Xtabond2 predetermined variables

[Dana Baade]

Hi,

Xtabond2 allows us to use exogenous, predetermined and endogenous variables. I have a regression in the form of

y=yt-1+x1_t-1+x2_t-1+i.year

where my independent variables are thus all lagged one period. I furthermore use fixed effects. I was wondering if I am allowed to treat xt2_t-1 as strictly exogenous for either difference or system gmm. My thinking is that it is correlated with previous error terms and that it was thus predetermined and should thus automatically be instrumented gmm style in stead of iv style. But I am not sure.

My question is thus: should lagged explanatory variables always be treated as being at least predetermined? Or can they be exogenous with respect to the error idiosyncratic error term?

[Sebastian Kripfganz]

If a variable is correlated with previous error terms, then it cannot be strictly exogenous. This is independent of whether the regressor is lagged or not. The decision to lag or not lag a regressor should typically not be made based on concerns about exogeneity/predeterminedness of that variable. The relevant question is if you expect the regressor to have a contemporaneous effect on the dependent variable. If yes, you should not lag it. If you only expect a delayed effect, then lagging can be appropriate.

The following might be helpful as well:

• xtdpdgmm: Stata command for GMM estimation of linear dynamic panel data models

[Dana Baade]

Thank you Sebastian Kripfganz,

I am very much thankful for your help. I read the presentations. I have very strong reasons to believe there is lagged effect between x2 and y. I threfore only lagged this variable to be in accordance with previous work and the real world. To be clear: a lagged variabe (x2_t-1) can thus still be exogenous with respect to the idiosyncratic error and included in the iv style instruments?

[Sebastian Kripfganz]

To be clear: a lagged variabe (x2_t-1) can thus still be exogenous with respect to the idiosyncratic error and included in the iv style instruments?

Yes.

[Dana Baade]

Thank you, very clear