In certain circumstances, a running sum is a great way to predict the future. For example, if I take all the products I have on-hand, and add all the products I plan to build in the next few weeks, and subtract out the demand I plan to sell in those weeks, I should end up with my projected on-hand.
In most businesses, keeping track of inventory is a really good idea, because it means your business' capital isn't just sitting idle and depreciating in value. We ain't sellin' fine art 'round these parts!
The first order problem is, there are often a lot of gaps in the data. For a given part number, perhaps we don't plan to sell or build that item in week 2. If there's no Wk 2 record for a given part, then a running sum tool will simply move on from the Wk1 record to the Wk3 record. If I then display the example results by week, one can see, there's no record for Wk 2 when, in fact, we would have had 100 units laying around in that week.
The way to fix this is to separately summarize to every possible product, and every possible time at the lowest granularity, then append one to the other so you have a complete, empty, combination of product and time. Then, join the empty data set on the data set that has values by products and times in the same join and make sure you keep all the empty time records on the side of that join.
I wish I could take credit for this one, but this is all thanks to Mike Mixon, @mix_pix. You can check his twitter feed below.
