Calculate the Number of loans the company had over the past 5 years

[Agent1592]

I would like to calculate the number of loansa company had for the past 5 years. What is the SAS SQL code to do this?

Here is the have table:

CompanyID	LoanID	FYEAR
1004	5006	1989
1004	5006	1990
1004	6460	1990
1004	5006	1991
1004	6460	1991
1004	7970	1991
1004	5006	1992
1004	6460	1992
1004	7970	1992
1004	5006	1993
1004	7970	1993
1004	5006	1994
1004	7970	1994
1004	5006	1995
1004	7970	1995
1004	7970	1996
1004	34433	1996
1004	34441	1996
1004	7970	1997
1004	34433	1997
1004	34441	1997
1004	45856	1997
1004	46572	1997
1004	49136	1997
1004	7970	1998
1004	34433	1998
1004	34441	1998
1004	45856	1998
1004	46572	1998
1004	49136	1998
1004	34433	1999
1004	34441	1999
1004	46572	1999
1004	49136	1999
1004	34433	2000
1004	34441	2000
1004	100472	2000
1004	34433	2001
1004	34441	2001
1004	111500	2001
1004	34433	2002
1004	34441	2002
1004	109872	2002
1004	119544	2002
1004	34433	2003
1004	34441	2003
1004	109872	2003
1004	119544	2003
1004	109872	2004
1004	119544	2004
1004	153686	2006
1004	153686	2007
1004	165317	2007
1004	153686	2008
1004	165317	2008
1004	153686	2009
1004	165317	2009
1004	187804	2009
1004	165317	2010
1004	197270	2010
1004	187804	2010
1004	197270	2011
1004	187804	2011
1004	197270	2012
1004	187804	2012
1004	197270	2013
1004	187804	2013
1004	197270	2014
1004	241601	2014

[Reeza]

 See example 12 + a WHERE clause. 

http://www.listendata.com/2014/04/proc-sql-select-statement.html

If you run into issues post the code and log and explain your issue in detail. 

[Agent1592]

I tried this code but I am not sure if it is right?

proc SQL;
create table want  as
select a.* , count(unique b.LoanID) as NUMLOANS
from have as a
left join have as b
on a.CompanyID=b.CompanyID and a.FYEAR=b.FYEAR+5; quit;

[Reeza]

You join a table when you have information in another table or rows that you need. 

Nothing in your question of indicates a need for a join. Remove the join and use a where. I would recommend hardcoding the years first and then making it dynamic after you sort out the rest of the code. 

Unless you were looking for a moving average? Nothing in your question mentions this though. 

[Agent1592]

No I do not want a moving average. I would like a moving count of the number of loans over the past 5 years. Thank you.

Is this correct?

proc SQL;
create table want as
select * , count(unique LoanID) as LOANFREQ
from have where FYEAR<=FYEAR+5
group by LoanID,CompanyID, fyear
order by CompanyID,fyear,LoanID; quit;

[Reeza]

LoanID can't be in your group statement since you want a count for it..

If you want a moving count (my mistake in average vs count) you do need a join. If you search through my posts in last week there's a good example of the three different methods in someone else's question. 

[Astounding]

If I'm interpreting your question correctly, this is too complex an approach.  When you say the past 5 years, don't you want:

• One observation per company
• A count of number of loans in the time period FYEAR >= 2011

If so, that's much easier.  SQL can do this in one step.  PROC FREQ can do it in two steps:

proc freq data=have;

tables CompanyID * LoanID / noprint out=counts;

where fyear >= 2011;

run;

proc freq data=counts;

tables CompanyID;

run;

[Agent1592]

Thanks, I am looking for a moving count though. Not only 2011-2016 but since inception.

[mkeintz]

Rolling counts through (ugh!) SQL.  Assuming you have one record per loan in dataset HAVE, and each loan is a five year loan, orginating in year FYEAR:

Edit: forget the companyid:

proc sql;
  create table want as 
    select companyid, year, count(*) 
    from (
       (select companyid, fyear+0 as year from have) outer union corresponding
       (select companyid, fyear+1 as year from have) outer union corresponding
       (select companyid, fyear+2 as year from have) outer union corresponding
       (select companyid, fyear+3 as year from have) outer union corresponding
       (select companyid, fyear+4 as year from have)
        )
    group by companyid, year;
quit;

A far better way would be:

data need (keep=year)/view=need;
  set have;
  do year=fyear to fyear+4;
    output;
  end;
run;
proc freq data=need;
  tables companyid *year /list;
run;

[Astounding]

Since you want a moving count, I would modify what you started doing:

proc SQL;
create table want  as
select a.* , count(unique b.LoanID) as NUMLOANS
from have as a
left join have as b
on a.CompanyID=b.CompanyID and a.FYEAR=b.FYEAR+5; quit;

quit;

The first fix, is what gets counted.  (You'll have to check my syntax on this since my SQL is suspect.)

count(distinct b.CompanyID, b.LoanID)

Secondly, change to a WHERE condition instead of a join:

where a.CompanyID=b.CompanyID and (a.FYEAR >= b.FYEAR) and (a.FYEAR - 4 <= b.FYEAR)

You can adjust the time periods if I didn't get the 5-year window correctly.

