这个是乱弹琴吧，用最后一个数值作为第一个非缺失值？

You are right, the above second method still does not work so well for complex conditions. Just think there should be a simpler method than the first one. Try the following in your data:

clear
input id year yvar
100 1995 23
100 1996 45
100 1997 66
200 1995 .
200 1996 .
200 1997 33
200 1998 .
200 1999 45
300 1995 .
300 1996 .
end

egen firstyvar=first(yvar), by(id)        
bysort id (firstyvar): replace firstyvar=firstyvar[_n-1] if missing(firstyvar)

g m=mi(yvar)
bys m id (year): g f=yvar if _n==1
bys id (year): egen first=min(f)
drop m f

这个是对的，他重新生成了一个变量。