SAS can analyse according to the formatted value of a date, so just apply a WEEK format to the date variable. Choose the date format according to how you want a week to start(Sunday or Monday) ... View more

The DO loop could be extended to collect the min and max asset value or return, during the 'history to date' ... View more

There might be a problem if a company has tlmore than one period of history. That data step would treat all data for a company as one history ... View more

Reply to @trungcva112 The array can be a fixed size. That needs to be large enough for the company with the most history. I thought 1200, to allow for 100 years of monthly data. If there is less history it is not a problem. Like all SAS statistical functions only non-missing values will be used. At the start of each data step iteration the whole array is set missing. Each value is inserted into the array just after each set statement and a STD() calculated. Each data step iteration ends when all rows of a company have been read. ... View more

sorry I posted code from SAS Studio (so line-breaks are missing) Shortly I'll post the code with line breaks ... View more

if this is only qtr/monthly intervals, it might be a good case to demonstrate array handling. SAS/IML would also simplify the resultant solution. What follows uses only a data step and assumes the monthly data arrive in month order within each company data progressive_stddevs ; do item= 1 by 1 until( last.company ) ; set your_data ; by company ; array cogX(1200) ; * 1200 months, probably enough per company? ; cogX(item) = cogsq ; n_cogs = n( of cogX(*) ) ; if n_cogs > 1 then COGSTD = std( of cogX(*) ) ; output ; end ; drop cogX: ; run ;  the array is filled progressively until end of the company, when it will be emptied by restarting the data step iteration The STD() function ignores empty cells in the array  ... View more

proc summary data= your_data ; var _numeric_ ; output n= ; proc transpose data= _last_( drop= _FREQ_ _TYPE_ ) out= non_miss_counts(rename= col1 = n ) ; run ; revised to provide counts of non-missing (proc means / summary statistic N) rather than the NMISS statistic ... View more

proc summary data= your_data ; var _numeric_ ; output nmiss= ; proc transpose data= _last_( drop= _FREQ_ _TYPE_ ) out= nmiss_counts(rename= col1 = nmiss ) ; run ; should work in any old release unless you have columns in your_data  named _type_ or _freq_  --> then I wouldn't know ... View more

Is that a pattern - every 4th cell belongs to the same obs? And these are all numeric? I like the idea of reducing the process - into a single pass (well I/O is the biggest performance blocker) Try proc iml ... View more