Will We Be Able To See Trending?

[Non Sequor]

There are a lot of ways to do trend analysis and they all tend to fail to predict anything that you don't already know. I'd say that the hardware sales are too prone to fluctuation to make any reasonable projections.

Off the top of my head, I can think of a reasonable model for software sales though. You could model weekly sales as something like Ae^(-rt)+Be^(-st) where t is time in weeks since release and the rest are constants. A is a large constant representing initial sales, with r determined by the drop off in subsequent weeks. B is a smaller constant representing sustained sales, with s being determined by the longevity of sales (i.e. the "legs"). It may also be possible to incorporate holiday sales into the model by assuming they're proportional to A and adding something like an Ae^(-qt)*(sin(pi*(t+n)/104))^m term where n is 52 minus the number of weeks between release and the middle of the holiday sales season and where m is some even number that is sufficiently large that this term only has an impact over the appropriate weeks. q would have to be set by guesswork though, but it might be possible to get a reasonable value by setting it equal to r times some fudge factor determined by comparing release and holiday sales for past titles.

Ignoring all of my holiday guesswork, we could fit sales data to this curve by using least squares to get a best fit for s based the sales from week x on, where x is whatever week can be considered the cutoff between release sales and sustained sales. Then we would set B equal to the sales in week x divided by e^(-sx). Then A would be the release week sales minus B, and we'd set r by least squares. After gathering enough data, this could be used to create a projection for lifetime sales for a game.

Of course this is all pretty naive, but lots of big businesses use naive trend models all the time, so if all of the popular kids are doing it, I'm in.