Currently, the first_stage regression will produce point estimates in the periods after everyone is treated (assuming those fixed effects are zero). This is not correct and should instead drop those years all together and use the last treated group as the "control group"
Currently, the first_stage regression will produce point estimates in the periods after everyone is treated (assuming those fixed effects are zero). This is not correct and should instead drop those years all together and use the last treated group as the "control group"