estimate two step GMM with moderating variable in stata? I'm stuck at it. nothing is helping me out. looking forward to your response.
@@CrunchEconometrix thank you so much for your respose. i've estimated by including interactions by my results are not significant. my results with pannel regression are significant but with GMM they are inignificant.
Hello Adeleye, Thank you for your support. This videos really help me while I'm doing an essay of my thesis. But I have some answers: 1) Is necessary that the series to be stationary? That is, mean and variance also do not change over time? Its is possible to work with series that are integrated - I(0) and series that I(1)? 2) How much lags should I have to specify in the: gmm(l.DepVar Var1 Var2, lag(? ?) collapse). Best Regards, Alex.
Hi Alexandre, it appears you did not watch the first background video to these GMM tutorials. I suggest you do for clearer understanding about details relating to the technique. For instance, the issue of "integration" does not arise in panel GMM. Thanks.
Hello, once again many thanks for your great tutorials and invaluable assistance. Since I am a beginner with the GMM models, I would like to ask you whether it is correct to include in the gmmstyle an endogenous variable which was excluded from the explanatory variables? In other words, it appears ONLY in the gmmstyle along with the lagged dependent variable. The model works. No errors show up. However, I would like to be 100% sure, that it is correct.
Many thanks for you ultra-swift reply It seems to me that when choosing instruments, it is quiet easy to pick endogenous instruments (e.g. lagged independent variables). Much harder is to use exogenous instruments. Personally, I have a huge problem to choose such an exogenous instrument which would make me 100% sure that it is not correlated with country/region fixed effects. Am I correct in thinking that is much safer to use lagged independent variables as endogenous instruments rather than exogenous instruments? In my small panel (N=16, T=8), I use all 5 independent variables as lagged endogenous instruments. Using different lags and adding/removing lagged exogenous instruments I am trying to “pass” both the AR2 test and Hensen test (
@@zbigniewm4712 I have simplified GMM explanation by estimation different simulations to help researchers. So, follow my guidelines and align your study.
Dear professor, I have a question. Am I right that we don't have to use every exogenous variable in the equation also in the iv parentheses? For example, in the video you don't use inflation as iv.
Thank you, Ngozi. Your posts are helpful. I have a problem with xtabond2 specification. I am thinking I have done all I needed to do and I had "a no observation feedback" What next should I do?
Your video explain me well. However, I have some questions regarding year dummies. Can you explain more about year dummies? Why do you use 1 and 0 instead of 0,1,2,3... for the year dummies? For example year 2006 =0, year 2007=1 etc. if you use 1 and 0 it seems that if it is 0 refer to which year? I see your data is strongly balanced, what about if the data is unbalanced panel and how to put year dummies? Whether put 1 if all the variable exist? Look forward your reply. Thanks
Hi Annur, year dummies in this context are coded 1, 0. For example, 2006 takes 1, other years take 0. 2007 = 1, others = 0. I will also suggest you read more about this. Thanks.
Hi, many thanks for your invaluable tutorials! I would like to ask you whether in the two-step system GMM the coefficient of the lagged dependent variable can be statistically insignificant? Does it have any impact on the interpretation of other statistically significant coefficients?
Hi Dr, I am working on the 7 countries of WAEMU. The study period is 1995-2015. The number of countries is then lower than the period. I would like to know if it is possible to calculate the arithmetic mean of 5 years, 5 years to reduce the study period in order to be able to apply the GMM system method?Can I average every 5 years so that the number of observations equals 5, where T will be less than N?
Hi Boukare, to estimate GMM, N must be sufficiently larger than T. As it stands, your N is very small. But you can try to average the years as suggested to see if you obtain any meaningful results.
I am using Two-Step SysGMM, Real GDP Per capita (lny) is a dependent variable, the independent variable is a dummy (free trade area), control variables are a set of 4-groups: i. Trade (openness, intra-community export, per capita value-added), ii.Human and physical capital variables are 4 (education, gross capital formation, and age dependency) , iii macroeconomic stability (3 proxies), iv. iv. Institutional quality variable. the problem is IV validity, Hansen test persists giving 1 value????. would you advise!!!
@@CrunchEconometrix thank you so much! your video has been extreamly helpful for my thesis. i just want to confirm, if my p-value for AR(1) is 0.2, and everything else is ok, is the model still valid?
Dear Ms. Adeleye, What happens in the xtabond2 if having two endogenous variables? More specifically, I have the function: Yt=F(Yt-1; X1; X2; X3). I see your commands consider Yt-1 as an endogenous variable. But now, suppose X1 is also endogenous variable, how will the command change? Adding X1 in gmm() like: gmm(Yt-1 X , lag(2 .) collapse)? Thank you in advance
@@CrunchEconometrix Hi. Thank you for your response. I did see that part before sending my comment. Exactly I had a misunderstanding about endogenous and predetermined variables. So sorry about that. Now I understand that mod can be an endogenous variable, and how to resolve in this case
Hello brother My problem is also same. I need to lag 3,ahould i add like Yt-3? And anothe suppose another endogenous variable x then, i have to add in gmm(). I need the to know brogher if possible. Thank you
I have a question! I did all the steps and I need to use a difference GMM. However, my p-values for the regressors are all very high... despite my AR(2) and Sargan and Hansen statistics looking very good. What is my issue? I have a lagged variable for the dependent variable so I must use this type of model, but I can’t figure out where I’m going wrong. I hope you have some answer with the limited info I’m giving you.
If it helps, I have three dummy variables in my model. I cannot turn most of my variables into natural logarithms because these continuous variables are growth rates and also take negative values. I have one continuous variable that only has positive values (it’s ‘government revenue as a share of GDP). Would making that a natural log be helpful? I’m such a rookie at this hahaha.
Also, I did not create year dummies because my number of groups is only 10, my T = 8 and anything I tried with year dummies makes my Z > N which shouldn’t be the case. Is that hindering me? Sorry if I’m asking too much
Hello Mam..This is a great video. But I am facing a problem for running a dynamic panel model using the command xtabond2 in stata 11. It shows " unrecognized command: xtabond2 r(199)". Please suggest me how can I recover this. Thank You...
Hi there, it implies you are yet to install the syntax. Do so with either "ssc install xtabond2" OR "findit xtabond2" OR "help xtabond2". Type in the COMMAND BOX w/o the inverted commas. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix I am from India. As per your instructions I type the command. But this message was shown ssc install xtabond2 proxy host not found fmwww.bc.edu/repec/bocode/x/ either 1) is not a valid URL, or 2) could not be contacted, or 3) is not a Stata download site (has no stata.toc file). r(660);
@@CrunchEconometrix I tried all the three options. But all three options not working. showing a messages "help for xtabond2 not found" or proxy host not found www.stata-journal.com/software/sj9-1/ either 1) is not a valid URL, or 2) could not be contacted, or 3) is not a Stata download site (has no stata.toc file).
Dr, I wonder if you could explain how to find significant differences between two coefficients. I run system gmm on two groups or samples. Thank you Dr👌
CrunchEconometrix Hi Dr. Sorry on my earlier message was a bit confusing. What I mean is how to compute the statistical significant differences between two coefficients. For example, the difference between coefficient A and B is statistically significant at 10%, pvalue 0.004.
Hi Ms Adeleye, thank you very much for ur smart videos. I have already shared the link with colleagues and are very happy with them...I am running twostep system GMM. My queries: 1. Is the independent variable MUST be significant or in another word if all or most of control variables are significant but the independent variable is not, is that acceptible?
2. Is it acceptable to report different results obtained from the changing of IVs and compile them in one table? or just report the best one? Because I have ended up having 6 results, in which R2 & Hansen are insignificant, but the independent is significant in ONLY 2 models 3. Is the lagged dependent variable (in case it is significant) included in the long-run GMM coefficient estimates?
Hi Yassin, thanks for watching my videos. Always remember that results are data-specific and cannot be manipulated. If you have insignificant results, you may decide to change your variables with better proxies. I have a video that addresses point 3, kindly watch it. Keep watching...keep sharing! 💕
Hello Professor, Thank you for the tutorial. Thanks to you I am almost done with my analysis. There is just one question that for estimating two step system GMM is it mandatory to first estimate one step System GMM? I have done my analysis without one step system GMM. Will this be okay?
Hi Lufta, there's dofile for 2SDiff GMM. That of 2SSys GMM is VERY obvious from the video which I also clearly explained. Get the 1SSys GMM and you'll know what to do.
