Panel Data IV regression with first stage regression only using and predicting for data after a certain year

[Youngsun Hong]

Hello everyone,

I have a set of panel data from 2016 to 2024. I am trying to regress Y on X. X is only relevant from 2019, so the values for X from 2016-2018 are all 0. I want to do an IV regression using Z, but the first stage regression X = a + bZ + u, would be using Xs for all the years, and predict X^ for all years, even those before 2019 that should stay 0 when I use a standard xtivreg command. Is there a way to work around this?

[George Ford]

My inclination would be to have different coefficients for 2016-2018 and 2019-2024 using a time dummy. You're assuming the coef are zero in the early years, which may or may not be true. Or, I suppose you could use a control function and cnsreg to impose the constraint.