Hi everybody,
I am currently trying to set up a simulation exercise in Stata SE 14.0 and could use some input. In particular, I have a panel data set with firms and years and would like to solidify my findings from a difference-in-difference estimation by running a simulation, which randomly assigns firm-years to the "treated" (1) or "untreated" (0) groups.
The basic equation to be estimated is yit = aj + ab + ay + b treat * inti,t-1 + eit. I want to replace the real data for "treat" (1/0) with simulated values and re-run the original estimation 1000 times.
I am using the user-written command reghdfe (available from SSC, highly recommended) to deal with a faster estimation of regressions with multiple fixed effects. (Note that this is largely irrelevant for my question.)
The code I am running looks like this:
I have not worked with simulations before but this estimation will take days even on the fastest PC I have at my disposal. Is there any more efficient way to set this up?
Cheers,
Karsten
I am currently trying to set up a simulation exercise in Stata SE 14.0 and could use some input. In particular, I have a panel data set with firms and years and would like to solidify my findings from a difference-in-difference estimation by running a simulation, which randomly assigns firm-years to the "treated" (1) or "untreated" (0) groups.
The basic equation to be estimated is yit = aj + ab + ay + b treat * inti,t-1 + eit. I want to replace the real data for "treat" (1/0) with simulated values and re-run the original estimation 1000 times.
I am using the user-written command reghdfe (available from SSC, highly recommended) to deal with a faster estimation of regressions with multiple fixed effects. (Note that this is largely irrelevant for my question.)
The code I am running looks like this:
Code:
set seed 12345
postfile simul beta using simulation, replace
forvalues i = 1/1000 {
gen order=runiform()
sort order
gen plac=0 if treat!=.
replace plac=1 if _n<=3590 & treat!=.
drop order
sort gvkey year
gen ltreat_int=l.treat*int
reghdfe y ltreat_int, a(year industry bank) cluster(gvkey)
loc beta=_b[ltreat_int]
post simul (`beta')
drop ltreat_int
drop plac
}
postclose simul
use simulation, clear
summarize
Cheers,
Karsten