Hi Prof Thank you for your kind videos' I'm trying to apply GMM in my study with an equation model as follows: CR = B0+ B1CRit-1+B2Git+B3BAit+B4DUit+B5AIit+B6ADit+e Where: CR, corporate reputation, G gender, BA board activity, DU CEO duality, AI audit committee independent, AD audit committee diligence I applied system GMM using the following command xtabond2 CR l.CR G BA DU AI AD Banking Insurance OFS Y2020 Y2019 Y2018 Y2017 Y2016, gmm (l.CR l.DU l.AI ) iv(G BA DU AI AD, equation(level)) twostep robust orthogonal small the question is, is it possible to use the lagged variable after gmm (e.g X1) and the same variable with the original version after iv (e.g l.x1)? Is this allocation of variables acceptable? Thanks in advance
Hi Barhom, I did something similar in the examples. Endeavor to watch all the series. But if the lagged BA and DU are in gmm( ) it implies they are assumed to be endogenous hence, you cannot include them in iv( ). Thanks
hello Nogzi.thank you for your helpful videos. I have a question. how do I have to sort my dofile dofile and make the matrix? how can I find some information about it? I would be very thankful if you could help me.
1. can I use lag (unified or different lags) in xtabond2 part? 2. in case 1 is true, are the lags must be the same in xtabond2 an IV part? 3. Would u provide me with reference for the formula: calculating the long-run effects for the Kth parameters: βk /(1-Φ) and for the year dummies
To the best of my ability, the GMM video tutorials that I've provided are well-explained. But, if u're still in doubt on what to do, kindly seek more information and clarification on the internet. I have also provided references at the end of each video. Thanks for sharing, keep sharing too! 💕
can i include the independent var (explanatory var as endogenous with the lag of independent var in the gmm part and exclude it from the IV part (exogenous) ?
Hi Ms. Adeleye! I'm doing my dissertation and I have a question about GMM in two steps and one step. Which one is more efficient? Which articles deal with the topic? Thanks!
Hi professor. I have been watching your videos they are really helpful. I have a query. I have a sample of 370 firms with 10 year time period. I divide my sample into two subsamples based on certain criteria. However, while applying system GMM two-step estimator in this case, the stata output doesn't report the Arellano and bond second-order serial autocorrelation. it gives a dot instead. Kindly advise on this.
Hi Yukti, thanks for the positive feedback and remarks on my TH-cam videos. Deeply appreciated! Dot appears if there is insufficient observations, specification error etc. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hi mam, thanks for the response, also kindly emphasise that what should be done in case of insufficient observations or specification error. I am from India
@@yuktibajaj8398 Replace the variable with the missing observations with a close proxy. Model specifications can be corrected by changing the functional form of the model: log-log, log-level, level-level...I'll appreciate it if you can share the link to my TH-cam Channel with your friends and academic community in India 🇮🇳 for awareness. Thanks 😊
@@CrunchEconometrix Thanks a ton mam, I will try doing the same and respond with the results. I always share your channel to my juniors scholars working on GMM.
Hi Mam Great videos indeed. I got a lot to learn. However, when I am trying my data set in two-step System GMM, three variables annual GDP growth rate, Bond market structure (Domestic credit to the private sector as a % of GDP) and stock market structure (market capitalization as a % of GDP) are omitted despite trying so many simulations. They are my independent variables. Kindly advice what to do?
Hi Yukti, thanks for the positive feedback on my videos. Deeply appreciated! 💕 😊. The ommission can be due to: (1) collinearity between the two variables or (2) insufficient data points from both. Stata often gives the reason for the omission. Check the upper part of the output table. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hi mam, I am from New Delhi, India. I have tried but the output is silent about the omission. Is there a way to show you the output
Hello Dr., am working on a cross section data for 16 countries with 16 years period. Is there a need to create year dummies and if yes why and how many year dummies do i generate
@@CrunchEconometrix From Pakistan. I have 38 countries in my sample but GMM considers only 33, Why is this. I have few missing day. Is it ok to report results without robust option, because robust makes the p-values insignificant?
Hi Prof, Your videos are very comprehensive. We really appreciate your dedication and hard Work. Problem: System GMM 1- Can we include regional dummies(option_i.Region)/factor variable (1=Asia,2=Africa,3=America,4= Europe) to capture regional effect in our model. 2-If I split the sample in two groups, the number of instruments got exceeded by number of cross sections, with collapse option, results turned to insignificant. 3- As per your video, it is pre-requisite to generate year dummies before applying system GMM, is it captured policy shock? i.e i have data from 2000-2018, I want to generate a year dummy which is equal to 1 after a particular time (1=2014,2015,2016) & 0 otherwise(2000-2013). Should we include both dummies in our model? Your suggestion is requested to deal with the above-mentioned issues.
Hi Asif, my GMM videos are well-explained. To date, they are best on TH-cam that shows you exactly how to go about GMM estimations. I did not use regional dummies. Follow what I did. If instruments are still more than groups after using the collapse command, reduce the number of regressors and re-estimate the model. My advise: GMM by itself a complicated technique so don't make it excessively complex...keep it simple to get interpretable results. Thanks.
@@CrunchEconometrix Ma'am, actually my hansen is coming out to be significant as well. I don't know, what should I do. Is it acceptable to present my studies with significant Hansen? Thanks in advance for ur reply.
@@ks12253 Nah K, nobody will take your results seriously with a statistically significant Hansen. It evidences that your instruments are not "good". You may need to run several simulations like I did in the videos and also "juggle" the instruments set. Not one straight measure in addressing this problem, but several.
Thanks crunch econometrix your video helps me alot...but here is a problem that my results of my lag dependant variable shows coefficient of lag dependant as omitted😢please help me
@@CrunchEconometrix Thnkxx..one thing in which I am confused that we use L.Dependant(lag of dependant /independent variable in estimation)then is there need to first generate the lag of variable like which we do in eviews or there is no need to generate it in data set?....Stata itself generate it
@@CrunchEconometrix Thank You so much For these videos on GMM which helps me alot...May you get many more success in your life.. my last question is that if I didn't include year dummies in my model but include theses as instrument...is it good or not...because my results are significant when doing in this way
@@CrunchEconometrix Good afternoon Dr. Adeleye Please i was running one system GMM, and from what you taught us during the seminar last year in Ghana. You said if the number of instrument is greater than the number of groups, it means your model is not good. So you have to reject it. I run my analysis with the collapse option and still have the number of instruments greater than the number of observation. How do I correct this? I used three firm level factors as exogenous regressors, no predetermined variable. Thank you. Hope to hear from you soon
@@addokwabena7066 That implies that you have too many explanatory variables. Remember, instruments are CREATED from the explanatory variables. You may need to reduce them. Thanks.
@@CrunchEconometrix grateful. I re-watched your video on understanding GMMamd realized I got it wrong from the beginning.(the number of groups was less than the number of years). Reduced it and it worked. Can you help on how to perform Mann-Whitney test in Stata? I
@@CrunchEconometrix thnkxx....Will you please tell about static vs dynamic panel....and if we have static panel then where T is small than N(no of observation )then still for static panel we use gmm or not
Hi Laraib, static panels do not include the lagged depvar as a regressor while dynamic panels do. Static panels require N > T are limited to use of FE, RE, and pooled OLS. Dynamic panels require N > T (use GMM) or N < T (use panel ARDL, DOLS, FMOLS etc).
Thanks for watching my video, deeply appreciated! Please watch my videos on GMM to understand the differences between 1-step and 2-step, hence, their degrees of freedom won't be the same. Basically, because the former uses more instruments.
@@ulinafg465 Thanks for watching my video, deeply appreciated! The F-test has its manual underlying computation which I don't get involved in because I allow the software to compute it for me.
Hi Waleid, due to abuse and unethical conduct some datasets and all Stata dofiles used in my videos are no longer free. They are available on my website upon payment. Here's the link cruncheconometrix.com.ng/shop. Thanks
I am Gilbert from Tanzania. I am having problems estimating my panel data using GMM. I did follow the different simulations in your videos and I ran one of them and I had my explanatory variables well significant, with expected signs, but only the Hansen p value was 1.I am stuck here. Which direction can you please advice me to take, in order to arrive to results with which I can publish my paper? In addition, I read in Roodman's book (2009) and there is a part on Hansen test that says "The Sargan/Hansen test should not be relied upon too faithfully, because it is prone to weakness", can I use this as an argument to proceed with my results as long as the tests are not good? Thank you alot.
