Rolling 3SLS regression with nlcom/lincom in STATA

[raghav goyal]

Hi, I am trying to run a rolling 3SLS regression in stata and need to find the coefficient and SE of a linear combination

My regular 3SLS regression looks like this:

reg3 (eq1: Q1 P1 Shock Year_21 Year_22) (eq2: Q2 P3 Shock Year_21 Year_22), endog( P1 P3) exog( Lagged_Shock Lagged_Shock)

And I need this:

nlcom (_b[ P1]+ _b[P2])/2

The program to write the rolling regression is

program drop _all

program roll, rclass
version 9.2
syntax varlist(ts) [if]

reg3 `varlist' = `instruments' if `if', endog(`endog') exog(`exog')

nlcom (_b[P1] + _b[P2]) / 2
mat coefs= r(b)
return scalar lcoefs = coefs[1,1]

mat vars = r(V)
return scalar lvars = vars[1,1]
end

rolling, window(20) saving(rslt, replace): roll Q1 Q2 P1 P2 Shock Shock Year_21 Year_22 Lagged_Shock Lagged_Shock, endog(P1 P2) exog(Shock Shock Year_21 Year_22) instruments(Lagged_Shock Lagged_Shock)


However,
it gives the error:

options not allowed
an error occurred when rolling executed roll


Can anyone help?