Had you ever made a Fourier analysis of the console sales? It seems perfect for to me. You have a signal that is someway periodic (T= 1Year), your measure with a constant frequency (1 Week time frame). You could analyze the how the amplitudes compare to each other and could try to predict the future of a console with it (lets say Christmas). I would to it, just to see if I'm right. But it would take me weeks to take the data from the page to a file ;)

Fourier analysis would be mostly useless in predicting game/system sales ... Much like the stock markets you're not going to find much of a secret pattern which can be exploited; about all you would notice is that in the final 1/4 of a year sales are up dramatically as compared to the rest of the year.

Well i think it would help. I think the Source was asking some time back, what consoles can be compared to the last generation, because on base 12 time higher Christmas sales, the other had 14 times higher. So he seems to think you can learn from the past that way. The Fourier analysis would just take that to a higher mathematical level.

And if you look at the weekly sales there is a frequent patter, Christmas. And why are the post of prediction the sales of the next month of TheSource and others so popular and high discussed. Isn't prediction the future sales one of the main thing in this forum?

I did a month by month fourier analysis a few months ago, but there are so many variables on console sales that can totally screw up the analysis.

IE:

Last generation, the PS2 had a very steady path for sales - similar to most other systems (Xbox, N64, PS1), in the fact that US sales for the EOY was very standard - it depicted a rising and falling of the peak sales, with years 3 and 4 (holidays) being the strongest. This is typical of most every system (save certain elements of the GC due to Zelda:TP, N64s OOT, and XBs H2 for Nov).

However, I never found any good-as-gold system for extrapolating a week to week measure of the end of the year. EOY Christmas sales are entirely determined by momentum and the right Holiday strategy, and not on a simple chart extrapolation.

A good example would be PS2 and X360 last year: both broke a few smaller molds on their sales. PS2 broke it because sales were so high for it's 6th US year. X360 broke it in the fact that it's EOY sales were quite a bit higher than the sales pattern through the rest of the year.

Likewise, it'd be hard to extrapolate where sales will be at the end of this year. Using very traditional analysis that wouldn't work; one would judge that the Wii will do 3m units in December alone in the US. However, production capacity cannot allow that. Likewise, these sales could go elsewhere (x360), or nowhere at all.

So then it's really just upto our prediction league ideas. There are so many variables and what-ifs for this Christmas, exactly like last year. Last year, X360 hardware was beyond mediocre, but the X360 managed to outsell any previous Xbox year in December. Likewise, Wii had one of the best first-December showings.

Basically though, for Oct-Dec, your sales will be anywhere from 6x (versus a normal per-month basis) on a very weak console, upto 15x for a very strong console. In the PS2's case, it was averaging about 300,000 units/month for many calendar months, and wound up with around 2.5-3.5m for many, many Nov/Dec NPDs. Likewise, October can be very telling, or not at all. October was a crappy month for the PS2 and X360, but both wound up doing great sales vs. October.

So again, it's all up to the prediction league. :-p

Form my experiancen with Fourier analysis of NMR signals it simply wouldn't work in this enviroment. You need alot of constants for the FID to be converted from the time domain to the frequency domain and I think that there is no way that you could get reliable data for such constants with regards to console sales data.

 Edit: this is a chemists PoV. Someone who knows more about console stats should be able to tell you more.

As others have noted, with all the unknown variables I can't imagine a Fourier Analysis would be useful. At the least though various statistical analysis should be useful to find ranges and other such somewhat useful info. However, I would also imagine stickball has already done that (particularly given his #2 spot on the prediction league).

 First off: I'm impressed so much people here know what a Fourier Analysis is. Didn't excpect that :o)

Nice to know you have tried. And there will never be a way to predict the Christmas sales precise, thats for sure. Month by Month is not enough for a good Analysis.  You only have 12 points, looking to predict good 2 of them and they have not a constant Frequency. That won't even work in a physical environment :o).

All you are saying is true. But it is not like I just want to make that and say, I know how it will went. There is still the thing to do, where someone calculates the price of the console, etc. in. I thought of it as a tool for people trying to predict the near future (TheSource e.g.), not a wonder! But nice to know that I am not the first to think about that.

 Is there a way to get to the weekly sales easily, without get a week of and type them one by one from the homepage. All i can think of is write a html parser. 

All the problems you describe i have to solve every day at my work. The PS2 graph really looks like a current of a electrical window regulator of a car. And thats something I have to analyze a lot :o)

like ben said: is there a way other than writing an html parser to gather the data?

fourier analysis: like others say, doesn't appear useful. should be able to explain the periodicity anyway.

(given this is a video game site, i imagine about 15-25% of the posters know fourier analysis? it's only freshman calculus after all)

shams is a dirty cocksucker, who takes it in the ass. just like all the wii fans.

Not blue3.