@@CrunchEconometrix well, I am getting a serial correlation in running xtserial. My variables are significant in fixed effects but all are insignificant in GMM. I dont even know what to do:)
@@CrunchEconometrix But robust in two step GMM is used for Windmeijer (2005) finite sample correction to correct for download bais of SE. In literature, no where it is mention that it is to connect heteroscedastic in two step GMM.. Plus it is for finite sample( what does it mean).What is the definition of finite sample.
@@CrunchEconometrix In my case in two-step GMM , when I am applying robust all my coefficients become insignificant ( around 1000 observations).. why so.
Yes, you can if make the assumption that the explvar is either endogenous or predetermined. I did a similar simulation in both the 1-step and 2-step videos. Watch them to understand what I mean. Thanks.
Hi beloved guest/subscriber, you have discovered my amazing TH-cam Channel tailored specifically for you and other beginners and intermediate users. Please do not keep me to yourself (lol). Kindly share my videos and links with your students, colleagues and academic community so that they too can SUBSCRIBE and learn with ease….and for the global community to be aware that applied econometrics can be simplified. My teaching approach is very practical. I adopt a do-as-I-do style. Many thanks to those who have supported me by telling others. Once again, CrunchEconometrix loves to teach, support my Channel with your subscription, your likes, feedbacks and sharing my videos with your cohorts. Follow me on Facebook, Twitter and Reddit. Love you all, greatly!!! :)
CrunchEconometrix thank you so much for the eye opening videos. I posted a question on your blog about an ARDL model and a financial crisis. Hopefully you will be able to attend to at the convenience of your schedule. Once again thank you 🙏🏿
Hi Andrea, my sincere apologies. Time constraints has not allowed me to visit my Blog as I should. So, please always post your queries on the comment section of the respective TH-cam video. I'll respond from there. Thanks.
@@simbarasheandrea My apologies, Simba. Not sure I responded to this. I'll appreciate if you always post your queries on TH-cam.
tThank you for your video. I have a question. My thesis topic is predictability of stock return ( alexakis. et.al. 2010)Use GMM for estimation. I have read from journal. It show that they did unit root before. In this case if it possible to use d.l.x. (Both lag and first difference) when running xtabond 2. Could you please help me?
Thank you very much Dr. for your swift answer. Love from your student (Ethiopia).
Thanks so much, Farid. Much love from Nigeria 🇳🇬🥰
Your videos have helped me incredibly while doing my bachelor's thesis. Thank you sooo much!
Congratulations Mar, so very excited hearing this positive feedback! 💕 😊 Glad to be of help....may I know from where (location) you are reaching me?
Thank you fellow alumna...watching from the University of Sussex
Thanks, Laurent for the encouraging words. Deeply appreciated! Sussex to the world 🌎 🥰🙏
interesting and informative video
Hi Zafir, anks for the positive feedback, Sir. Deeply appreciated! Please may I know from where (location) you are reaching me?
Hi Ma'am, Your videos are of great help ma'am. They have simplified GMM so much.
Ma'am, I am a PhD scholar at IIT Delhi, India in the area of finance and have been using your videos for my PhD. They have proved to be beyond helpful. I am in the last leg of my PhD and am facing some issue in the result.
I am getting no values for AR, Sargan and Hansen test. I will be immensely grateful for the help.
Hi Prashant, this problem when there are TOO many missing observations in the data. Thanks.
@@CrunchEconometrix Thank you for the response ma'am, but all the variables except 3 have 100% data. the 3 variables jave 99% data. Dependent Variable too has 100% data
Prashant, execute the syntax "mdesc" to know the exact percentage of missing observations per variables.
Hi @Prashant, did you fix your problem? I face a similar problem when all my variables have 100% data but Sargan and Hansen's test shows no values.
Dear professor, thanks for your videos. They are very helpful! I already followed your steps and made estimations with 8 different specifications: System GMM and Difference GMM, one and two steps, and ortoghonal and not ortoghonal for my master degree investigtion, because I wanna know if inequality and redistribution affected the colombian economic growth, and I wanned to have robust results. I kindly ask, besides Sargan, Hansen, AR(1) and AR(2), which other tests can i apply to asure that the estimations are free from any other problem. Thank you very much for your attention.
Thanks for the encouraging feedback, Nicolás. Hansen and AR(2) are the important diagnostics.
I am learning alot thanks to you Miss, @ 3:38 you say "AR2 not significant tells you that your model does not suffer from second order serial correlation" you mean it does suffer (no?)
Once the AR(2) > 0.05, the model does NOT suffer from 2nd-order serial correlation. That null hypothesis of 2nd-order serial correlation CANNOT be rejected.
Thank you so much! As I am a Beginner to GMM, I have a basic question: based on what criteria do you select the variables for iv() and gmm()?
Hi Francis, kindly watch the INTRODUCTORY GMM video. I gave some explanations. Thanks
Thank you again!
You are welcome, Farid🥰
Thank you very much for really helpful videos! I have a question. For a panel data estimation with t (11 years) and n (19) I used GMM like you taught us. Besides, what method can I apply in order to support my results? Thanks.
Hi Merve, you can perform static analysis (FE, RE, IV-GMM) for robustness checks.
@@CrunchEconometrix Thank you professor. I really appreciate your help!
great job
Thanks, Romanus for the positive feedback on my video, deeply appreciated! Kindly share my TH-cam Channel link with your friends, students and academic community on social media for awareness. They'll learn some useful tips and skills too...thanks 😊
Dear Dr.Adeleye : Thank you for your wonderful videos. I am learning a lot from you. I estimated 1-step Sys GMM and 2-step sys GMM but some coefficients are significant in 1-step but not significant in 2-step and both have good Hansen and Sargan results. Which should I report in my manuscript?
Hi Martin, I will always go for the GMM that gives me the best results. Do the same.
Great videos mam.Your videos helped me a lot in understanding gmm and with your help, I was able to write a research paper. Thank you very much. I was hoping if you could make some videos on estimating simultaneous equation using system gmm.I.m unable to find the commands for the same on stata.
Hi Naina, I'm encouraged by your commendation and positive feedback. Deeply appreciated! 😀 I'll do my best to make the videos once I fully understand the techniques. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix thank you once again, mam. Lots of love and best wishes from India.
Hello Dr. again! How are you? I hope you 're doing well. I have few questions waiting for your light answers to reflect on. No.1 Is it mandatory to include robust option when running xtabond2 since most of my variables of interest are not significant? No.2 When I include y*, the same thing happen again including y*. So do we expect y* to be significant? Thank you so much!
Hi Farid, as I explained in the video the "robust" option controls for heteroscedasticity. Please watch the clip on "GMM Year Dummies" to understand the relevance. Thanks
Dear Prof. Ngozi Adeleye,
I have watched most of your TH-cam videos on GMM, and they are beneficial videos. I have a question to ask you, and I hope you will provide your expert advice on my query.
I have panel data with N>T case. After estimating the results using the two-step system GMM, should we test for cross-sectional dependency? As I understand, the existing tests proposed by the Breusch and Pagan (1980) LM test, Pesaran (2004) scaled LM test, Pesaran (2004) CD test, and Baltagi et al. (2012) bias-corrected scaled LM test is to be used after Fixed effect/Random effect test. They do not refer to the two-step system GMM case. I highly appreciate your advice in this regard. Thank you very much, prof.
Hi Sri, I gave the background to the GMM technique. You may need to watch it or check other online resources for more information regarding your queries. Thanks.
Thank you ma'am for your wonderful lessons, I am trying to apply the two-step system GMM, using the xtabond2 command, but unfortunately not able to get the result. Instead, I got this error "////// invalid. r(198); could you pls shed some light in this regard.
Hi Farhan, you got the error message due to wrong typing of the code. At the end of /// move to the next line to continue the coding.
I have unbalanced panel data of 218 companies and I taken lagged dependent variable. Also taking Interactive Term, it is the product of financial crisis(dummy variable) and Div.payout( independent variable). At first I run simple OLS and after that I test for heteroscedasticity and then VIF, Correlation for testing multicollinearity. I also test Dicky-fuller augmented test for checking stationarity.After that I run two step system Dynamic Panel Regression with Robust of Arellano-Bover and Blundell -Bond estimation. Now my question is that whether auto correlation is must for panel data anlysis? and also whether Sargen test is compulsory for panel data..
Hi Sour, my advice is that you watch all my GMM videos for better understanding and follow my guide. Don't make it too complicated than it already is.
