* This program was used to generate the output in the paper
* "The Stata module felsdvreg to estimate a linear model with two high-dimensional fixed effects"
* (Revised version of "Using Stata for a memory saving fixed effects estimation of the three-way error component model")

*Manual FEiLSDVj estimation by means of xtreg
*chosing first firm effect per group as reference
sjlog using fels1, replace 
use felsdvsimul
tabulate i, generate(p)
tabulate j, generate(f)
xtreg y x1 x2 f4-f5 f7-f9 f11-f12 f14-f15, fe i(i)
sjlog close, replace

sjlog using fels2, replace
predict peffxt, u
table i, contents(m peffxt)
sjlog close, replace

*FELSDV estimation by means of felsdvreg.ado
sjlog using fels3, replace
felsdvreg y x1 x2, ivar(i) jvar(j) feff(feffhat) peff(peffhat) mover(mover) group(group) xb(xb) res(res) mnum(mnum) pobs(pobs)
sjlog close, replace

*Display firm effects and person effects
sjlog using fels4, replace
table j, contents(m feffhat)
table i, contents(m peffhat)
sjlog close, replace

*FELSDV estimation by means of felsdvreg.ado, with extra output
sjlog using fels5, replace
felsdvreg y x1 x2, ivar(i) jvar(j) feff(feffhat) peff(peffhat) mover(mover) group(group) xb(xb) res(res) mnum(mnum) pobs(pobs) noisily
sjlog close, replace

*Produce correlations between person and firm effects
sjlog using fels6, replace
correlate feffhat peffhat
correlate feffhat peffhat if group==1
sjlog close, replace

sjlog using fels7, replace
sort group
by group: egen pmean=mean(peffhat)
by group: egen fmean=mean(feffhat)
generate peffnorm=peffhat-pmean
generate feffnorm=feffhat-fmean
generate gmean=pmean+fmean
table group, contents(m gmean)
correlate feffnorm peffnorm
correlate feffnorm peffnorm if group==1
sjlog close, replace