topic Re: Calculating Product for a selected observations of the same variable in SAS Programming

Calculating Product for a selected observations of the same variable

ksachar — Sat, 25 Feb 2017 05:50:25 GMT

I need to calculate modified deitz return (Which is basically geometric mean of periodic return after adjusting Cash Flow and weighted Cash flow). You calculate monthly return, and for larger periods take geometric mean for it. Say you have monthly return and you can calculate annual return by taking Geometric mean.

I am manipulating data in SAS EG so I can use the dataset in SAS VA for a Report.

Geometric Mean=(1+r1)*(1+r2)...... -1

r1, r2== are monthly returns.

The idea is to select a random period Date_From and Date_To and the pre-calculated Rate of return come up.

I have calculated the monthly return with adjusted cash flow, I am strugling to get a Geometric mean/Cumulative Return.

I HAVE:

Client	Financial Product	Date	Return	Return +1
Customer A	Product A	Jan2016	2%	$1.02
	Product A	Feb2016	3%	$1.03
	Product A	Mar2016	1%	$1.01
	Product A	Apr2016	7%	$1.07
	Product A	May2016	-5%	$0.95
	Product A	Jun2016	3%	$1.03
	Product A	Jul2016	-0.50%	$1.00
	Product A	Aug2016	3%	$1.03
	Product A	Sep2016	2%	$1.02
	Product A	Oct2016	5%	$1.05
	Product A	Nov2016	1%	$1.01
	Product A	Dec2016	-1%	$0.99
Customer A	Product B
....	....
.....	.....
.......	.......
.........	.........
.........	.........

I WANT:

Financial Product	Date_From	Date_To	GM_Formulae	Cumulative Return
Product A	Jan2016	Feb2016	(1+r2)-1	2.00%
Product A	Jan2016	Mar2016	(1+r2)*(1+r3)-1	5.06%
Product A	Jan2016	Apr2016	....	....
Product A	Jan2016	May2016	.....	.....
Product A	Jan2016	Jun2016	.......	.......
Product A	Jan2016	Jul2016	.........	.........
Product A	Jan2016	Aug2016	.........	.........
Product A	Jan2016	Sep2016	....	....
Product A	Jan2016	Oct2016	.....	.....
Product A	Jan2016	Nov2016	.......	.......
Product A	Jan2016	Dec2016	(1+r1)(1+r2)....(1+r11)-1	23.16%
Product A	Feb2016	Mar2016	.........	.........
Product A	Feb2016	Apr2016	....	....
Product A	Feb2016	May2016	.....	.....
Product A	Feb2016	Jun2016	.......	.......
Product A	Feb2016	Jul2016	.........	.........
Product A	Feb2016	Aug2016	.........	.........
Product A	Feb2016	Sep2016
Product A	Feb2016	Oct2016
Product A	Feb2016	Nov2016
Product A	Feb2016	Dec2016

For 12 months there will be 66 combinations of months

The date combinations are easy to create. But I am having trouble calculating Rate of return for all teh 66 combinations. Since it is teh same variable.

There are ways to this in excel. Since, actual table is millions of rows thats not a possibility. Can this be done in SAS.

Re: Calculating Product for a selected observations of the same variable

Reeza — Sat, 25 Feb 2017 06:03:34 GMT

Of course you can do this in SAS 🙂

I'm assuming your question is how do you do this in SAS?

When googling I suggest limiting search to lexjansen.com or include SAS NOTES.

1. Why are you missing months in your data?

2. How do you account for missing months?

3. How do calculate the rate at the yearly boundaries, does it reset at some point?

4. How big is your dataset - 5 million is different than 200 million.

There's a geomean function but that would mean transposing your dataset. The best methods are to look at PROC EXPAND, usually under time series in EG or convert to logs and then use proc means and revert back to regular scale. The second option is detailed here: http://support.sas.com/kb/25/366.html

Re: Calculating Product for a selected observations of the same variable

Ksharp — Sat, 25 Feb 2017 07:12:24 GMT

If I was right, that is not geometric mean.