Dear Professor, I am using two step SYS GMM for my analysis. I have a regression in Stata like this: xtabond2 y l.y x1 x2 x3 where x1 is an endogeneous variable meanwhile x2 x3 are exogeneous. Should it be right this line code in Stata? xtabond2 y l.y x1 x2 x3i.year, gmm( l.y l.x1, collapse) iv( x2 x3 i.year, eq(level) robust ecc... Should I include the lag of my endogenous variable in twice ( ie xtabond2 y l.y x1 l.x1 x2 x3...) or it is okay if I include the lag of endogenous variable ( l.x1) just in gmm()?
Hi Simone, I have responded to you on this.
Thank you for all your videos. They were very useful and helpful. Keep shining. I have a question please: Some references put the option lag () in the xtabond2 command synthax, for example lag (4 7). Could you please explain what's the meaning of this option and when we should use it? Thank you in advance.
I will refer you to go through some of the references listed at the end of the video to fully understand the xtabond2 routine.
Hello Dr. Adeleye, is it necessary to have year dummies if the depvar is not in log form?
Hi Farid, year dummies can be included in regression models regardless of whether the depvar is transformed into natural logarithm or not.
Prof. I would like to know if it is as well possible to run this estimation with the use of EVIEWS ?
I have actually checked EVIEWS but I couldn't find the system GMM option.
Hi Ulrich, not sure if EViews has that menu. But you can post your query on any EViews platform for more constructive feedback.
Hello mam... your doing great job... i tried this method on my data but AR (1) , AR(2), sargan test and hasen test are significant... now what should i do?
I showed in the GMM videos how to correct these. You may want to watch the videos in sequence.
Thank you for the video! It was very helpful. Nevertheless, while listening some questions crossed my mind and I would really appreciate if you can answer:
1) Is Windmeijer correction automatically applied when we use xtabond2 command? And if not what additional command should be used to account for it?
2) If AR(2) test is not significant that means that the model does not suffer from second order serial correlation. But what if AR(1) is also not significant (example: 0,200) does that indicate a problem ?
3) I think in every two-step regression a warning appeared saying "Two-step estimated covariance of moments in singular. Does that also indicate a problem and what does it mean exactly?
4) Related to the 3rd question if the program tells you "The difference in Sargan/Hansen test may be negative is that a problem?
Thank you once again !
Hi Akim, thanks for the positive feedback. Deeply appreciated. I gave some clear explanations in the video, but where in doubt, kindly refer to the references at the end of the video for more information about the GMM technique. Thanks.
Great videos series. it really help me on my dissertation. one question please. do we need to do diagnostic data test such as multicollinearity test, normality test, heteroskedasticity test, outlier test before do two step system gmm? thank you very much
Hi Hamba, not at all. Only correlation analysis is required which is a PRE-ESTIMATION test. Kindly watch my video on MULTICOLLINEARITY on why this is essential in ALL empirical analysis. Thanks.
@@CrunchEconometrix Thank you very much for your response. definitely will go through your video. One more question please.... For two step system gmm, is it necessary to include "robust" option, since when I applied it, many of my variables become not significant? when I am not using the "robust" nearly all my variable is significant and AR(2) and hansen test is good. What is your opinion? thank you very much
Well, the "robust" option controls for heteroscedasticity.
@@CrunchEconometrix all right. Thank you very much. 🙏
Hi, Thanks for your video.
What can be the cause of the problem with the following warning? (condition with my estimation: n of instruments < n of groups)
"Warning: Two-step estimated covariance matrix of moments is singular."
Using a generalized inverse to calculate optimal weighting matrix for two-step estimation.
Hi Xez, thanks for the positive feedback. Deeply appreciated! I'll suggest you don't worry about it. It appears in all Stata GMM output. Concentrate on the results. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hi there and thanks for your answer. I reached you from Germany. Why do you think that using a generalized inverse W Matrix is nothing one should worry about? I read so far different explanations why this could occur, and some are indeed worrying to me.
@@Xez1919 I focus on the GMM results. But if you are still interested in the "warning", kindly post on Statalist.org for more constructive feedback.
Thank you maam for your great lessons, I have 10 indep.variables and 4 control variables out of them six are endogenous. Now should I include the lag of only this six or all as the Instruments? Dear pls reply it's one month my work is stuck at this point.
Farhan, my 9 GMM videos are well explained with clear examples. Adapt what I did to suit your data and variables.
@@CrunchEconometrix dear ma'am can I used all the explanatory variables as Instruments or just the endogenous variables?
Kindly follow the guide shown the 9 GMM videos.
Xtabond2 y l.y x1 x2 x3 x4 gmm (y x1 x2 x3 x4, lag (a b)) noconstant twstep. Dear maam is is a valid command or not?
You can modify the code used in my GMM videos. They are very visible.
Hi Ms Adeleye, many thanks for your swift respone. I am running the Two-Step System GMM. In some specifications, the outcomes for time-invariant variables (e.g. coefficients) are reported as omitted. When I change the specifications (e.g.by removing some variables), the outcomes for the time-invariant variables are normally presented. Why is that?
Hi Mogila, variables are dropped of there's multicollinearity. Stata will report that at the top of the output Table.
Dr Adeleye, your videos have been so useful for me! I wonder if you could help me with two questions.
Is it possible to include many dummy variables in the two-step system GMM? To
be more precise, I want to include a dummy for the year (y), region (h) and
technological intensity (g). Following your example: xtabond2 lngdp l.lngdp
lngfcfgdp lnlabr infl lnintuse lnmob lnfts y* h* g*, gmm(l.lngdp lnmob,
collapse) iv(lngfcfgdp lnlabr lnintuse lnfts y* h* g*) twostep
robust orthogonal small
Hi Elizabeth, thanks for the positive feedback. Deeply appreciated! I suggest you use the dummies sparingly. That is, one at a time to avoid your model breaking down due to insufficient instruments. Please may I know from where (location) you are reaching me?
Hello Professor, how do we export the test Hansen to a word doce? When I use asdoc I only get the table standard part with the coefficients. Thank you
Vicente, depending on the export command not all the statistics in the table are "exported". You will have to write out some.
Dr.
I take lag 3 for my dv in two step system gmm.
Iif i take that then i can make the AR1 AND AR2 INSIGNIFICANT ONLY.
IS IT OK IF I TAKE LAG 3 OF DV?
HOW TO JUSTIFY THIS DR?
PLEASE, I NEED TO KNOW DR.
IF AR1 is significant then is it ok to report on thesis?then what is the justification?
Thank you dr.
GMM syntax is easily manipulated. So, you can try your simulations to see the outcomes.
Hi Ms. Adeleye, thnx very much these very useful tutorial videos. if I include limited dummies years as instruments ONLY (NOT in xtabond2 part), shall I put in the output GMM table "time dummies" YES or NO?
@UCK9hD254JKbCZ4Bf8Iz1s7g Hi dear Adeleye. Firstly thnx very much for your prompt response. Secondly I do apologies if I posted my question in inappropriate way (I am not an English speaker and beginner in GMM). I was trying to come direct to my question. However, let me re-posted as follow: I am running system GMM, my explanatory variable is dummy (xtabond2 lny L1.lny FTA ORECs lnOP lnICEXT lnPCDVA SEPG DCPSGD PSAV , gmm(lny,lag(2 2) collapse)iv(L2.lny l1.FTA OREC
s lnOP lnICEXT lnPCDVA DCPSGD y9-y13, equation (level) ) nodiffsargan twostep robust orthogonal small. While I was shuffling variables I found the best model result as when I use dummy years as instruments , that I do by "ONLY" to best of my knowledge. if is that true, shall I put yes or no for time dummies in output table?...hopefully I could manage to make it clear.
Apologies accepted Yassin, based on refs given and my experience, that will be NO because year dummies were not included in your analysis.
@@CrunchEconometrix Thank you very, dear. right now I came to know about "Yes" or "No" in the output table. nice day
@@CrunchEconometrix Another question, is that "No" accepted in system GMM result?
@@yassinyahia2453 Well, a reviewer will want to know why you did not control for time variations. Please watch my video on that (if you haven't) to know the essence of time dummies in GMM procedure.
Hello Ngozi, Thanks for your commitment and effort.
