endogenous vs predetermined variables in xtivreg

[graziano ceddia]

Dear Professor @Wooldridge
I am using xtivreg to estimate a FE panel data with some endogenous independent variables. How can I determined if these variables are truly endogenous or pre-determined (in which case I could use their one-year lag as instruments)? Is there any specific test to distinguish strict endogeneity from pre-determined?
At the moment I have run an Hausman test (under the assumptions that the variables are exogenous vs. non-exogenous). The result tells me that the variables are not exogenous (i.e., at best pre-determined). Yet, I would like to know whether they are pre-determined or strictly endogenous.
Could anyone help?
Thanks a lot for your assistance.

[Manh Hoang Ba]

Following the within transformation, both endogenous and predetermined variables are treated as endogenous, while strictly exogenous variables remain strictly exogenous. Consequently, when implementing FE2SLS/RE2SLS (xtivreg), it is unnecessary to distinguish between endogenous and predetermined variables. Rather, the original regressors need only be divided into two categories: strictly exogenous variables and all other variables.