Instrumental variables / 2sls

[Mateo Capone]

Dear all,

I have a panel with 10T and around 65000 observations.

I would like to use the instrumental variables estimation approach in order to take care of endogeneity concerns.

My model is: Var1 = L.Var1 + Var2 + Var3 + Var4 + Var5 + Var6 + Var7 + Var8 + Var9 + yeardummy.

I would like to consider all my independent variables as endogenous and their first lag as instrument, translating it to Stata: ivregress 2sls Var1 yeardummy (L.Var1 Var2 Var3 Var4 Var5 Var6 Var7 Var8 Var9 = L2.Var1 L.Var2 L.Var3 L.Var4 L.Var5 L.Var6 L.Var7 L.Var8 L.Var9).

Appreciate any help or idea, as I think something is missing or is not correct.

Thank you.

[Sebastian Kripfganz]

I assume "10T" means that you have 10 observations (time periods) per group. Thus, you are trying to estimate a short-T dynamic panel data model.

Your approach only works if there is no unobserved group-specific heterogeneity (fixed effects). Otherwise, the lagged levels used as instruments will still be correlated with these unobserved effects. If you are assuming away those unobserved effects, then I do not see a reason for instrumenting the lagged dependent variable, unless you are assuming that there is first-order (but no higher-order) serial correlation in the error term.

A GMM approach might be more appropriate for your data:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.