I have a question, in my 2 step sys GMM result, my first GMM regression AR(1) =0.380 while AR(2) =0.775. AR(1) is obviously out of place. how can i reduce my AR(1), or can i ignore my AR(1) since AR(2) is more important(as you said in one of your videos). The hansen test is 0.807 and sargan test is 1.000, how can i reduce the sargan result or can i ignore sargan result and use just hansen??
In my 2nd regression, the AR(1) and AR(2) was good, but none of the variable was significant. What could be the reason??? when i try to change the variable and lag, I got omitted result(blank) for all the variables, can i accept this result??.
best wishes
In Addition, what's the criteria for picking the internal and external instrument?? or is it by random?
Hi Okolo, I addressed most of your queries. My advice is that you watch the 9 GMM series in sequence for better understanding of the technique.
Instruments in iv( ) are determined by several factors: theory, related studies, and/or intuition.
@@CrunchEconometrix Thank you
Dr Adeleve thank you for this video. I have some questions I would be grateful to answer me. In my model I have 1 IV and 6 moderators and 6 control variables. So can I I apply all moderators in one commands like (xtabond2 dv l.dv IV IV*modert1 IV*modrat2.... IV*modrat6 control variables y*, gmm(l.DV) iv(IV IV*modrat1.... IV*modrat6 control variables y*, equation(level))
Caacay, I am very certain that you won't get any meaningful results due to the many regressors. You will have little or degrees of freedom left. Trim down your regressors and follow the steps shown in my video to engage the analysis.
@@CrunchEconometrix thank you dr for replying. can i get your email! i have a copule of questions. thank you again.
Please post them here and keep them very brief. Thanks.
@@CrunchEconometrix okay dr. my question is, depending on what we select the iv (instrument variables)?
Regarding your replay "trim down your regressors" how i do that? is it by deleting some of the control variables or by applying only one moderator then the other ...etc.
thank you dr
Trim down is clear. Reduce.
hi dear doctor，just wanted to ask if the coefficient of lagged dependent variable（instrumental variable）is not significant，shall we continue the test，or the model its not valid？thanks
Hi Qing, you can continue. Model is valid if it passes the diagnostics.
@@CrunchEconometrix dear doctor，thanks for your fast reply，sorry to ask one or two more questions，
1. usually i do correlation and coefficient test plus VIF test for the variables，and then i start GMM test，is there any other test before executing GMM test？
2. as you said，even the coefficient of lagged depended variable（IV）is not significant，so model is still valid? then，another question is，after running GMM test，I do run Sargan test and Abond test, but in many cases, even I go for two-step and VCE test, though AR(2) is more than 5%,( so we accept the results?), but AR(1) sometimes still less than 5%, so in this case shall I accept the VCE results or should go for other test?
Thank you for your kind answer, I am really eager to capture GMM model especially for longitudinal research work,thank you so much dear doctor.
Qing, my GMM videos are explicit on what to do.
@@CrunchEconometrix thank you very much，will learn again
May I ask what the different between 1step and 2step gmm? When should I use 1-step, and when another?
Hi Krysten, kindly watch the foundation GMM video. Well explained, thanks.
Dear Ngozi ADELEYE hope you are doing? how to fix ivstyle invalid issue in stata while estimating GMM estimation. after understanding your model Explanatory, control, and endogenous variable. I have put exactly mine model in the same sequence for estimation. Please guide me.
Tahirah, I'm not familiar with the error. But you can click on the Stata error code and follow the prompt on identifying the problem which will guide you on fixing it.
Dear Adeleye, how can I have access to do file that you mentioned at 2:13 in the video? thanks
Hi Seyede, dofiles are no longer free but available on my website upon payment of a token fee after which you are allowed a one-time download. Kindly find the link cruncheconometrix.com.ng/shop/
good morning Ma, after creating the year dummies, I can see about seven years, how did you now generate yr*. i cant access your do-file
Hi there, y* is the short form of including all the year dummies into the analysis. Also, due to abuse and unethical conduct Stata dofiles used in my videos are no longer free but available on my website upon payment. Here's the link cruncheconometrix.com/view/datashop.php
Dear Dr. Adeleye, I have a question regarding the panel data I'm using. In my panel, N
GMM is not feasible, though there's time series GMM that allows for N < T, you may have to read up on that.
@@CrunchEconometrix I understand. Thank you!!
Very informative video ma'am.do we include lagged values of independent variables as instruments or other variables outside the model not taken initially?
Anuradha, yes you can include them.
Dear Ngozi ADELEYE hope you are doing? I am really inspired by your work. I am using System GMM for my Ph.D. dissertation but most of the variables have negative values even dependent variables also containing most of the negative values. I am unable to take the log of dependent variables when I take the log this made the panel unbalanced and the sample size will also be reduced? how do I fix this issue please guide me?
Hi Tahirah, you can change the variables with negative values to their closer proxies or use them as they are.
Hello, your video is very helpful but i do not understand the criteria to choose the instruments
Hi Vincente, thanks for the encouraging feedback. I explained "instruments" in the 1st GMM. You may want to watch it.
Good morning Professor, your videos are really helpful. I am following all the steps of the tutorial .But the problem is my Hansen Test is .341. It is the lowest among all the steps. Now what should I do? Will my analysis be correct if it is .341?
Thanks in advance. Have a nice day.
Hi Barnona, the Hansen is good. Proceed.
@@CrunchEconometrix Thank you so much. ❤
i have two specific questions.
I took data for 23 Pakistani banks for 12 quarters and using system GMM in this research using STATA. I am unable to understand what is meant by the term ‘instrument’ here since I just used two endogeneity variables but system is showing number of instruments to be 65.
In my initial result, number of instruments is greater than number of groups; Would system GMM be affected by this? does 'collapse' is the only option to reduce it?
I need help in this regard.
i am using xtdpdsys command at stata for my research
Zaira, my 9 GMM videos are detailed enough to make you wrap up your work. I advise you watch them in sequential order. Thanks.
I used xtabond2.
i saw your GMM video and they are so important for me but my question is can i use a two step system GMM for 8 countries and a period of 15 years?
Thanks, Dawit for the encouraging feedback, deeply appreciated ❤️. But NO.
GMM is suited only for N>T panel structure. I explained this in the introductory video.
Mam, When I run the xtabond2 commands in stata as a panel data analysis for the period of 2005-2018. then result omit two years due to collinearity problem i-e 2005 and 2015, I know that 2005 omit due to base year but 2015 omit due to collinearity how to solve this problem.
Hi Majid, I will tell you not to worry about this. Stata will usually drop several year dummies due to collinearity. What is important is that is you obtain good results and your models passes the mandatory diagnostics.
Thanks a lot Mam but How we will justify it in my paper
This is not an issue, Majid. I have told you what is important.
I applied two step system gmm, but most of my variables are not significant, while Sargan and Hansent tests hold a p-value of more than 0.05 and there is no serial correlation in the second order. Dear ma'am pls guide me in this regard.
Hi Farhan, this is not a problem. You can change your variables OR try all the 4 GMM variants: one-step/2step diff/system to know which one gives you the best results. All GMM variants are relevant to the literature. I have used the difference GMM in recent publications due to my data not being robust for system GMM analysis.
Good morning prof.
Could you please (if possible) throw a bit of lights on this problem I encountered with system GMM?
I followed the video on Difference and System GMM to do an analysis. For this work-balanced panel, I used the system GMM, however, the results have been very discouraging. In your videos, you emphasized on the collapse command which I have used, orthogonal, Robust as well, but the results are still not good. AR1 & AR2 both insignificant , Sargon and Hansen also are both insignificant. All my variables are insignificant as well.
I am wondering what the problem could be.
Hi Ateh, from my videos you would have known that insignificant diagnostics are good. But your model failed the coefficients test. The silent rule is that at least 50% of coefficients must be statistically significant. Change your regressors, run several simulations ane check your outcomes.
Thank you for the videos, they're very clear and precise; I'm trying to use system GMM with my dataset but I'm not sure about its adequacy, because it has 17 periods but 11 variables. In your opinion could the results be unreliable? Should I use other techniques like PMG or similar mentioned in one of your videos? Thank you so much :)
Hi Ciak, thanks for the feedback... deeply appreciated. Make sure you watch the foundational GMM it's features and determine if your data fits the technique. Kindly do same with the panel ARDL foundational video. Thanks.
@@CrunchEconometrix I've watched your videos about the adequacy of SysGMM with my dataset and I know that I can use this technique. Just not sure about the meaning of Hansen excluding group and Difference tests for GMM instruments levels.
Is it possible to report wald test statistics and t test statistics together
Hi Collin, yes you can.
