Out of sample prediction using a two way fixed effect

[Lily Ksh]

Dear members,

When using a two way fixed effect model, "i.year" is added in the command which estimates the coefficients of the time dummies. If we estimate a two way fixed effect model from 2000-2005, how do we include these time effects for the out of sample prediction from 2006 - 2010?

Any advice is appreciated.

[FernandoRios]

Hi Lily
Unfortunately, you cant do that. Since you never observe 2006-2010 data, you cannot use make predictions based on how your model is, unless you have very strong assumptions.
You could say, for example, that the Fixed effects in those years are zero, or equal to the last observed year (2005). Or use some projection based on what you observe in 2000-2005.
HTH

[Lily Ksh]

Hi Fernando,

Thanks. That makes sense. How can I code it so that it stores all of the coefficients of the time dummies and then estimates the average of the estimated coefficients?

[FernandoRios]

all coefficients are stored in e(b). However, if you are using a linear regression model, I think I would prefer using -reghdfe- (ssc), so you can store the fixed effect in a separate variable.
HTH