GM should like : ((1+r1)*(1+r2)) **1/2

data have;
infile cards expandtabs truncover;
input Client & $20.	FinancialProduct & $20.	 Date : monyy5.	Return : percent8.;
format date monyy. return percentn8.2;
cards;
Customer A	Product A	Jan2016	2%	$1.02
Customer A 	Product A	Feb2016	3%	$1.03
Customer A 	Product A	Mar2016	1%	$1.01
Customer A 	Product A	Apr2016	7%	$1.07
Customer A 	Product A	May2016	-5%	$0.95
Customer A 	Product A	Jun2016	3%	$1.03
Customer A 	Product A	Jul2016	-0.50%	$1.00
Customer A 	Product A	Aug2016	3%	$1.03
Customer A 	Product A	Sep2016	2%	$1.02
Customer A 	Product A	Oct2016	5%	$1.05
Customer A 	Product A	Nov2016	1%	$1.01
Customer A 	Product A	Dec2016	-1%
;
run;
proc sql ;
create table temp as 
select a.client,a.financialproduct,a.date,b.date as _date,b.return
 from have as a,have as b
  where a.client=b.client and a.financialproduct=b.financialproduct
        and a.date lt b.date
   order by a.client,a.financialproduct,a.date,b.date ;
quit;
data want;
 set temp;
 by client financialproduct date;
 retain cum 1;
 if first.date then cum=1;
 cum=cum*(1+return);
 mean=cum-1;
run;

Re: Calculating Product for a selected observations of the same variable

Reeza — Sat, 25 Feb 2017 07:16:10 GMT

It looks like a compound return rate?

Re: Calculating Product for a selected observations of the same variable

mkeintz — Mon, 27 Feb 2017 23:31:45 GMT

Let's say you have data set have, sorted by customer/product/date, where date is a true sas date variable. You also have the additional vars RET and RETPLUS1, and there are no gaps in the time series.

You want rolling cumulative returns for 2-months, 3-months, .... 12 months:

Erroneous program removed.

This program takes advantage of temporary arrays, which have their values automatically retained. We also use the fact that

3-month cumret = (prior 2-month cumret + 1) * (1 + ret) - 1

and generally

J-month cumret = (prior {J-1}-month cumret + 1) * (1+ret) - 1

Editted. See the comments below. ~~So the program first updates 2-month-cumret, then 3-month-cumret, ... 12-month-cumret~~

Thanks to @ksachar's response, I realize that I wan't paying attention to my own advice. First the program should loop from 12 to 2, not 2 to 12. I.e. update 12 month cumret first, then 11 month, etc. And I should update R{1} AFTER the loop, not before. The corrected program follows:

data want (drop=N nmonths);
  set have (rename=(date=todate));
  by prod;
  format fromdate yymmddn8.;

  array r {12} _temporary_;
  if first.prod then do;
    N=0;
    call missing(of r{*});
  end;
  N+1;

  do nmonths=12 to 2 by -1;
    if N<nmonths then continue;
    r{nmonths}=(1+ret)*(1+r{nmonths-1})-1;
    cumret=r{nmonths};
    fromdate=intnx('month',todate,1-nmonths,'end');
    output;
  end;

  r{1}=ret;
run;

Re: Calculating Product for a selected observations of the same variable

ksachar — Mon, 27 Feb 2017 17:43:34 GMT

Thanks a lot @Ksharp. You are correct. It is compounded return. If the time period exceeds 12 months , then I have to calculate the annualized return too.
My real problem is far more complicated than what I presented in "I HAVE". I tried to adjust your code but I am missing something.

Can you help me modify your code for the data bellow (I removed the last variable since you didn't used in your code)

