I am about to estimate a dynamic panel data model via xtabond2 module in Stata 15 (or Dynamic Panel Data wizard in Eviews 10).
The model includes 7 explanatory variables which build a common sample properties of N=31 and T=7... .
Before estimation, I calculated the matrix of correlations between explanatory variables and unfortunately encountered many values around 0.8 or above.
The question here is that: Is multicollinearity an issue in this case, which applies system GMM estimators, as it is with Fixed Effects estimators?
Multi collinearity is not a problem when you utilize instrumental analysis that isolate own effect of independent variables from group effect and other variable effect.
Thank you Dawood Mamoon , and Daniel Meierrieks for your useful and detailed explanation.
Theoretically speaking, multicollineartiy stems from denominator of variance expression for multiple regression OLS estimators. However, I don't know whether the variance of iterated GMM estimator contains such an expression?
Anyway, I estimated the models in Eviews 10 (with both orthogonal deviation and first difference transformation) and observed interesting results: in models prone to multicollinearty, if orthogonal deviation method were applied, the highly-correlated variables' coefficients would be all significant; but, in first difference case, the coefficients were not robust and significant, which may occur as a result of multicollinearity (؟).
In addition, is variance inflation applicable where there is no R? In case, what would be the VIF formula?
Dear Hamed
I think the reason of non-significant results for your explanatory variables is using first difference. Usually economical variables after differencing are non-significant.
VIF = 1 / (1 - R^2) where R^2 is coefficient of determination of your model.
Dear Seyed Mahdi Amir Jahanshahi
Thank you for your answer,
As the dynamic panel data models are estimated after n-steps in Eviews 10, the matter is about the determination of R^2, especially when no intercept is included.
Dawood Mamoon Sir, could you please assist me by providing any sources for your statement "Multicollinearity is not a problem when you utilize instrumental analysis that isolates own effect of independent variables from group effect and other variable effects".
Thanks in Advance
- Badges
- Science topic