The macro variable created by call symput isn't available until the data step ends so if you ran this on a fresh session you would get a note in your log telling you it couldn't be resolved then on the second run it would hold the last value from the first run and so on....   If you tell us exactly why you want to do something like this (it isn't clear from the example) we can offer an alternative. ... View more

This is an interesting question and to be honest I was a little surprised to realise their wasn't an inbuilt function to give us the number of distinct values across a range of variables.   There's only one change I'd suggest to Kurt's answer - if you don't add the length statement then the value of the flag variable is truncated for the second observation so:   data want; set dt1; length flag $3; array vars {*} a1-a3; count = 0; flag = 'no'; do i = 2 to dim(vars); if vars{i} ne vars{i-1} then do; count = count + 1; flag = 'yes'; end; end; drop i; run; The other thing that occcured to me was how would you want missing values treated?     ... View more

I think this should do the trick - note that if FORD for July is missing then flag will also equal 1 for that month - you'll need to amend the code if that isn't what you want   data have; input account $ transactionamount month $; datalines; FORD 100 JAN FORD 100 FEB FORD 100 MAR FORD . APR FORD . MAY FORD . JUN FORD 100 JUL FORD 100 AUG FORD 100 SEP FORD 100 OCT FORD 100 NOV FORD 100 DEC TOY 100 JAN TOY 100 FEB TOY 100 MAR TOY 100 APR TOY 100 MAY TOY 100 JUN TOY 100 JUL TOY 100 AUG TOY 100 SEP TOY 100 OCT TOY 100 NOV TOY 100 DEC ; run; proc sql; create table have_sorted as select * from have order by account, case when month="JAN" then 1 when month="FEB" then 2 when month="MAR" then 3 when month="APR" then 4 when month="MAY" then 5 when month="JUN" then 6 when month="JUL" then 7 when month="AUG" then 8 when month="SEP" then 9 when month="OCT" then 10 when month="NOV" then 11 when month="DEC" then 12 else 99 end ; quit; data want; set have_sorted; by account; if first.account then counter=1; else counter=counter+1; lag1=lag1(transactionamount); lag2=lag2(transactionamount); if missing(transactionamount) and missing(lag1) and missing(lag2) and counter>2 then flag=1; else flag=0; retain counter; run; ... View more

I believe you're overthinking the problem - if you sort your data by accountno and month then, in a data step, use the lag function to look back one, two and three periods by accountno then you'll find it much easier. I suspect your solution would run into multiple complexities with a rolling three month span.   One thing to be careful of with lag() though is - "don't use it inside conditional statements" or you'll get incorrect results.   If you post a datastep to create some sample data and some desired output we can give you a more detailed answer. ... View more