Thank you so much Ms. Adeleye... You really help me. I'm a beginner in stata and also on statistics. Do you have a tutorial or tips of how to plot margin effects after gmm. I tried to use margin and marginplot but stata shows "warning: cannot perform check for estimable functions"
U're welcome, Budi😄...I'm glad to be of help but I have no idea about plotting margin effect. Why not post the query on Statalist.org I'm sure you will get answers on what to do. May I know from where (location) you are reaching me?
@@CrunchEconometrix Oke Ms, I'll try it. I'm from Bali province, Indonesia. I am an undergraduate university student, I've seen many papers use GMM thus I searched about it and found your videos. I've watched all your GMM video series, those are really helpful 😊😊
@@budimudita5649 Awesome! Please spread the word about my videos to your colleagues and academic community in Indonesia 🇮🇩 for awareness. They'll learn some useful tips and skills too...thanks 😊
Hi, much appreciated content dear but I've a question. how can we estimate two step GMM with moderating variable in stata? I'm stuck at it. nothing is helping me out. looking forward to your response.
Hi Naila, same way you include interaction terms in static models. Simply include the interaction and estimate. That's all.
@@CrunchEconometrix
thank you so much for your respose. i've estimated by including interactions by my results are not significant.
my results with pannel regression are significant but with GMM they are inignificant.
Sometimes different techniques produce different outcomes. Nothing to worry about.
Thank you!!!always!
U're welcome, girl! 💕 😊
Hello Adeleye,
Thank you for your support. This videos really help me while I'm doing an essay of my thesis.
But I have some answers:
1) Is necessary that the series to be stationary? That is, mean and variance also do not change over time? Its is possible to work with series that are integrated - I(0) and series that I(1)?
2) How much lags should I have to specify in the: gmm(l.DepVar Var1 Var2, lag(? ?) collapse).
Best Regards,
Alex.
Hi Alexandre, it appears you did not watch the first background video to these GMM tutorials. I suggest you do for clearer understanding about details relating to the technique. For instance, the issue of "integration" does not arise in panel GMM. Thanks.
@@CrunchEconometrix Thanks! I'm going to re-watch the videos.
Hello, once again many thanks for your great tutorials and invaluable assistance. Since I am a beginner with the GMM models, I would like to ask you whether it is correct to include in the gmmstyle an endogenous variable which was excluded from the explanatory variables? In other words, it appears ONLY in the gmmstyle along with the lagged dependent variable. The model works. No errors show up. However, I would like to be 100% sure, that it is correct.
Oh sure, Zbig, you can do that based on the assumptions made on the variable. I did a simulation on that in the series. Take a look to see what I did.
Many thanks for you ultra-swift reply It seems to me that when choosing instruments, it is quiet easy to pick endogenous instruments (e.g. lagged independent variables). Much harder is to use exogenous instruments. Personally, I have a huge problem to choose such an exogenous instrument which would make me 100% sure that it is not correlated with country/region fixed effects. Am I correct in thinking that is much safer to use lagged independent variables as endogenous instruments rather than exogenous instruments? In my small panel (N=16, T=8), I use all 5 independent variables as lagged endogenous instruments. Using different lags and adding/removing lagged exogenous instruments I am trying to “pass” both the AR2 test and Hensen test (
@@zbigniewm4712 I have simplified GMM explanation by estimation different simulations to help researchers. So, follow my guidelines and align your study.
Dear professor, I have a question. Am I right that we don't have to use every exogenous variable in the equation also in the iv parentheses? For example, in the video you don't use inflation as iv.
Very correct, Merve. You don't have to.
@@CrunchEconometrix thank you very much. Your videos are very helpfull.
hi,
I am planning to use a panel VAR approach under system GMM (two system GMM). can I get the do-file of panel VAR approach under system GMM?
Thanks
I have no videos on panel VAR. Thanks.
Thank you, Ngozi. Your posts are helpful. I have a problem with xtabond2 specification. I am thinking I have done all I needed to do and I had "a no observation feedback" What next should I do?
That is an indication that one of the variables has no observations left after the underlying algorithm (instruments) have taken place.
@@CrunchEconometrix Thanks, Ngozi. Is it possible which one?
Is it possible to know which one?
@@oroufuooro2770 Check through your dataset.
@@oroufuooro2770 Check through each variable in your dataset.
Your video explain me well. However, I have some questions regarding year dummies. Can you explain more about year dummies? Why do you use 1 and 0 instead of 0,1,2,3... for the year dummies? For example year 2006 =0, year 2007=1 etc. if you use 1 and 0 it seems that if it is 0 refer to which year? I see your data is strongly balanced, what about if the data is unbalanced panel and how to put year dummies? Whether put 1 if all the variable exist? Look forward your reply. Thanks
Hi Annur, year dummies in this context are coded 1, 0. For example, 2006 takes 1, other years take 0. 2007 = 1, others = 0. I will also suggest you read more about this. Thanks.
@@CrunchEconometrix do you have a reference for this? I appreciate your help
You can search online for that.
Hi, many thanks for your invaluable tutorials! I would like to ask you whether in the two-step system GMM the coefficient of the lagged dependent variable can be statistically insignificant? Does it have any impact on the interpretation of other statistically significant coefficients?
Hi Zbig, it's a rare occurrence to have it to be insignificant. But if that is the case, it does not have any impact on other coefficients.
Hi Dr, I am working on the 7 countries of WAEMU. The study period is 1995-2015. The number of countries is then lower than the period. I would like to know if it is possible to calculate the arithmetic mean of 5 years, 5 years to reduce the study period in order to be able to apply the GMM system method?Can I average every 5 years so that the number of observations equals 5, where T will be less than N?
Hi Boukare, to estimate GMM, N must be sufficiently larger than T. As it stands, your N is very small. But you can try to average the years as suggested to see if you obtain any meaningful results.
I am using Two-Step SysGMM, Real GDP Per capita (lny) is a dependent variable, the independent variable is a dummy (free trade area), control variables are a set of 4-groups: i. Trade (openness, intra-community export, per capita value-added), ii.Human and physical capital variables are 4 (education, gross capital formation, and age dependency) , iii macroeconomic stability (3 proxies), iv. iv. Institutional quality variable. the problem is IV validity, Hansen test persists giving 1 value????. would you advise!!!
You have too many control variables.
i was wondering would my result be invalid if i fail to reject AR1 test ?
Not at all, Zaid. AR(2) is more crucial. You should fall to reject that. AR (1) can be statistically significant but not AR(2).
@@CrunchEconometrix thank you so much! your video has been extreamly helpful for my thesis. i just want to confirm, if my p-value for AR(1) is 0.2, and everything else is ok, is the model still valid?
Zaid, I gave clear interpretation of GMM results. Adapt the interpretation to suit your analysis.
Dear Ms. Adeleye,
What happens in the xtabond2 if having two endogenous variables?
More specifically, I have the function: Yt=F(Yt-1; X1; X2; X3). I see your commands consider Yt-1 as an endogenous variable. But now, suppose X1 is also endogenous variable, how will the command change? Adding X1 in gmm() like: gmm(Yt-1 X , lag(2 .) collapse)?
Thank you in advance
Hi Dao, that is correct. If you had watched the entire GMM series you would have seen where I did that with the "mob" variable. Thanks.
@@CrunchEconometrix Hi. Thank you for your response. I did see that part before sending my comment. Exactly I had a misunderstanding about endogenous and predetermined variables. So sorry about that. Now I understand that mod can be an endogenous variable, and how to resolve in this case
Hello brother
My problem is also same.
I need to lag 3,ahould i add like Yt-3?
And anothe suppose another endogenous variable x then, i have to add in gmm().
I need the to know brogher if possible.
Thank you
I have a question! I did all the steps and I need to use a difference GMM. However, my p-values for the regressors are all very high... despite my AR(2) and Sargan and Hansen statistics looking very good. What is my issue? I have a lagged variable for the dependent variable so I must use this type of model, but I can’t figure out where I’m going wrong. I hope you have some answer with the limited info I’m giving you.
If it helps, I have three dummy variables in my model. I cannot turn most of my variables into natural logarithms because these continuous variables are growth rates and also take negative values. I have one continuous variable that only has positive values (it’s ‘government revenue as a share of GDP). Would making that a natural log be helpful? I’m such a rookie at this hahaha.
Also, I did not create year dummies because my number of groups is only 10, my T = 8 and anything I tried with year dummies makes my Z > N which shouldn’t be the case. Is that hindering me?
