no observation error in Arellano-Bond estimator

[Enza Testa]

Hi guys, I have a 5-years panel and I did a pooled and fixed/random regressions. Now I'm trying to run an Arellano-Bond estimator but Stata tells "error 2000, no observations". I can't understand the problem since I worked with the same panel with others models. I looked for the solution in some topic and someone suggests someting about string variables but I haven't or about missing values and I have many. Is it a problem for this kind of model? Any suggestions? Thanks a lot

[Clyde Schechter]

I know very little about Arellano-Bond, but the one thing I do know about it is that, as with any Stata estimation command, an observation can be included in the estimation only if all of the variables in the command have non-missing values. In the case of -xtabond-, this applies also to all of the designated lags of the dependent variable. So, you will probably find that when you put all of those restrictions together, you have no observations left. If that is not the case, I suggest you post some example data, and it is likely that somebody who is familiar with using -xtabond- will jump in and help.

[Sebastian Kripfganz]

There might be another typical source of mistake that could lead to this problem. Do you mean by "5-years panel" that you have a panel data set with 5 consecutive observations or do you mean a panel where all the variables are 5-year averages (as sometimes done in the empirical macroeconomic growth literature)? In the latter case, Stata might try to create lags based on the previous year's observation while the previous observation is actually 5 years in the past. To figure that out, please report what Stata gives back when you type

xtset

In particular the reported delta is of interest.

[Enza Testa]

My variables are 5-years averages from 1970 to 2014. So I have 9 observation for each country and each variables to capture a long run effect as the empirical macroeconomic growth literature suggests. This is what I obtained typing xtset

. xtset
panel variable: id (strongly balanced)
time variable: year, 1970 to 2014
delta: 1 unit

[Sebastian Kripfganz]

That is exactly what I was suspecting. The following should solve your problem:

xtset id year, delta(5)

The option delta(5) tells Stata that consecutive observations are 5 units (years) apart from each other. With that information, the Arellano-Bond estimator can compute all the lags accordingly in the usual way.

[Enza Testa]

Now it seems work. Thank you so much. Can I ask you someting more specific about the estimator? Is the first time I use it and I have many doubts.

[Sebastian Kripfganz]

Now it seems work. Thank you so much. Can I ask you someting more specific about the estimator? Is the first time I use it and I have many doubts.

You can post your question here on Statalist. I or others might respond as usual. I would suggest to start a new topic because the problem discribed here is solved. In any case, please use the search functionality of this forum first to find out if somebody else had a similar question before that might have been answered already. Also, please have a look at the Stata documentation for the respective command (xtabond and/or xtdpd) and the examples therein. Further useful guidance can be found in the following article:

• Roodman, D. (2009). How to do xtabond2: An introduction to difference and system GMM in Stata. Stata Journal 9 (1), 86-136.

[Enza Testa]

Ok ok thank you. I look at this paper and in case I post a new question.

[Enza Testa]

Hi guys, I have again the same problem. Stata tells me "no observations". I put the delta option as suggested above and if I run the model with my first dependent variable everything is ok,but when I change the dependent variable (it's just another GDP specification) I have the same error, I can't understand why. The second dependent variable has fewer observations than the first one. It can cause the problem?