Summation of the first 5 highest observations

[abdulla]

gvkey	fy	TA
1700	1995	3311.662
1700	1995	1599.234
1700	1995	564.76
1700	1995	560.047
1700	1995	656.226
1700	1996	577.841
1700	1996	1696.431
1700	1996	1081.963
1700	1996	1727.069
1700	1996	4182.832
1700	1996	2068.554
1700	1996	841.204
1700	1996	.
84578	1995	1312.852
84578	1995	1131.176
84578	1995	3700.925
84578	1995	1099.786
84578	1995	848.657
84578	1995	1067.566

The above is my data and I want to get the following result. 

gvkey	fy	TA	Sum
1700	1995	3311.662	6691.929
1700	1995	1599.234	6691.929
1700	1995	656.226	6691.929
1700	1995	564.76	6691.929
1700	1995	560.047	6691.929
1700	1996	4182.832	10756.85
1700	1996	2068.554	10756.85
1700	1996	1727.069	10756.85
1700	1996	1696.431	10756.85
1700	1996	1081.963	10756.85
1700	1996	841.204	.
1700	1996	577.841	.
1700	1996	.	.
84578	1995	3700.925	8312.305
84578	1995	1312.852	8312.305
84578	1995	1131.176	8312.305
84578	1995	1099.786	8312.305
84578	1995	1067.566	8312.305
84578	1995	848.657	.

I want to get the sum of the first 5 highest observations. I have used the following code. 

proc sort data= have;
by gvkey fy decending TA; run;

Proc sql;
create table want  as
select *,sum(TA) as sum
from have
group by gvkey, fy;
quit;

I get the sum but if I have more than 5 observations in the TA column, I get sum for all of the observations. I want the sum for only the first 5 highest observations. And most importantly I don't want to lose any other rows also. I need to have all the rows in my output. Can anyone please help?

[novinosrin · Accepted solution]

Hi @abdulla 

data have;
input gvkey	fy	TA ;
cards;
1311	1995	3311.662
1311	1995	1599.234
1311	1995	564.76
1311	1995	560.047
1311	1995	656.226
1700	1996	577.841
1700	1996	1696.431
1700	1996	1081.963
1700	1996	1727.069
1700	1996	4182.832
1700	1996	2068.554
1700	1996	841.204
1700	1996	.
84578	1995	1312.852
84578	1995	1131.176
84578	1995	3700.925
84578	1995	1099.786
84578	1995	848.657
84578	1995	1067.566
;

proc sort data= have;
by gvkey fy decending TA;
run;

data want;
 do _n_=1 by 1 until(last.gvkey);
  set have;
  by gvkey;
  if _n_<=5 then sum=sum(sum,ta);
 end;
 do _n_=1 to _n_;
  set have;
  if _n_>5 then call missing(sum);
  output;
 end;
run;

[abdulla]

I need it by fy too. Now I am getting a lot of missing for fy.
I have data like the following.
gvkey fy
1311 1995
1311 1995
1311 1995
1311 1995
1311 1995
1311 1996
1311 1996
1311 1996
1311 1996
1311 1996
1311 1996
1311 1996
1311 1996
1311 1997
1311 1997
1311 1997
1311 1997
1311 1997
1311 1997

[ChrisNZ]

Did you try to understand the solution given?

You just have to modify in a couple of places.

Please use this forum as an opportunity to learn.

data WANT;
 do _N_=1 by 1 until(last.FY);
  set HAVE;
  by GVKEY FY;
  if _N_ <= 5 then SUM=sum(SUM,TA);
 end;
 do _N_=1 to _N_;
  set HAVE;
  if _N_ > 5 then call missing(SUM);
  output;
 end;
run;

gvkey	fy	TA	SUM
1311	1995	3311.66	6691.93
1311	1995	1599.23	6691.93
1311	1995	656.23	6691.93
1311	1995	564.76	6691.93
1311	1995	560.05	6691.93
1700	1996	4182.83	10756.85
1700	1996	2068.55	10756.85
1700	1996	1727.07	10756.85
1700	1996	1696.43	10756.85
1700	1996	1081.96	10756.85
1700	1996	841.20	.
1700	1996	577.84	.
1700	1996	.	.
84578	1995	3700.93	8312.31
84578	1995	1312.85	8312.31
84578	1995	1131.18	8312.31
84578	1995	1099.79	8312.31
84578	1995	1067.57	8312.31
84578	1995	848.66	.

[Patrick]

You could use Proc Rank to also deal with ties if any.

proc rank data=have out=inter descending ties=dense ;
  ranks rank;
  by gvkey fy;
  var ta;
run;

proc sql;
/*  create table want as*/
    select 
      gvkey, 
      fy,
      TA, 
      sum(ta) as sum_ta
    from inter
    where rank<=5
    group by 
      gvkey, 
      fy 
    ;
quit;