Sorry if I’m asking too much
Insignificant coefficient is a function of multicollinearity. Watch my video on MULTICOLLINEARITY. It will guide you on what to do.
Hello Mam..This is a great video. But I am facing a problem for running a dynamic panel model using the command xtabond2 in stata 11. It shows " unrecognized command: xtabond2
r(199)". Please suggest me how can I recover this. Thank You...
Hi there, it implies you are yet to install the syntax. Do so with either "ssc install xtabond2" OR "findit xtabond2" OR "help xtabond2". Type in the COMMAND BOX w/o the inverted commas. Please may I know from where (location) you are reaching me?
@@CrunchEconometrix I am from India. As per your instructions I type the command. But this message was shown ssc install xtabond2
proxy host not found
fmwww.bc.edu/repec/bocode/x/ either
1) is not a valid URL, or
2) could not be contacted, or
3) is not a Stata download site (has no stata.toc file).
r(660);
I gave you 3 options not 1.
@@CrunchEconometrix I tried all the three options. But all three options not working. showing a messages "help for xtabond2 not found"
or proxy host not found
www.stata-journal.com/software/sj9-1/ either
1) is not a valid URL, or
2) could not be contacted, or
3) is not a Stata download site (has no stata.toc file).
Dr, I wonder if you could explain how to find significant differences between two coefficients. I run system gmm on two groups or samples. Thank you Dr👌
Hi Khairul, check the statistical significance between the 2 coefficients of the same variable and the signs if aligned with expectations.
CrunchEconometrix Hi Dr. Sorry on my earlier message was a bit confusing. What I mean is how to compute the statistical significant differences between two coefficients. For example, the difference between coefficient A and B is statistically significant at 10%, pvalue 0.004.
I have no idea what you mean, Khairul. Besides a pvalue of 0.004 equates to 1% significance.
Hi Ms Adeleye, thank you very much for ur smart videos. I have already shared the link with colleagues and are very happy with them...I am running twostep system GMM. My queries:
1. Is the independent variable MUST be significant or in another word if all or most of control variables are significant but the independent variable is not, is that acceptible?
2. Is it acceptable to report different results obtained from the changing of IVs and compile them in one table? or just report the best one? Because I have ended up having 6 results, in which R2 & Hansen are insignificant, but the independent is significant in ONLY 2 models
3. Is the lagged dependent variable (in case it is significant) included in the long-run GMM coefficient estimates?
Hi Yassin, thanks for watching my videos. Always remember that results are data-specific and cannot be manipulated. If you have insignificant results, you may decide to change your variables with better proxies. I have a video that addresses point 3, kindly watch it. Keep watching...keep sharing! 💕
@@CrunchEconometrix Thank you very much for your prompt response and sincere help
GREAT , THXXXXXXXX
Thanks, Abu for the encouraging words! 🙏🥰
Hello Professor, Thank you for the tutorial. Thanks to you I am almost done with my analysis. There is just one question that for estimating two step system GMM is it mandatory to first estimate one step System GMM? I have done my analysis without one step system GMM. Will this be okay?
You can do either. Or both for robustness check.
Hello Group, I did not find two-step Diff/system GMM stata do file from crunch econometrics website. Would you mind to suggest me how can I get it?
Hi Lufta, there's dofile for 2SDiff GMM. That of 2SSys GMM is VERY obvious from the video which I also clearly explained. Get the 1SSys GMM and you'll know what to do.
Hi Prof
Thank you for your kind videos'
I'm trying to apply GMM in my study with an equation model as follows:
CR = B0+ B1CRit-1+B2Git+B3BAit+B4DUit+B5AIit+B6ADit+e
Where: CR, corporate reputation, G gender, BA board activity, DU CEO duality, AI audit committee independent, AD audit committee diligence
I applied system GMM using the following command
xtabond2 CR l.CR G BA DU AI AD Banking Insurance OFS Y2020 Y2019 Y2018 Y2017 Y2016, gmm (l.CR l.DU l.AI ) iv(G BA DU AI AD, equation(level)) twostep robust orthogonal small
the question is, is it possible to use the lagged variable after gmm (e.g X1) and the same variable with the original version after iv (e.g l.x1)?
Is this allocation of variables acceptable?
Thanks in advance
Hi Barhom, I did something similar in the examples. Endeavor to watch all the series. But if the lagged BA and DU are in gmm( ) it implies they are assumed to be endogenous hence, you cannot include them in iv( ). Thanks
@@CrunchEconometrix Thank you prof
hello Nogzi.thank you for your helpful videos. I have a question. how do I have to sort my dofile dofile and make the matrix? how can I find some information about it? I would be very thankful if you could help me.
No idea, Atefe. You may need to check other online resources. Thanks.
1. can I use lag (unified or different lags) in xtabond2 part?
2. in case 1 is true, are the lags must be the same in xtabond2 an IV part?
3. Would u provide me with reference for the formula: calculating the long-run effects for the Kth parameters: βk /(1-Φ) and for the year dummies
To the best of my ability, the GMM video tutorials that I've provided are well-explained. But, if u're still in doubt on what to do, kindly seek more information and clarification on the internet. I have also provided references at the end of each video. Thanks for sharing, keep sharing too! 💕
@@CrunchEconometrix Thank you very much for patience and response
can i include the independent var (explanatory var as endogenous with the lag of independent var in the gmm part and exclude it from the IV part (exogenous) ?
I've responded to this.
Hi Ms. Adeleye! I'm doing my dissertation and I have a question about GMM in two steps and one step. Which one is more efficient? Which articles deal with the topic? Thanks!
Hi Marcos, watch these videos in the sequential order as advised DON'T skip any and you have your answers. Jot some notes while doing so...thanks.
Ok , thanks!
Hi researcher.In dynamic panel data case.if T is greater than N .can we apply diff/sys Gmm in these type of modal
No Hamza. Watch the 1st video in this series and my video on "Basics of Panel ARDL" and you get a clearer understanding.
Your video is very helpful thank you. Is ₦995.00 equals to 2.61 U$ dollars?
Can't say really. It depends on the cross conversion at the time of purchase.
Hi professor. I have been watching your videos they are really helpful. I have a query. I have a sample of 370 firms with 10 year time period. I divide my sample into two subsamples based on certain criteria. However, while applying system GMM two-step estimator in this case, the stata output doesn't report the Arellano and bond second-order serial autocorrelation. it gives a dot instead. Kindly advise on this.
Hi Yukti, thanks for the positive feedback and remarks on my TH-cam videos. Deeply appreciated! Dot appears if there is insufficient observations, specification error etc. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hi mam, thanks for the response, also kindly emphasise that what should be done in case of insufficient observations or specification error.
I am from India
@@yuktibajaj8398 Replace the variable with the missing observations with a close proxy. Model specifications can be corrected by changing the functional form of the model: log-log, log-level, level-level...I'll appreciate it if you can share the link to my TH-cam Channel with your friends and academic community in India 🇮🇳 for awareness. Thanks 😊
@@CrunchEconometrix Thanks a ton mam, I will try doing the same and respond with the results. I always share your channel to my juniors scholars working on GMM.
how to use system gmm if dependent variable has negative values?
Anuradha, you can use GMM.
Hi Mam
Great videos indeed. I got a lot to learn. However, when I am trying my data set in two-step System GMM, three variables annual GDP growth rate, Bond market structure (Domestic credit to the private sector as a % of GDP) and stock market structure (market capitalization as a % of GDP) are omitted despite trying so many simulations. They are my independent variables. Kindly advice what to do?
Hi Yukti, thanks for the positive feedback on my videos. Deeply appreciated! 💕 😊. The ommission can be due to: (1) collinearity between the two variables or (2) insufficient data points from both. Stata often gives the reason for the omission. Check the upper part of the output table. May I know from where (location) you are reaching me?
@@CrunchEconometrix Hi mam, I am from New Delhi, India.
I have tried but the output is silent about the omission. Is there a way to show you the output
Hello Dr., am working on a cross section data for 16 countries with 16 years period. Is there a need to create year dummies and if yes why and how many year dummies do i generate
Hi Nana, kindly watch the video on GMM year dummies. Thanks.
I am running System GMM, using xtabond2 command. Can Second order auto-correlation (Arellano and Bond) in two-step System GMM be negative?
Hi Ambreen, be concerned with its p-value which cannot be negative. May I know from where (location) you are reaching me?
