I am trying to use xtabond2 to measures the impact of financial liberalisation and banking crisis in 16 countries from period 1973-2005. I include time dummies in the dynamic panel data model.
I intend to use GMM estimator and used the below command;
xtabond2 GDPPCR logGDPPCl logPC FL Bcrisis Investment Openness Govsize logEducation logInflation IYear1974-Iyear2005, gmm(logGDPPCl logPC FL Bcrisis, laglimits (1 1) collapse) ivstyle( Investment Openness Govsize logInflation logEducation, equation(level)) ivstyle(IYear1974-Iyear2005) small robust arlevels nodiffsargan
In the results appear that there is 47 instruments.
How can I do to reduce it?
Thank you for your comments.
what do you mean by 47 instruments? if it mean 47 predictor variables in the model, you have different alternative approaches to reduce the number predictor variables.
one of the widely used method is to run factor analysis and based on th underlying correlation structure among the variables, you can identify representative varaible for each component of factor analysis i.e., the variable with highest factor loading from each component will be selected for further analysis,
if you provide more details, i can provide more inputs for your analysis
with regards
Hi Alej,
For GMM, instrumental variables should not be correlated with your main variables, that is why we use lags (1st or/and 2nd lag) to construct instrumental variables. In order to reduce the number of your instrumental variables, you have to regress lags of your main variables on your dependent variable first and then select the ones with the highest Adj R^2.
For more details see:
https://teach.business.uq.edu.au/courses/FINM6905/files/module-1/readings/GMM%20in%20Finance.pdf
Hope this helps
Dear Shrikanth and Mo Sherif,
Thank you for recommendations. I cheked the paper Mo suggest and other literatura and finally I can resolve my problem.
I use the below command:
xtabond2 GDPPCR logGDPPCl logPC FL Bcrisis Investment Openness Govsize logEducation logInflation, gmm(logGDPPCl logPC FL Bcrisis, laglimits (1 0) collapse) ivstyle(Investment Openness Govsize logInflation logEducation) small orthogonal leveleq robust arlevels nodiffsargan
The results show only 16 instruments. This number is = to the number of countries I use as a sample.
My best regards,
Alejandro Santana
Hi Alejandro,
I am having the same issue with my instruments being greater than the group. I look at your code, what I didn't understand is that in the bracket after gmm. that's the gmm style instrument should be just the dependent variables right? or can you add other variables too. can you explain the choice of your variables inn gmm and iv style bracket? I would really appreciate it.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Thermodynamics