Panel estimation in system window
Posted: Sat Mar 13, 2010 5:56 pm
Am I correct in thinking that if I constrain the coefficients (apart from the intercept) across each equation in a SUR framework to be identical, I effectively have a panel?
What I'm really trying to achieve is to see if my SUR (with lagged dependent variable) can be restricted to a dynamic panel structure (my data has a time dimension much larger than the cross-section).
Also, am I correct in thinking that the point of GMM estimators like Anderson-Hsiao and Arellano-Bond is to avoid the bias that arises from the correlation between fixed effects on the lagged dependent variable and the idiosyncratic error?
However, if a SUR with cross-equation coefficient restrictions becomes a dynamic panel, I wouldn't even need fixed effects right? Surely I can account for the fixed effects by letting the intercept term differ across equations, so there is no Nickell bias?
What I'm really trying to achieve is to see if my SUR (with lagged dependent variable) can be restricted to a dynamic panel structure (my data has a time dimension much larger than the cross-section).
Also, am I correct in thinking that the point of GMM estimators like Anderson-Hsiao and Arellano-Bond is to avoid the bias that arises from the correlation between fixed effects on the lagged dependent variable and the idiosyncratic error?
However, if a SUR with cross-equation coefficient restrictions becomes a dynamic panel, I wouldn't even need fixed effects right? Surely I can account for the fixed effects by letting the intercept term differ across equations, so there is no Nickell bias?