Sorry for the hassle, but I just notice the solution gives the wrong fixed_time (every 2 seconds relative to 09:12 instead of 09:11). I changed the code a bit by subtracting 1 from the fixed_time and fixed_opentime variables (I haven't yet fully understand how the code works), so it seems to give the required result: Thanks again,  SK   data want (drop=nxt_: _: fixed_opentime); /* Read the openflag trade, followed by all trades (include 2nd read of openflag trade*/ set trades (where=(openflag=1) in=optrade) trades (in=alltrades) end=end_of_tr; /*by ticker; */ if optrade then do; /* if optrade set opentime and delete this record ... it will be read again*/ opentime=timestamp; fixed_opentime=round(1+opentime,2)-1; _open_minus_60=fixed_opentime-60; /*Minimum acceptable fixed_time*/ _open_plus_60=fixed_opentime+60; /*Maximum acceptable fixed_time*/ delete; end; retain opentime fixed_opentime _open_minus_60 _open_plus_60; format opentime fixed_opentime _open_minus_60 _open_plus_60 time8.0; /* Read ahead one record to get next fixed_time*/ if end_of_tr=0 then set trades (firstobs=2 keep=timestamp rename=(timestamp=nxt_timestamp)); else nxt_timestamp=_open_plus_60+2; fixed_time=round(1+timestamp,2)-1; nxt_fixed_time=round(1+nxt_timestamp,2); format fixed_time nxt_fixed_time time8.0; if nxt_fixed_time=fixed_time then delete; /*keep only final rec for each fixed_time*/ if nxt_fixed_time<_open_minus_60 then delete; /*drop records that are too early*/ if fixed_time>_open_plus_60 then stop; /*replace "stop" by "delete" if sorted by ticker/timestamp */ do fixed_time=max(fixed_time,_open_minus_60) to min(nxt_fixed_time-2,_open_plus_60) by 2; output; end; run;   ... View more

I apologize for the delayed response - I fell ill. Thanks for the complete solution, I highly appreciate it. SK ... View more

Thanks again for the help. As for your remarks: I'm using the proc sort because the first data step used to merge the open time places the opentime observation as the first row whereas according to its timestamp (09:00:11) it should be in the middle. In your reply, you mentioned "you can do that selection in the first data step" - how? I didn't succeed doing that (I can do it using proc sql). The code for "filling the holes" got different results than I expected, I would greatly appreciate it if you can elaborate.  Here are the results I'm looking for (the red rows are the "holes" I filled manually): TimeStamp X OpenFlag opentime fixed_time 09:00:01 98 0 09:00:11 09:00:01 09:00:03 93 0 09:00:11 09:00:03 09:00:04 83 0 09:00:11 09:00:05 09:00:04 83 0 09:00:11 09:00:07 09:00:04 83 0 09:00:11 09:00:09 09:00:11 62 1 09:00:11 09:00:11 09:00:13 60 0 09:00:11 09:00:13 09:00:15 54 0 09:00:11 09:00:15 09:00:16 29 0 09:00:11 09:00:17   Thanks again,  SK ... View more

Thanks for the prompt reply. Since I need the code to extrapolate the last known value, I changed the code as following. As you can see there are two "missing" observations 09:00:07 and 09:00:09 both of them should get the last known value of x=83. I will appreciate any help in filling out the missing observation in  an efficient way (I used proc expand but it is sort of a black box for me). Thanks,  SK   Data Trades; input TimeStamp time9. X OpenFlag; Format TimeStamp time9.; datalines; 09:00:00 61 0 09:00:01 98 0 09:00:02 54 0 09:00:03 93 0 09:00:04 83 0 09:00:10 42 0 09:00:11 62 1 09:00:12 46 0 09:00:13 60 0 09:00:14 92 0 09:00:15 54 0 09:00:16 29 0 ; run; data want; set trades (where=(openflag=1)) trades (where=(openflag=0)); if openflag=1 then opentime=timestamp; retain opentime; format opentime time8.0; fixed_time=opentime+ceil((timestamp-opentime)/2)*2; format fixed_time time8.0; run; Proc Sort data=want; by fixed_time timestamp; run; Data Want; set want; by fixed_time timestamp; if last.fixed_time; run; ... View more