@@CrunchEconometrix From Pakistan. I have 38 countries in my sample but GMM considers only 33, Why is this. I have few missing day. Is it ok to report results without robust option, because robust makes the p-values insignificant?
@@ambreensultan5781 Robust options also controls for heteroscedaticity...and which pvalue are you referring to in this case?
Hi Prof, Your videos are very comprehensive. We really appreciate your dedication and hard Work.
Problem: System GMM
1- Can we include regional dummies(option_i.Region)/factor variable (1=Asia,2=Africa,3=America,4= Europe) to capture regional effect in our model.
2-If I split the sample in two groups, the number of instruments got exceeded by number of cross sections, with collapse option, results turned to insignificant.
3- As per your video, it is pre-requisite to generate year dummies before applying system GMM, is it captured policy shock? i.e i have data from 2000-2018, I want to generate a year dummy which is equal to 1 after a particular time (1=2014,2015,2016) & 0 otherwise(2000-2013). Should we include both dummies in our model?
Your suggestion is requested to deal with the above-mentioned issues.
Hi Asif, my GMM videos are well-explained. To date, they are best on TH-cam that shows you exactly how to go about GMM estimations. I did not use regional dummies. Follow what I did. If instruments are still more than groups after using the collapse command, reduce the number of regressors and re-estimate the model. My advise: GMM by itself a complicated technique so don't make it excessively complex...keep it simple to get interpretable results. Thanks.
@@CrunchEconometrix Thank You for a prompt response. I agreed that your videos are the best online source to learn and follow.
@@asifrazzaq2883 Thanks Asif. Don't use regional dummies but time dummies. You can use regional dummies for pooled OLS.
Thanks
U're welcome, Asad.
Hello Ma'am, I'm doing system GMM and my sargan test is coming significant. What should I do? Thanks in advance& your videos are great.
Hi K, as explained in my GMM videos, worry more about the Hansen statistic not Sargan.
@@CrunchEconometrix Ma'am, actually my hansen is coming out to be significant as well. I don't know, what should I do. Is it acceptable to present my studies with significant Hansen? Thanks in advance for ur reply.
@@ks12253 Nah K, nobody will take your results seriously with a statistically significant Hansen. It evidences that your instruments are not "good". You may need to run several simulations like I did in the videos and also "juggle" the instruments set. Not one straight measure in addressing this problem, but several.
Thanks crunch econometrix your video helps me alot...but here is a problem that my results of my lag dependant variable shows coefficient of lag dependant as omitted😢please help me
Hi Laraib, re-estimate the model by playing around with lags and instruments set.
@@CrunchEconometrix Thnkxx..one thing in which I am confused that we use L.Dependant(lag of dependant /independent variable in estimation)then is there need to first generate the lag of variable like which we do in eviews or there is no need to generate it in data set?....Stata itself generate it
@@CrunchEconometrix Thank You so much For these videos on GMM which helps me alot...May you get many more success in your life.. my last question is that if I didn't include year dummies in my model but include theses as instrument...is it good or not...because my results are significant when doing in this way
Yes, Stata does that.
Amen 🙏 to your prayers, Laraib. Yes, go ahead if that approach gives you a better result.
Need do file can you share your website
Hi Miral, here is the link cruncheconometrix.com.ng/shop
2nd thing is that after lot of tries I could not install xtbond2 package in stata 14....is there some issue
Hi Laraib, there should not be any challenge. Post this on Statalist.org for more constructive feedback.
@@CrunchEconometrix Good afternoon Dr. Adeleye
Please i was running one system GMM, and from what you taught us during the seminar last year in Ghana. You said if the number of instrument is greater than the number of groups, it means your model is not good. So you have to reject it. I run my analysis with the collapse option and still have the number of instruments greater than the number of observation. How do I correct this? I used three firm level factors as exogenous regressors, no predetermined variable. Thank you. Hope to hear from you soon
@@addokwabena7066 That implies that you have too many explanatory variables. Remember, instruments are CREATED from the explanatory variables. You may need to reduce them. Thanks.
@@CrunchEconometrix grateful. I re-watched your video on understanding GMMamd realized I got it wrong from the beginning.(the number of groups was less than the number of years). Reduced it and it worked. Can you help on how to perform Mann-Whitney test in Stata? I
Could I ask for do file, please? thanks
Hi Mahesa, due to abuse dofiles are no longer free but available upon payment of a token. Here's the link cruncheconometrix.com.ng/shop
Furthermore, in my model, the coefficients are significant, but the Hansen test does not give any result: Prob > chi2 = .
Hi Elizabeth, instruments problem. Reduce the dummies in the model and re-estimate.
Is it necessary to generate year dummies to run system gmm in stata?please tell me I'm so much confused
Hi Laraib, year dummies are essential to panel data analysis. I emphasized that in my video on GMM year dummies. You can also read more about it.
@@CrunchEconometrix thnkxx....Will you please tell about static vs dynamic panel....and if we have static panel then where T is small than N(no of observation )then still for static panel we use gmm or not
Hi Laraib, static panels do not include the lagged depvar as a regressor while dynamic panels do. Static panels require N > T are limited to use of FE, RE, and pooled OLS. Dynamic panels require N > T (use GMM) or N < T (use panel ARDL, DOLS, FMOLS etc).
@@CrunchEconometrix Thank you So Much😍
can i ask you why the F test in one step gmm and two step gmm have different degree of freedom?thank you
and can you help me to understand how we can get the f statistics?thank you so much
Thanks for watching my video, deeply appreciated! Please watch my videos on GMM to understand the differences between 1-step and 2-step, hence, their degrees of freedom won't be the same. Basically, because the former uses more instruments.
@@ulinafg465 Thanks for watching my video, deeply appreciated! The F-test has its manual underlying computation which I don't get involved in because I allow the software to compute it for me.
ok..thank you very much for your reply, really appreciate it..
@@ulinafg465 U're welcome dear...kindly share my videos with your students and academic colleagues. Gracias!
Could you please give use all the command in word file
Hi Waleid, due to abuse and unethical conduct some datasets and all Stata dofiles used in my videos are no longer free. They are available on my website upon payment. Here's the link cruncheconometrix.com.ng/shop. Thanks
I am Gilbert from Tanzania. I am having problems estimating my panel data using GMM. I did follow the different simulations in your videos and I ran one of them and I had my explanatory variables well significant, with expected signs, but only the Hansen p value was 1.I am stuck here. Which direction can you please advice me to take, in order to arrive to results with which I can publish my paper?
In addition, I read in Roodman's book (2009) and there is a part on Hansen test that says "The Sargan/Hansen test should not be relied upon too faithfully, because it is prone
to weakness", can I use this as an argument to proceed with my results as long as the tests are not good?
Thank you alot.
Hi Adon, we have resolved this via email.
Dear Professor,
Can we still use GMM if we have a cross-sectional dependence? If not, which method should we use?
Hi Mex, GMM can take care of cross-sectional dependence aside the fact that CSD affects panel will time dimensions above 20 years.
@@CrunchEconometrix well, I am getting a serial correlation in running xtserial. My variables are significant in fixed effects but all are insignificant in GMM. I dont even know what to do:)
@@mexzarkashiev2435 Yeah, I can relate to this. You'll have to keep performing several simulations till you get the right model.
two step is more efficient to correct heteroscedasticity and autocorrelation. Then why there is a need to use robust with two step system gmm.
Panika, it is still to correct for heteroscedasticity.
@@CrunchEconometrix But robust in two step GMM is used for Windmeijer (2005) finite sample correction to correct for download bais of SE. In literature, no where it is mention that it is to connect heteroscedastic in two step GMM..
Plus it is for finite sample( what does it mean).What is the definition of finite sample.
@@CrunchEconometrix In my case in two-step GMM , when I am applying robust all my coefficients become insignificant ( around 1000 observations).. why so.
Panika, I suggest that you read Roodman (2009, 2014) on the nitty-gritty of the xtabond2 GMM syntax to clear your doubts.
Panika, I can't tell why. Yours may be an isolated case.
The bots are strong in this comment section.
How do you mean, Jesse?
can i include the independent var (explanatory) as endogenous with the lag of independent var in gmm part not as exogenous (with IVs part)?
Yes, you can if make the assumption that the explvar is either endogenous or predetermined. I did a similar simulation in both the 1-step and 2-step videos. Watch them to understand what I mean. Thanks.
@@CrunchEconometrix thnx very,, thumb up for ur great job
Thanks very much