Useful programs

These programs should (and will) be on a repo on Github.

Replace top-coded variables with Pareto distribution mean

Values for variable are top-coded above a threshold. The variable is known/thought to follow a Pareto distribution. The Pareto distribution if estimated on a number of variables below the top-coded threshold (say the Pareto distribution is estimated on values in the top 90-99 pct bin if the top-coded values correspond to the top 1pct).

This program is taken from Blundell, Pistaferri and Preston (2017) paper. Needs 3 globals: year, cens_var and topcode.

cap program drop pareto
program define pareto
    preserve
    keep if year == $year
    egen fr_gt_y=rank($cens_var),field
    replace fr_gt_y=fr_gt_y-1
    egen tot_y=sum($cens_var!=.) 
    replace fr_gt_y=fr_gt_y/tot_y 
    keep if $cens_var<$topcode 
    xtile qy=$cens_var,nq(10) 
    keep if qy==10 
    collapse fr_gt_y,by($cens_var) 
    replace fr_gt_y=ln(fr_gt_y) 
    replace $cens_var=ln($cens_var) 
    reg fr_gt_y $cens_var 
    restore 
    gen b=-_b[$cens_var] 
    replace $cens_var=$cens_var*(b/(b-1)) if $cens_var==$topcode & year==$year 
    end