Dynamic panel data model

[Mohammad Azeem Khan]

Hello everyone,
I have some problem with the results of the xtabond2 command. When I used the following command " xtabond2 dep var Independent variables year*, gmm(first lag
dep var) iv(strictly exogenous variables among the explanatory variables) nocons twostep robust"
for my dynamic panel model, it gives me the probability value equal to 1 for the Hansen statistics. How should I interpret such a value? Is there any error with the model?

Is there any issue if the number of instruments exceeds the number of cross-sectional groups in the models because it is written in some papers on dynamic panel GMM that the no. of instruments should be lower than the number of groups.
What is the meaning of putting the "year*" in the command?

Thanks.

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions; provide Stata code in code delimiters, readable Stata output, and sample data using dataex. We work better with results in the text rather than as pictures (which are often hard to read).

You should look up Roodman's working paper on xtabond. If you're not careful, xtabond can give you too many instruments relative to sample size. As far as I know, year has the same role in xtabond as elsewhere - a control for time (but often it is done i.year to allow dummies for different years rather than assuming a linear influence of date).

[Mohammad Azeem Khan]

OK thanks, I will follow your advice and then I will post it again.

[Mohammad Azeem Khan]

Hi statalist members,

Can we include time fixed effect and cross-sectional entity specific effect in the random effect model like the fixed effect model by putting the following command?

xtreg depvar independent vars i.time i.id ,re

Does it work in the random effect model?

Moreover, if the Hausman test prefers either fixed or random effect models when we have not included time effect in both the fixed and random effect models (i.e no i.time command). Now if we include time effect in the models, do we need to again run the Hausman test? or the model preferred in the first case (whether fixed or random) will also remain robust here?

Thanks
Azeem

[Mohammad Azeem Khan]

I am running the panel data with fixed effect model in following manners

xtreg logz_score Log_HHIL2 i.year,fe

, where log_zscore is the dependent variable and HHIL2 is the independent variable which is varying over time but not across panels. The coefficient for Log_HHIL2 is very high ( i.e. -2131.4 ) and all the coefficients for years have same t statistics and standard errors. But when I drop the year effect using following command

xtreg logz_score Log_HHIL2, fe

, I am getting normal coefficient for Log_HHIL2. Kindly help me, I am posting my results for a clear understanding of my problem.

[Mohammad Azeem Khan]

Please ignore the attached results files, only considers the images. I am Sorry for the inconvenience.