[Ksharp]

But you didn't post the output yet.

data have;
infile cards truncover expandtabs;
input CompanyID	LoanID	FYEAR;
cards;
1004	5006	1989
1004	5006	1990
1004	6460	1990
1004	5006	1991
1004	6460	1991
1004	7970	1991
1004	5006	1992
1004	6460	1992
1004	7970	1992
1004	5006	1993
1004	7970	1993
1004	5006	1994
1004	7970	1994
1004	5006	1995
1004	7970	1995
1004	7970	1996
1004	34433	1996
1004	34441	1996
1004	7970	1997
1004	34433	1997
1004	34441	1997
1004	45856	1997
1004	46572	1997
1004	49136	1997
1004	7970	1998
1004	34433	1998
1004	34441	1998
1004	45856	1998
1004	46572	1998
1004	49136	1998
1004	34433	1999
1004	34441	1999
1004	46572	1999
1004	49136	1999
1004	34433	2000
1004	34441	2000
1004	100472	2000
1004	34433	2001
1004	34441	2001
1004	111500	2001
1004	34433	2002
1004	34441	2002
1004	109872	2002
1004	119544	2002
1004	34433	2003
1004	34441	2003
1004	109872	2003
1004	119544	2003
1004	109872	2004
1004	119544	2004
1004	153686	2006
1004	153686	2007
1004	165317	2007
1004	153686	2008
1004	165317	2008
1004	153686	2009
1004	165317	2009
1004	187804	2009
1004	165317	2010
1004	197270	2010
1004	187804	2010
1004	197270	2011
1004	187804	2011
1004	197270	2012
1004	187804	2012
1004	197270	2013
1004	187804	2013
1004	197270	2014
1004	241601	2014
;
run;


proc SQL;
create table want  as
select b.* ,(select count(unique a.LoanID) 
from have as a where a.CompanyID=b.CompanyID and 
a.FYEAR between b.FYEAR and b.FYEAR+5) as NUMLOANS
from have as b;
quit;

[Agent1592]

Thanks you. Sorry this is the want table.

[Agent1592]

Want table.

CompanyID	LoanID	FYEAR	N_LOANS_1YEAR	AVG_LOANS_5YEAR	MAX_LOANS_5YEAR
1004	5006	1989	1		1
1004	5006	1990	2		2
1004	5006	1991	3		3
1004	5006	1992	3		3
1004	5006	1993	2	2.2	3
1004	5006	1994	2	2.4	3
1004	5006	1995	2	2.4	3
1004	7970	1996	3	2.4	3
1004	7970	1997	6	3	6
1004	7970	1998	6	3.8	6
1004	34433	1999	4	4.2	6
1004	34433	2000	3	4.4	6
1004	34433	2001	3	4.4	6
1004	34433	2002	4	4	6
1004	34433	2003	4	3.6	4
1004	109872	2004	2	3.2	4
1004	153686	2006	1	2.8	4
1004	153686	2007	2	2.6	4
1004	153686	2008	2	2.2	4
1004	153686	2009	3	2	3
1004	165317	2010	3	2.2	3
1004	197270	2011	2	2.4	3
1004	197270	2012	2	2.4	3
1004	197270	2013	2	2.4	3
1004	197270	2014	2	2.2	3

[Ksharp · Accepted solution]

Do you have to use SQL ?

data have;
infile cards truncover expandtabs;
input CompanyID	LoanID	FYEAR;
cards;
1004	5006	1989
1004	5006	1990
1004	6460	1990
1004	5006	1991
1004	6460	1991
1004	7970	1991
1004	5006	1992
1004	6460	1992
1004	7970	1992
1004	5006	1993
1004	7970	1993
1004	5006	1994
1004	7970	1994
1004	5006	1995
1004	7970	1995
1004	7970	1996
1004	34433	1996
1004	34441	1996
1004	7970	1997
1004	34433	1997
1004	34441	1997
1004	45856	1997
1004	46572	1997
1004	49136	1997
1004	7970	1998
1004	34433	1998
1004	34441	1998
1004	45856	1998
1004	46572	1998
1004	49136	1998
1004	34433	1999
1004	34441	1999
1004	46572	1999
1004	49136	1999
1004	34433	2000
1004	34441	2000
1004	100472	2000
1004	34433	2001
1004	34441	2001
1004	111500	2001
1004	34433	2002
1004	34441	2002
1004	109872	2002
1004	119544	2002
1004	34433	2003
1004	34441	2003
1004	109872	2003
1004	119544	2003
1004	109872	2004
1004	119544	2004
1004	153686	2006
1004	153686	2007
1004	165317	2007
1004	153686	2008
1004	165317	2008
1004	153686	2009
1004	165317	2009
1004	187804	2009
1004	165317	2010
1004	197270	2010
1004	187804	2010
1004	197270	2011
1004	187804	2011
1004	197270	2012
1004	187804	2012
1004	197270	2013
1004	187804	2013
1004	197270	2014
1004	241601	2014
;
run;


proc SQL;
create table temp as
select CompanyID,FYEAR,count(distinct LoanID) as N_LOANS_1YEAR
from have 
group by CompanyID,FYEAR;

create table want  as
select *,case when 
(select count(*) from temp where CompanyID=a.CompanyID and FYEAR le a.FYEAR) lt 5 then .
 else (select avg(N_LOANS_1YEAR) from temp where CompanyID=a.CompanyID 
 and FYEAR between a.FYEAR-4 and a.FYEAR) end as AVG_LOANS_5YEAR,
(select max(N_LOANS_1YEAR) from temp where CompanyID=a.CompanyID 
 and FYEAR between a.FYEAR-4 and a.FYEAR) as MAX_LOANS_5YEAR
 
from temp as a;
quit;

[Agent1592]

No i don't necessarily need SQL. If there is an easier SAS code I can use that. Thanks