Customer A ClassA Product A Jan2016 2% $1.02
Customer A ClassA Product A Feb2016 3% $1.03
Customer A ClassA Product A Mar2016 1% $1.01
Customer A ClassA Product A Apr2016 7% $1.07
Customer A ClassA Product A May2016 -5%
Customer A ClassA Product A Jun2016 3%
Customer A ClassA Product A Jul2016 -0.50%
Customer A ClassA Product A Aug2016 3%
Customer A ClassA Product A Sep2016 2%
Customer A ClassA Product A Oct2016 5%
Customer A ClassA Product A Nov2016 1%
Customer A ClassA Product A Dec2016 -1%
Customer A ClassA Product A Jan2015 -2%
Customer A ClassA Product A Feb2015 0.5%
Customer A ClassA Product A Mar2015 3%
Customer A ClassA Product A Apr2015 4%
Customer A ClassA Product A May2015 6%
Customer A ClassA Product A Jun2015 -1%
Customer A ClassA Product A Jul2015 5%
Customer A ClassA Product A Aug2015 8%
Customer A ClassA Product A Sep2015 -6%
Customer A ClassA Product A Oct2015 4%
Customer A ClassA Product A Nov2015 1.3%
Customer A ClassA Product A Dec2015 .8%
Customer A ClassA Product B Jan2016 3%
Customer A ClassA Product B Feb2016 7%
Customer A ClassA Product B Mar2016 9%
Customer A ClassA Product B Apr2016 -2%
Customer A ClassA Product B May2016 -1%
Customer A ClassA Product B Jun2016 -4%
Customer A ClassA Product B Jul2016 -4%
Customer A ClassA Product B Aug2016 1%
Customer A ClassA Product B Sep2016 -2%
Customer A ClassA Product B Oct2016 3%
Customer A ClassA Product B Nov2016 -5%
Customer A ClassA Product B Dec2016 6%
Customer A ClassA Product B Jan2015 -1%
Customer A ClassA Product B Feb2015 3%
Customer A ClassA Product B Mar2015 -1%
Customer A ClassA Product B Apr2015 2%
Customer A ClassA Product B May2015 -3%
Customer A ClassA Product B Jun2015 3%
Customer A ClassA Product B Jul2015 -9%
Customer A ClassA Product B Aug2015 2%
Customer A ClassA Product B Sep2015 1%
Customer A ClassA Product B Oct2015 -2%
Customer A ClassA Product B Nov2015 3%
Customer A ClassA Product B Dec2015 -2%

Thanks once again.

Re: Calculating Product for a selected observations of the same variable

ksachar — Mon, 27 Feb 2017 21:46:15 GMT

@mkeintz Thank you so very much for responding to my post. I tried the code but it is not giving me correct cumret. Will you be able to provide me the modified code on this sample data (I have added another variable). Also, Can you share the output so I can compare? want to make sure I am not making any mistake

If I need to extend the code to 24 months, will it work if I change :

array r {12} _temporary_;

array r {24} _temporary_;

AND

 do nmonths=2 to 12;

 do nmonths=2 to 24;

Re: Calculating Product for a selected observations of the same variable

Ksharp — Tue, 28 Feb 2017 02:17:01 GMT

Add one more code.

CODE NOT TESTED.

data want;
 set temp;
 by client financialproduct date;
 retain cum 1;
 if first.date or month(_date)=1 then cum=1;
 cum=cum*(1+return);
 mean=cum-1;
run;

Re: Calculating Product for a selected observations of the same variable

lakshmi_74 — Wed, 01 Mar 2017 04:26:57 GMT

data customer_dup(keep=cum_sum date_from date_to return prev_return) ;
retain cum_sum prev_return date_from start_pos position;
set customer_dup;
by customer product year;
if (first.customer or first.product or first.year) then
do;
put obs=;
prev_return=return_cal;cum_sum=return_cal;date_from=month;date_to=month;start_pos=obs;
end;
else
do;
date_to=month;
cum_sum=prev_return*return_cal;
prev_return=cum_sum;
end;final_cum=cum_sum-1;output;
if (last.customer or last.product or last.year) then
do;
do num=tot_group-1 to 1 by -1; position=start_pos+1;
do i= 1 to num by 1;
set customer_dup point=position;
if (i eq 1) then
do;date_from=month;date_to=month; prev_return=return_cal;cum_sum=return_cal;end;
else
do;
date_to=month;cum_sum=prev_return*return_cal;prev_return=cum_sum;
end;final_cum=cum_sum-1; output;
position=position+1;
end; start_pos=start_pos+1;
end;
end;
informat date_from date_to MONYY7.;
format date_from date_to MONYY7.;
run;

Re: Calculating Product for a selected observations of the same variable

ksachar — Mon, 20 Mar 2017 17:37:31 GMT

This works perfectly. My initial requirement was for a data for 6 years (i.e. 72 months and different combinations for 72 months-- which is nCr =(72C2) = 2556 ). After discussing with @lakshmi_74lakshmi i made minor changes to the code; removed YEAR in BY statement, anywere in first.year, last.year etc. and the retun was calculated for each Product group (date combinations).

Re: Calculating Product for a selected observations of the same variable

ksachar — Mon, 20 Mar 2017 17:39:09 GMT

@lakshmi_74 @Ksharp @mkeintz @Reeza Thank you for finding time and helping me resolve this.