Standard deviation of firm i's sales from year t-4 to t

[Fer_]

Hello everyone, 

I have dataset which has gvkey, year, and sale. I need to calculate standard deviation pf sale for each firm (identified by gvkey) between year of t to t-4.

For example  standard deviation of sale for 1996 from 1996 to 1995, 1994, 1993, 1992. 

I appreciate your help. 

gvkey	year	sale
1004	1992	524
1004	1993	765
1004	1994	442
1004	1995	510
1004	1996	324
….	….	…
1019	1990	2900
1019	1991	3600
1019	1992	5624
1019	1993	4527
1019	1994	6570
1019	1995	4298
1019	1996	4657
1019	1997	5498
1019	1998	6790
….	…	…

[sbxkoenk]

Hello,

You need a moving standard deviation with size of the window equal to 5 (t-4, t-3, t-2, t-1, t).

You can easily do this with PROC EXPAND (SAS/ETS).
Here are some transformations that will be of interest to you :

• CMOVSTD window Centered moving standard deviation
• CUSTD (n is optional) Cumulative standard deviation
• MOVSTD window Backward moving weighted standard deviation:

Try for example this:
Take care! : I am not sure if this will calculate std on (t-4, t-3, t-2, t-1, t) or on (t-5, t-4, t-3, t-2, t-1).
Please check ... if it's not including the current line t (as you want) , I will tell you how to "shift" the window.

proc expand data=sashelp.pricedata out=abc;
 by region line product;
 convert sale = sale_MOVSTD / transformout=(MOVSTD 5);
run;

proc datasets library=work NoList Nodetails memtype=DATA;
 modify abc;
  label sale_MOVSTD = "sale moving std window size = 5";
run;
QUIT;

BR, Koen

[mkeintz]

@sbxkoenk 

You can actually assign the label during the PROC EXPAND, thereby avoiding the need for a subsequent PROC DATASETS.

The price?  An unavoidable warning note, as in:

31   proc expand data=sashelp.pricedata out=abc;
32    by region line product;
33    convert sale = sale_MOVSTD / transformout=(MOVSTD 5);
34    label sale_MOVSTD = "sale moving std window size = 5";
WARNING: Variable SALE_MOVSTD not found in data set SASHELP.PRICEDATA.
35   run;

[Reeza]

data want;
array p{0:4} _temporary_;
set have; 
by gvkey year;

if first.gvkey then do;
call missing(of p{*});
num_years=0;
end;
num_years+1;

p{mod(_n_,5)} = sale;

*if you only want to calculate this once you have 5 years of data;
if num_years>=5 then sale_std = std(of p(*));

*if not conditional remove the IF portion;
run;

I would probably recommend PROC EXPAND as it will deal with missing data and years better, however, here's a rough data step approach. You will need to tailor to deal with missing years/data if necessary.